Ethics is becoming increasingly important in society. But despite its importance, it can be a complex subject to teach. Students often have preconceived ideas and are not aware of the consequences of these views. Games offer a fun, low-risk way to explore the consequences of our actions; they allow the players to explore other viewpoints and see the results of their choices.

This project will modify a previous Part IV project to convert a board game into a card game and change the domain from robo-ethics to AI ethics.  Previously, three Part IV projects designed and evaluated a board game for teaching ethics in a robo-ethics course at the university. Based on the evaluations, we want to simplify the game and make it easier to use in a large course. You can choose the aspects of ethics you are interested in and extend the games to include them. Alternatively, the previous projects have suggested some new directions for future versions that you could implement.

This project is an extension of previous Part IV projects and is a collaboration with the Design Programme.

Location: MDLS (405-569)

Children are often fascinated with robots and want to control (program) them. Currently, most programming environments are screen-based, requiring the use of either a mouse and keyboard or a touch screen. This project will investigate using physical coding blocks (tangibles) as an alternative approach for programming robots and writing software programs.

The project will primarily focus on creating an embedded system to detect and interpret a chain of physical programming blocks, generate the intended software program, and display the associated output. It will interface with an existing software system for controlling mBot2 or Nao robots. While the project can cover any aspect of the programming environment, the focus will be on designing and implementing a tangible programming system.

This project is an extension of previous Part IV projects.

Location: Embedded Systems (405.760, Lab)

One of the primary challenges in software maintenance is identifying the specific locations within a codebase that require modifications. Change Localisation Time (CLT), defined as the time needed to pinpoint where changes are necessary, significantly impacts the overall duration of software modification processes. Therefore, reducing CLT is crucial for enhancing developer productivity.

This project aims to address the issue of CLT through a two-stage approach: First, conduct a comprehensive review of existing literature to identify various techniques that have been proposed to reduce CLT. Second, the identified techniques must be assessed to determine their effectiveness in lowering CLT.

By identifying and evaluating relevant techniques, the project seeks to recommend the most effective strategies for minimising CLT. As a result, it will contribute to more efficient software maintenance practices.

Note: Student-supplied project

Location: MDLS (405-569)

There is a vast gender imbalance in engineering, with few women entering the specialisation. Entry rates have remained unchanged despite attempts to encourage women. One potential reason is that children start to form opinions early in life (before ten years of age), which then influence the subjects they choose in schools, limiting their choices for university.

In this project, you will explore some STEM activities that can be used in primary schools and their on girls. This project builds on previous projects that developed a robotics programming environment and a set of electronic components for teaching. You can choose to use either of these environments or look at developing your own. In particular, we are looking at how to improve girls’ perceptions of STEM and help them form positive opinions of the topic.

This project is an extension of previous Part IV projects.

Location: MDLS (405-569)

The Formula SAE competition is introducing an autonomous vehicle category to the control of the Formula SAE cars. To meet this challenge, we seek to develop a fully autonomous control system that enables the car to learn to drive itself through Reinforcement Learning. Reinforcement Learning is a machine learning approach that seeks to enable a robot to learn to operate based on its interaction with the environment without human input or control.

Prior work has developed a means of enabling a Formula SAE car to autonomously navigate through reinforcement learning-based control on a simulated F1Tenth race car and the real world (https://github.com/UoA-CARES/autonomous_f1tenth). This project will extend this work to enable learning to race against opponents' cars - a critical but deceptively challenging task for autonomous vehicles.

This project will require strong programming skills, specifically in Python, and will require the students to come frequently into the robotics lab to work on the project. This project is not suitable for remote development due to the requirement of working with and testing on the physical vehicle. Prior experience with Pytorch or ROS (1 or 2) would be beneficial but not required, as we will teach you those tools as part of the project.

This project falls into the Robot Learning Team based out of the CARES research group (https://cares.blogs.auckland.ac.nz/). We will seek to publish this project's results at an international conference.

Location: Robotics (405.652, Lab)

Technologies become more human-centred when they adapt to the user's preferred way of interaction. Given the convenience of verbal expression, it's no surprise that the widespread adoption of AutomaticSpeech Recognition (ASR) has enabled smart devices to serve in roles ranging from voice assistant to language interpreter. In the case of dysarthria, a speech condition commonly secondary to neuromotordisorders, a functional ASR system can provide essential accommodation to promote patient autonomy. However, ASR does not work equally well for people overlooked during the system's design, resulting in unfortunate usability disparities across demographics, accents, and disabilities.Existing mitigation requires significant effort by the disadvantaged user to improve ASR accuracy through personalized speaker adaptation, leaving an expensive and troublesome barrier to accessing assistive technologies.In this project, you will investigate state-of-the-art machine learning methods to leverage personalized speaker adaptations for the benefit of all speakers. Possible directions include applying novel techniques such as cross-domain ensemble distillation [1], though you can also propose an original approach after conducting a literature review.

Location: HASEL (405.662, Lab)

The inverse problem of electrocardiography essentially requires one to predict what is happening on the heart surface of a patient (called EGMs), given their body surface potentials (called ECGs). If a good prediction can be made, then cardiologists can use these predictions to easily and efficiently plan heart surgeries.

We have developed mathematical models for predicting EGMs from ECGs. Vison transformer AI models, in particular Dino v2 (https://github.com/facebookresearch/dinov2) from Meta AI has been used to learn EGMs represented as heat maps from ECGs also represented as heat maps. We have got very good preliminary results.

The aim of the project is to extend the AI model to handle complex arrhythmias (heart problems). In particular we want to study mixing deep learning models with Bayesian statistical mixture density models in this project.

Location: Embedded Systems (405.760, Lab)

Parallel computing and the use of GPUs (Graphics Processing Units) has become extremely important for machine learning and in today's IT world in general. To efficiently use them one needs to divide a machine learning model or program into tasks/operations and to schedule them onto the processors or cores of the GPU or system. Theoretically this is addressed by task scheduling where a program is described by a directed acyclic graph, a so called task graph. The primary objective of task scheduling is the fast execution of a model or program, but the energy consumption becomes more and more important too. So called slack reclamation algorithms slow down the execution of non-crucial tasks so that idle times (the slack) are filled and the energy consumption is reduced. But recently the question was raised how much slack is actually present in schedules produced by modern scheduling algorithms. This project will investigate this important question, whose answer can have a strong consequence on the design of future scheduling algorithms and in turn the sustainability of large computing jobs like machine learning.

Location: MDLS (405-559)

All satellites need to function with a limited power supply. This is especially true for small form-factor satellites like CubeSats. This also forces the on-board processing to work with low power. At the same time, processing data in the satellite can significantly reduce the need to transmit data, which itself is power and energy hungry.In this project we will therefore investigate the design and implementation of power efficient  algorithms for imaging using techniques based on synthetic-aperture radar (SAR). SAR has many applications and is widely used, but usually in larger satellites.The special focus of this project will be on the use of Neural Processing Units, which are designed for very power-efficient machine learning applications. Here, we will investigate how NPUs can be used in the domain of SAR, for which NPUs were not designed. Through this you will learn about the architecture and programming of the newest compute devices for machine learning. We will consider AMD Ryzen AI CPUs and NVIDIA Jetson devices that possess NPUs and are realistic devices for satellite integration.

Location: MDLS (405-559)

Computer technology has truly conquered our work and life. While we strongly benefit from it, computing consumes an ever increasing fraction of our energy production. Making computing more energy-efficient is therefore of paramount importance, not least due to emerging computing needs of the sensational advances of machine learning and artificial intelligence. Technological advances have made computers more energy-efficient over the years, usually paired with smart approaches to benefit from these advances. For example, the speed of processors can be dynamically adjusted to reduce the power consumption and a lot of research has focused on algorithms using this. However, due to technological developments the movement of data has become a significant source of energy consumption in the computation of a program. Almost all modern computers have more than one processor and data moves between the processors and the memories. Avoiding or minimising this data movement can significantly reduce the energy consumption of computers.In this project you will investigate how the duplication of tasks can reduce the energy consumption. So instead of transferring the result of an operation or a task, it is recomputed (a task is duplicated) at the destination. You will develop scheduling algorithms that can not only allocate and order sub-tasks of a program onto the processors, but also decide when it is beneficial to duplicate a task. This will reduce the energy consumption of computing and make it more sustainable.

Location: MDLS (405-559)

The gender gap in engineering continues to be a significant issue, with both academia and industry being dominated by men. This topic continues to receive a growing amount of research attention, looking at both why women are less likely to go into engineering, and why women engineers may not be retained in industry.

This work aims to help understand how gender stereotypes impact the perceptions of who should become an engineer. We plan to follow the recent work from Cutrupi et al. [1], in which they asked school aged children (Primary and Intermediate) to draw a software engineer, in order to collect their perceptions and investigate whether gender stereotypes still persist. This work will include, helping to conduct the “draw an engineer” sessions, analysing the output of these sessions, and preparing a research paper.

[1] Cutrupi et al., “Draw a Software Engineer Test - An Investigation into Children’s Perceptions of Software Engineering Profession” at ICSE 2023

Location: MDLS (405-569)

Those with dyslexia often face significant barriers in reading and writing clearly. In addition to immediate challenges with spelling and grammar, writers with dyslexia can struggle with ordering and expressing their ideas, choosing language to match their desired tone, and writing with clarity and precision. This can have a significant impact in educational, professional and social contexts.

While general spelling and grammar tools (e.g., Grammarly) are rapidly improving, they are often insufficient when applied to dyslexic text. The goal of this project is to investigate state-of-the-art language models (LLMs), as well as other dyslexia specific techniques to dyslexic text correction. Subsequently, a prototype text correction tool will be developed, in the form of a text editor, web-plugin, or similar.

Location: HASEL (405.662, Lab)

The project implements reinforcement learning-based strategy to effectively mitigate system frequency challenges across a multi-machine power network. The proposed framework divides the interconnected network into smaller areas, each containing various generators. The load frequency control (LFC) is to be implemented at the designated machines in each area, ensuring the frequency stability under load demand variations and bus faults.

Location: Control & Signal Processing (405.722, Lab)

The Multi-Terminal Direct Current (MTDC) power system enhances the power system’s capacity and efficiency while easing the integration of Renewable Energy Resources (RES). However, the DC link capacitors discharge an enormous high-frequency current in the event of a Direct Current (DC) line fault. These enormous currents of high frequencies can instantly damage the sensitive electronic components attached to the DC transmission system, particularly the Voltage Source Converter’s (VSC) freewheeling diodes. Therefore, rapid and robust DC fault protection and coordination topologies are necessary to facilitate the future MTDC grid application. Recently, this study introduced two novel fault detection and localization topologies using the time derivative and high pass filter (HPF) response that identify faults in a few microseconds (µs), which is validated using simulations and a point-to-point hardware prototype. Further, simulation and hardware prototypes will evaluate these topologies and others for protection coordination strategies in the forthcoming MTDC grid. Therefore, the previously developed prototype of the two-terminal P2P system should be extended to a four-terminal radial MTDC system to verify protection and coordination topologies in this part IV project.

Location: Power Systems (405.628, Lab)

1. Background

The bidirectional flow of energy between electric vehicles (EVs) and energy storage systems is crucial for modern charging infrastructure. This project focuses on implementing a bidirectional DC-DC converter with an active-clamp forward topology to ensure efficient energy transfer and dynamic response. A dual-loop PI control strategy is used to achieve stable voltage and current regulation.

2. System Overview

Bidirectional DC-DC Converter

l    Function: Enables energy transfer between the energy storage system and EV.

¡ Boost Mode: Energy from storage flows to EV (discharge).

¡ Buck Mode: EV surplus energy is fed back to the storage system (charge).

Dual-Loop PI Control Strategy

l    Voltage Outer Loop: Ensures stable output voltage.

l    Current Inner Loop: Enables fast dynamic response and precise current tracking.

3. Design Components

l    Power Stage Design

¡ Active-Clamp Circuit, GaN FETs for reduced switching losses.

¡ Transformer design for energy isolation.

l    Control Algorithm

¡ PWM-based dual-loop PI control for bidirectional energy regulation.

l    Testing and Validation

¡ Performance metrics: efficiency, stability, and dynamic response.

Location: Power Systems (405.628, Lab)

The integration of Photovoltaic (PV) systems with Hybrid Energy Storage Systems (HESS) in microgrid applications represents a crucial focus in research and development, aiming to improve both efficiency and reliability. Key control strategies employed in this field include Maximum Power Point Tracking (MPPT) for optimizing PV system performance and Proportional-Integral (PI) control for managing the HESS. MPPT algorithms are designed to maximize the energy harvested from PV panels. They continuously adjust the operating point of the PV system to ensure it operates at its maximum power point, despite changes in environmental conditions like irradiance and temperature.

PI control is a widely used technique for regulating the charging and discharging processes in the HESS, which often includes a combination of batteries and supercapacitors. PI controllers are valued for their simplicity and effectiveness in maintaining a stable state of charge and meeting load demands. By tuning the proportional and integral gains, the controller ensures efficient energy management, contributing to the overall stability and efficiency of the microgrid.

In this project, the PV system will be managed using MPPT algorithms, while the HESS will be regulated through a PI-based control strategy. The performance of the PI control will be compared to that of Model Predictive Control (MPC) to highlight its advantages and limitations in terms of stability, efficiency, and response time. The test-bed at the Power Systems Lab in the Department of ECSE, University of Auckland, will be used to experimentally validate the concept and explore potential improvements.

Location: Power Systems (405.628, Lab)

Abstract:

Deep Neural Networks (DNNs) are powerful but often act as "black boxes," making their decisions opaque to users. Building on the novel Perception Visualization (PV) framework introduced in the paper "Perception Visualization: Seeing Through The Eyes Of a DNN," this project invites students to explore and extend techniques for visualizing and interpreting DNN behavior. By combining saliency maps with semantic reconstructions, PV reveals what a model perceives alongside where it focuses, offering more profound insights into model reasoning. Students will explore the methodologies for visualizing deep neural networks (DNNs) and assess their effectiveness in enhancing model interpretability. This project will empower students to bridge the gap between model performance and human understanding, fostering trust in AI systems.

Project Objectives

• Implement the Perception Visualization (PV) on a pre-trained CNN.
• Explore enhancements to reconstruction quality using advanced techniques.
• Evaluate PV’s interpretability against traditional methods through quantitative metrics and user studies.
• Apply PV to a real-world problem (e.g., medical imaging, autonomous driving) to identify model failures and propose corrective actions (e.g., dataset augmentation).
• Build an interactive Python toolkit for generating PV explanations, integrating it with popular frameworks like PyTorch or TensorFlow.

Concluding Remark

This project presents an exciting opportunity for students to engage with cutting-edge research in AI interpretability. By exploring the innovative approach of Perception Visualization, students will contribute to the ongoing discourse in explainable AI while developing essential technical and analytical skills.

Location: Control & Signal Processing (405.722, Lab)

The project aims to develop and deploy an Autonomous Underwater Vehicle (AUV) capable of surveying marine vessels and underwater structures for biofouling and invasive species, particularly in New Zealand waters. The AUV will provide a safer, more efficient alternative to human divers, enhancing biosecurity surveillance efforts and enabling the large-scale assessment of biofouling levels, in line with regulatory guidelines.

New Zealand's marine ecosystems are at risk from the spread of invasive species, which can attach to the hulls of recreational and commercial vessels. Identifying and managing biofouling is crucial for protecting biodiversity and safeguarding commercial fisheries. Current biofouling survey methods are resource-intensive, relying on divers to visually inspect vessels, a practice that is both time-consuming and hazardous. The limited speed and scale of manual surveys mean that only a small number of vessels can be inspected.Kōtare Group Limitedseeks to address this challenge by deploying an autonomous underwater vehicle (AUV) to perform biosecurity surveillance on vessels and underwater infrastructure. By automating these surveys, we aim to significantly increase the efficiency, speed, and safety of biofouling inspections, ensuring compliance with the Ministry for Primary Industries' (MPI) biosecurity regulations.

Autonomous Survey Capability: The development of an autonomous or semi-autonomous system that can survey vessels and structures for biofouling without the need for human divers.Biofouling Detection: The system should be able to accurately assess biofouling surface coverage, measure thickness, and identify invasive species.

This project will include the development of underwater computer vision systems and underwater mapping of ship hulls.

Location: Robotics (405.652, Lab)

This is an industry-sponsored project by Crown Equipment Limited.

Student Benefits of this Project:

• You will gain experience working with an existing New Zealand company.
• You will be solving real-world problems.
• You will be exposed to a company that works on cutting-edge robotics and automation.
• You will be able to explore and gain an understanding/experience in topics that may not have been covered in your course material.

Problem Description: A pallet is a flat structure that supports goods while they are being transported by machines such as forklifts. Pallets are commonly wooden structures that, through wear and tear, may crack, lose panels, warp or lose rigidity. Crown is interested in using active depth cameras (time-of-flight and structured lighting) to identify defective pallets during typical day-to-day processing so they may be repaired or removed from circulation. This problem is more complex since the pallets may carry a load and could be partially occluded by semi-transparent wrapping. You will be provided with:

• A test pallet, wrapping and active depth cameras.
• Access to GPU machines for machine learning.
• Guidance in understanding the material required to solve the problem.

Applicable material: If you have an interest in this space, here are some relevant works to look over:

• Behind the Curtain
• 3D Object Recognition
• Pallet Recognition
• Region Proposal Networks
• Pallet Recognition
• Damaged Pallets

Location: Robotics (405.652, Lab)

This project addresses New Zealand’s growing brain health challenges, particularly among the elderly. Existing support systems lack practical, evidence-based tools for pre-assessment and advisory support. Additionally, accessible and effective solutions for early dementia detection and management are urgently needed, especially within diverse ethnic communities. We propose developing an AI-powered chatbot to enhance brain health and well-being while integrating dementia pre-assessment capabilities. This system will enable real-time conversations with an AI-driven expert, capable of understanding user queries, providing relevant information, and offering personalized support. It will also assess early signs of dementia through user interactions, behavioral patterns, and self-reported data, delivering tailored guidance to help users and caregivers manage risks effectively. Leveraging advanced large language models (LLMs) and a structured brain health framework, the chatbot will ensure meaningful, empathetic, and accurate interactions. Its effectiveness will be evaluated based on usability, reliability, and impact on brain health and dementia care outcomes.

Location: Control & Signal Processing (405.722, Lab)

The software industry lacks diversity. Recent research has theorised that software development tools can have inbuilt biases, making them easier to use for people who think in a certain way. For example, GitHub, a commonly used software tool, has been found to have missing information, unclear terminology, and unexplained processes. This can make GitHub non-intuitive for people with process-oriented learning styles as they prefer to have complete information before starting a task.

This project will select a common software development tool or practice and investigate potential inbuilt biases in-depth by designing a lab experiment. A set of recommended changes to the selected software tool or practice will be derived through this investigation.  These changes will be validated through further experiments if time allows.

Location: HASEL (405.662, Lab)

Participatory urban design has recently been highlighted as a key element to generate community wellbeing and safety (Anderson & Baldwin, 2017), and foster a cohesive, democratic neighbourhood (Sanoff, 2010) that in turn contributes to its resilience in facing potential adversities (Zautra et al., 2008). Its implementation, however, is challenging. Lack of motivation to participate (Del Gaudio et al., 2017) and lack of technical capacity to participate (Arslan & Nursen, 2024) are underscored as its main challenges.

This project seeks to tackle these challenges by leveraging the advantages of gaming and digital technology. It aims to build an AR gaming application to encourage and empower public participation in a neighbourhood public space design. Specifically, the game should serve as a digital tool to support: (i) community empowerment and assertion of their right to the city; (ii) improvement of social and spatial relation that contribute to inclusivity and cohesion; (iii) urban re-enchantment – play and enjoyment in the city.

References

Anderson, J., & Baldwin, C. (2017). Building Well-Being: Neighbourhood Flourishing and Approaches for Participatory Urban Design Intervention (pp. 313–337). https://doi.org/10.1007/978-94-024-0878-2_17

Arslan, Ö., & Nursen, K. (2024). Challenges and opportunities in participatory urban design: Highlights from the Turkish cases. METU Journal of the Faculty of Architecture, 40(2).

Del Gaudio, C., Franzato, C., & de Oliveira, A. J. (2017). The challenge of time in community-based participatory design. Urban Design International, 22, 113–126.

Sanoff, H. (2010). Democratic design: Participation case studies in urban and small town environments. (No Title).

Zautra, A., Hall, J., & Murray, K. (2008). Community Development and Community Resilience: An Integrative Approach. Community Development, 39(3), 130–147. https://doi.org/10.1080/15575330809489673

Location: Control & Signal Processing (405.722, Lab)

Aged care is increasingly benefiting from advanced technological solutions. Similar to AI-driven systems in the hospitality industry, intelligent assistants are now being introduced in rest homes and elderly care facilities to create personalized, smart care environments. These AI-powered systems not only reduce caregivers’ physical workload but also enhance healthcare accuracy and efficiency.This project aims to develop an AI-driven software solution designed to enhance aged care by integrating a virtual avatar for companionship and interaction. The avatar will act as a human-like interface, fostering a sense of presence and emotional connection for residents. Additionally, the system will collect and analyze activity-related data to identify potential health risks, enabling proactive care interventions. By promoting companionship and strengthening human connection, this solution seeks to improve both the physical and emotional well-being of residents, ultimately enhancing their overall quality of life.

Location: Control & Signal Processing (405.722, Lab)

Dementia is a complex and progressive neurodegenerative condition. Individuals living with dementia often exhibit wandering behavior in rest homes or care facilities. While this behavior may provide a sense of autonomy, it also poses significant safety risks, particularly the risk of falls. To address these challenges, this project aims to develop an advanced indoor activity tracking system that enhances safety, supports caregivers, and preserves the dignity of individuals with dementia.Building on an existing Ultra-Wideband (UWB)-based system, this project will expand its capabilities by integrating fall detection and advanced risk management features tailored for indoor environments in care facilities. The system will provide continuous, high-precision monitoring of movement patterns, enabling early risk detection and rapid response to incidents. Through comprehensive evaluation, the solution will be refined to maximize its effectiveness in reducing the risks associated with indoor wandering.

Location: Control & Signal Processing (405.722, Lab)

Introduction

Vegetation indices such as the Normalized Difference Vegetation Index (NDVI) and Normalized Burn Ratio (NBR) are critical tools for monitoring ecosystem health, fire severity, and post-disaster recovery. Traditional methods rely on manual calculation of these indices using spectral band ratios. This project proposes leveraging deep learning (DL) to automate and enhance the accuracy of vegetation index computation, fire severity classification, and predictive modelling for environmental monitoring. By integrating multispectral remote sensing data with DL techniques, the project aims to improve scalability and precision in large-scale ecological assessments.

This project bridges cutting-edge DL techniques with environmental science, offering students hands-on experience in solving real-world problems. Outcomes could aid agencies in disaster response, ecological conservation, and climate change mitigation.

Objectives

1. Develop a deep learning model to predict vegetation health (NDVI) and fire severity (dNBR) from multispectral imagery.
2. Compare the performance of DL-based approaches with traditional index calculation methods.
3. Create a scalable Python pipeline for processing satellite data, training models, and generating actionable insights.

Methodology

1. Data Collection & Preprocessing
• Possible Datasets:
• Landsat 8/9 (30m resolution, bands: Red, NIR, SWIR).
• MODIS (250–500m resolution, bands: Red, NIR, SWIR).
• Cold Springs Fire Dataset (pre- and post-fire NAIP/Landsat imagery).
• Tools: Google Earth Engine, USGS Earth Explorer, earthpy for data retrieval.
2. Model Development
• CNN Architecture: For pixel-wise regression (NDVI/NBR prediction) and segmentation (fire severity classification).
• Hybrid Models: Combine CNNs with recurrent networks.
• Transfer Learning: Fine-tune pre-trained models (e.g., ResNet) on multispectral data.

References

• Earth Lab Course Materials: Vegetation Indices in Python
• USGS Landsat Data Access: Earth Explorer
• Google Earth Engine: Dataset Catalog

Location: Control & Signal Processing (405.722, Lab)

With the development of LED technology, a new emerging advancement follows—lens technology. LEDs with lenses offer unique lighting conditions and a flexible way to control light. However, this potential remains largely unexplored, and, in fact, very few products provide effective lighting control.

In particular, emergency lighting in New Zealand lacks proper light control. Some European products already incorporate advanced light control techniques, and their approach appears highly promising. However, New Zealand and Australia have yet to adopt such techniques. One possible reason for this could be the reliance on classification systems that primarily indicate recommended spacing rather than optimizing light control.

Key Research Components:

Develop a process for calculating horizontal illuminance based on luminous flux distribution.

Ensure compatibility with LDT and IES file formats for seamless integration with lighting data.

Location: Control & Signal Processing (405.722, Lab)

Tertiary institutions commonly use SET (Student Evaluation of Teaching) surveys to evaluate student perceptions of courses and teachers. They were designed to be a straightforward approach to providing actionable feedback. However, international research has identified several shortcomings with the surveys, how they are applied, and interpreting the results. Despite these findings, SET surveys are still mandated for University of Auckland courses and at many other universities.

This project will evaluate international research on SET evaluations and suggest an alternate evaluation approach.

Location: MDLS (405-569)

Back-to-back classes, assignments, and meetings often leave students with little time to truly recharge. Even when breaks are scheduled, they’re frequently spent checking emails, reviewing notes, or preparing for the next task—activities that keep the brain in a constant state of work. This continuous mental engagement can lead to cognitive fatigue, reducing both focus and overall academic performance. While meditation is a commonly known tool for mental clarity, it requires practice and patience—something many people may find challenging to fit into their daily routines.

This project looks at an alternative approach: using active musical listening as a tool for rapid cognitive resets. Unlike passive listening, where music fades into the background, active listening involves consciously focusing on specific musical elements like melody, rhythm, and instrumentation. The project will explore the research question of whether active musical listening can help students switch more effectively between different cognitive demands. For example, a student transitioning from a technical engineering assignment to writing an essay might struggle with lingering analytical thoughts. Taking just a few minutes to engage with music in an active way could help clear these residual ideas and enter a fresh mental state for the next challenge.

Location: Radio Systems (405.736, Lab)

Background

Large parts of the world have no cellular coverage, making it difficult to monitor environmental conditions and hazards such as air quality, groundwater levels, or wildfire risks. These areas mostly rely on manual collection, which is expensive, slow, and can be dangerous. For example, the 2024 Bridge Hill/Castle Hill fire in Canterbury spread over 980 hectares partly because it wasn’t detected early enough.

Real-time data is also important for people in remote locations, like farmers and hikers, who rely on accurate information to stay safe and make decisions. While satellite services (e.g., Starlink) can provide internet access, they typically need more than 10 watts of power, which is far too high for low-power IoT (Internet of Things) sensors that usually work at milliwatts or even less. Many such sensors run on tiny batteries or have no batteries at all, making it essential to find a communication method that uses very little energy.

The Technology

This project explores how to use sunlight for communication in remote sensing. Instead of creating a new signal, Ambient Light Backscattering (ALB) uses existing light sources, like sunlight, to transmit data. Traditional radio systems need to both create and encode (modulate) their signals, which uses more power. ALB only needs to encode data into the light already present, greatly reducing power consumption.

The system works by using special surfaces (such asliquid crystals shuttesr in 3D glasses) that quickly switch between being transluscent and opaque. This creates an “on-off” pattern in the light (i.e. amplitude modulation), which human eyes can’t detect but photodetectors can pick up and decode into data. This approach enables extremely low-power communication, making it ideal for remote, low-power sensors.

This project

We are asking whether it’s possible to build a reliable system that:

1. Can raise alerts to nearby systems using only ambient light, even if sunlight levels change or are low (e.g., during hazards or bad weather).
2. Can provide more detailed information (like exact location or air quality measurements) when brighter light is available.

and do this in an energy-efficient way (self-powered). This could lead to new, energy-efficient ways of monitoring the environment in areas with limited or no cellular coverage.

This will be a highly hands-on project, involving tasks such as designing and building the electronics, implementing algorithms on low-power embedded systems, and managing the overall power budget.

Location: Radio Systems (405.736, Lab)

This project will be used to address the scarcity of available datasets for the detection of Mild Cognitive Impairments through audio analysis. Students will develop Machine Learning (ML) techniques for data augmentation to increase the diversity and volume of audio samples in existing datasets. They will explore different ML methods to create synthetic but realistic audio samples. Additionally, the project will involve generating transcripts from audio data using state-of-the-art automatic speech recognition techniques. The student will also refine and validate the generated transcripts to ensure their further usage in linguistic analysis and perform model development and evaluations.

This project will result in the generation of a dataset containing augmented audio and their transcripts, which can be used to train ML models in the future, and train ML models. The students will create a foundation for deeper linguistic and acoustic analysis by completing this project. This project aligns with ongoing efforts to use AI for early detection of dementia, offering students a practical experience with audio preprocessing and annotation tools, and deep learning techniques.

Location: HASEL (405.662, Lab)

Data privacy has become a pressing concern in the era of rapid technological advancements and the increasingly ubiquitous use of generative AI. The proliferation of online platforms and cloud storage systems has heightened the risk of personal information being collected and used without consent. While encryption methods exist for data storage and transmission, many organisations lack transparent data protection policies, leaving users vulnerable. Current methods often address encryption and compression separately, and rely on complex algorithms to achieve higher security levels.

This research explores the potential of compressive sensing (CS), a novel signal processing technique, to develop an integrated framework for secure data encryption and compression. CS leverages the sparsity of signals to encode and reconstruct data efficiently, offering inherent compression and encryption properties. Unlike traditional methods, CS can potentially perform these functions simultaneously, thereby reducing computational complexity while maintaining equivalent levels of security.

The study aims to evaluate how CS can address user privacy and server efficiency by investigating its ability to compress data and encrypt it to facilitate secure storage and transmission, whether through insecure channels or with untrusted cloud-based systems. This approach could offer a streamlined, cost-effective solution to modern data protection challenges, meeting the needs of both users and service providers. By combining efficiency with robust security, CS has the potential to redefine data privacy frameworks in the digital age.

Location: Control & Signal Processing (405.722, Lab)

A P4P in 2024 identified difficulties faced by first time users of debuggers. This project will use the results of that project to redesign some debugging features to be more inclusive and usable. Students would create a plugin for PyCharm with their redesigns and run an experiment with a new set of first time users to see if the redesigns make the debugger easier to use.

Location: HASEL (405.662, Lab)

Modern power electronics and computation demand improvements in density and efficiency. One way to achieve higher efficiencies is to move to cryogenics, that is, cooling to temperatures below -150 C. However, most commodity microcontrollers are rated at a minimum operating temperature of -40°C; anything below this is uncertified and can lead to decreased reliability and stability issues. The primary aim of this project is to understand the root causes of any decreased reliability under cryogenic temperatures. The secondary aim is to suggest, and if possible implement, solutions to these root causes. These results will be applied to improving the reliability of a relevant system, such as the computation of a complex algorithm, or a motor controller.

Location: Green Electronics (405.712, Lab)

The Ko Au nanosatellite mission is a student-led project to develop a 1U demonstrator for the Extremely Low Resource Optical Identification (ELROI) system developed by Los Alamos National Laboratories (LANL) (1). ELROI is intended in its final form to be a standalone module that could be attached to any satellite bus, transmitting a unique 127-bit identification code to allow for rapid identity verification. Current prototypes utilize diffuse laser light over a 140° angle, necessitating that the satellite point at the Earth to enable detection. Ko Au is intended to take this concept and transform it into an omni-directional broadcast signal from a 1U CubeSat, removing the need for all command-and-control functionality (2).

The lack of communication with the satellite necessitates a higher degree of robustness and autonomy than is typical for nanosatellites. This has prompted the decision to add redundancy by the inclusion of multiple onboard computers (OBCs), to make the system capable of tolerating the harsh radiation environment found in low Earth orbit. Traditional redundant systems are costly and too large for the proposed 1U platform, and so we propose the use of commercial-off-the-shelf (COTS) microprocessors and a software-based, decentralized redundancy architecture. Due to the limited power available on Ko Au, this must be a “cold” redundancy system, where only one OBC is active for the majority of operations. Such a system has been developed by York University for similar OBCs.

Location: Embedded Systems (405.760, Lab)

Inspecting plants for disease is a tedious and error-prone process that takes away farm workers’ time that could be better spent on other tasks. In previous years, students have investigated the development of a system to identify diseased tomato crops in indoor farms with a drone equipped with Computer Vision (CV), and Wireless Power Transfer (WPT) for easy charging without intervention required. We will investigate whether we can improve the accuracy of disease recognition through improving the CV training dataset and image preprocessing. We will also investigate methods to reduce the weight and size of the drone to increase flight time between charges, and subsequently investigate coil designs to optimise power transfer.

Location: Power Electronics (405.614, Lab)

In today's world, AI technologies like ChatGPT, Siri, and other generative AI systems have seamlessly integrated into our daily lives. These tools are used for conversations, learning new content, and solving complex questions. However, these AI systems are limited by their inability to express emotions and engage in natural, human-like interactions. While humans convey emotions through facial expressions, tones, nuances, and hand gestures, AI tools struggle to replicate these crucial aspects of communication, making interactions less comfortable and natural.

Building on the existing use of Large Language Models (LLMs) in conversational AI, our project aims to address these limitations by developing an advanced co-speech motion generative AI system for robots like Pepper and Silbot. Our goal is to create a dynamic and adaptive motion generation mechanism that responds to various inputs, such as keyframes, cameras, and voice commands, producing fluid and contextually appropriate movements. This will enhance the overall interaction experience with robots, bridging the gap between human and machine communication.

Beyond human-robot conversations, we expect this system to enable robot-to-robot communication in ways that are comprehensible to humans. Instead of relying solely on signal-based exchanges, robots will communicate using gestures, motions and voice in visually intuitive ways. This capability's core is the co-speech motion generative AI engine, which allows robots to synchronize verbal and non-verbal communication. This innovation bridges the gap between robotics communication and human understanding, enabling people to observe or participate in conversation between robots. For example, multiple robots could engage in a debate or collaborative discussion while clearly expressing themselves through verbal and non-verbal cues. Furthermore, this opens opportunities for different research areas related to education/debating.

Location: Robotics (405.652, Lab)

This project will develop a robot system to harvest targets, such as tomato and strawberry in smart farm environment. We have a working version that detects target, which is tomato, and measures its size and volume. We will have dfiferent targets, such as strawberry, to harvest targets using Deep Neural Network (DNN) models. Also we will design and develop different end-effectors, such as gripper to harvest targets. All necessary devices will be provided.

This project includes to:

• start from the current working version,
• detect targets, measure the ripeness, size, volume, and estimate the harvesting timing,
• design and develop robot harvesters.

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have an experience on vision processing and/or using DNN and 3D design/printing.

Location: Robotics (405.652, Lab)

This project will develop a robot system to express signs with facial emotion expressions. We have a working face robot and hand robot that express facial emotions and finger motions. We also have a well working sign understnading engines. This project will use all working systems to integrate a sign language expression robot system. May need to train Deep Neural Network (DNN) models to improve the accuracy of existing methods. All necessary devices will be provided.

This project includes to:

• start from the current working version,
• control a face robot to express emotions,
• control a hand robot to express signs,
• integrate two robots with sign understanding engine, and
• develop a GUI to show interaction and results.

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have an experience on vision processing and/or using DNN.

Location: Robotics (405.652, Lab)

This project will develop a newer version of robot soccer system that used low-cost motors with 3D printing. We have a working robot soccer system includes GUI, strategy SW, vision system SW, and robot system. We will update the current SW to be able to use AI models to generate strategy and control robots. May need to train Deep Neural Network (DNN) models to improve the accuracy of existing methods. All necessary devices will be provided.

This project includes to:

• start from the current working version,
• develop an AI based strategy system, and
• design and control new soccer robots.

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have an experience on vision processing and/or using DNN, and 3D design/printing.

Location: Robotics (405.652, Lab)

This project will develop a newer version of healthcare robot system based on the current working version. This project will convert the current Windows based SW to Ubuntu environment, and control robot arms and head to express its emotions. We will add face detection/recognition and chatbot if needed. May need to train Deep Neural Network (DNN) models to improve the accuracy of existing methods. All necessary devices will be provided.

This project includes to:

• start from the current working version,
• convert Windows based SW to Ubuntu environment,
• control robot arms and head to express emotions, and
• integrate existing chatbot and face processors.

Project scope will be decided after the meeting with supervisors. You will bring to the role a passion for research and engineering, excellent computing skills (including a high level of programming ability), and a strong sense of responsibility. It will be better if you have an experience on vision processing and/or using DNN.

Location: Robotics (405.652, Lab)

Machine learning (ML) algorithms for artificial intelligence (AI) have been used for many applications. ML models for AI are trained based on the application-specific dataset and then deployed on target execution platform to perform the required tasks such as object recognition and tracking. ML algorithms are computationally intensive so, deployment of these algorithms for embedded systems or devices at the edge is very challenging as not only achieving the required system performance and high-level of accuracy are essential, reducing energy consumption is also critical. As the complexity of ML models are changing reconfigurable and software-controlled hardware accelerators are required to manage and reduce the design complexities.

The aim of this research is to investigate the computational complexity of machine learning algorithms, especially, deep learning algorithms for embedded applications (such as object recognition) and develop a reconfigurable hardware accelerator, which is implemented on FPGA to satisfy the required performance and accuracy while minimizing the energy consumption.

Location: Embedded Systems (405.760, Lab)

Deploying machine learning (ML) algorithms on resource constrained devices to achieve a high-level of accuracy for practical cases is very challenging as the availability of limited resources has significant impact on the quality of results. TinyML is ML optimized to be deployed on resource constrained devices for AIoT (Artificial Intelligence of Things) applications such as Industrial IoT, healthcare devices, environment monitoring systems and agricultural applications. As the complexity of ML algorithms and the data size for TinyML algorithms are reduced compared to ML algorithms, their effectiveness for the target applications should be investigated.

The aim of this research is to deploy TinyML algorithms on FPGAs implemented as customized multi-processor systems on chip (using softcore Nios V processors, and hardware accelerators) or an extended single core processor and evaluate the performance and accuracy for some embedded applications. Architectural techniques at the instruction set architecture (ISA) level such as using Single-Instruction Multiple Data (SIMD) and Vector Processing will be investigated.

Location: Embedded Systems (405.760, Lab)

Large Language Models (LLMs) are achieving remarkable results in generating code from natural language prompts. While they  are effective for simple tasks, their ability to handle complex software systems—those involving multiple classes with interrelated functional and non-functional requirements—remains limited. This research project aims to explore and propose new methodologies for generating such complex software systems using LLMs. In particular, we will investigate how to specify complex requirements, break down complex tasks into smaller subtasks that LLMs can manage, combine the results, and verify the correctness of the produced software. Throughout this process, we will maintain a common context to ensure that the smaller components remain compatible and integrate well into a coherent system.

Location: HASEL (405.662, Lab)

Computing concepts encompass a broad range of topics, each with varying levels of complexity. Some key areas include, but are not limited to, algorithms and data structures, software architecture and operating systems. Certain concepts within these fields can be particularly challenging due to their abstract nature or counterintuitive behaviour.

This project aims to explore novel and innovative ways to communicate complex computing concepts, leveraging existing and modern techniques, such as visualisation, and AI/LLMs, with the latter focusing particularly on its multimodal input capabilities. To address this, a literature review will be conducted to identify a specific computing concept that could benefit from improved communication and learning methods. The goal is to develop a solution that effectively addresses the challenges of understanding the target computing concept, particularly in areas where existing tools fall short. The effectiveness of this research and tool will be evaluated based on its impact on students’ comprehension of the chosen concept.

Recent and past research has demonstrated that visualisation can help aid people's understanding of computing concepts. Additionally, advancements in AI and LLMs have shown improvements in visual and spatial reasoning. This project seeks to further advance this field by investigating how a combination of visualisation and multimodal AI can improve the communication and understanding of complex computing concepts.

Location: MDLS (405-559)

Free Space Optical Communication (FSOC) is revolutionising satellite-to-earth data transmission by leveraging infrared laser light instead of conventional radio frequencies. This approach enables significantly higher bandwidth while reducing the size and weight of transmitter and receiver hardware. However, the commercial adoption of FSOC faces a critical challenge: maintaining a cloud-free line of sight (CFLOS) between satellites and optical ground stations (OGS).

A network of all-sky cameras that capture detailed images of cloud coverage is operated in Auckland and at OGS sites across Australia and New Zealand. These all-sky images, combined with other data sources such as satellite imagery and meteorological observations, offer a rich dataset for cloud analysis. While we have been provided an existing algorithm from NIWA that identifies cloud coverage during the day, we currently lack the ability to predict short-term future cloud cover along satellite passes. Addressing this limitation involves software-driven research into integrating these diverse datasets, advancing image processing techniques, and developing robust predictive modelling approaches. The project focuses on applying both traditional statistical approaches and machine learning techniques to develop a robust cloud cover prediction system. This includes optimising deep learning models for geophysical phenomena, integrating real-world environmental variables, and balancing computational efficiency with accuracy. The insights gained through this investigation will extend beyond FSOC, contributing to broader challenges in geophysical modelling such as weather forecasting, and environmental monitoring.

Location: HASEL (405.662, Lab)

Background

Large parts of the world have no cellular coverage, making it difficult to monitor environmental conditions and hazards such as air quality, groundwater levels, or wildfire risks. These areas mostly rely on manual collection, which is expensive, slow, and can be dangerous. For example, the 2024 Bridge Hill/Castle Hill fire in Canterbury spread over 980 hectares partly because it wasn’t detected early enough.

Real-time data is also important for people in remote locations, like farmers and hikers, who rely on accurate information to stay safe and make decisions. While satellite services (e.g., Starlink) can provide internet access, they typically need more than 10 watts of power, which is far too high for low-power IoT (Internet of Things) sensors that usually work at milliwatts or even less. Many such sensors run on tiny batteries or have no batteries at all, making it essential to find a communication method that uses very little energy.

The Technology

This project explores how to use sunlight for communication in remote sensing. Instead of creating a new signal, Ambient Light Backscattering (ALB) uses existing light sources, like sunlight, to transmit data. Traditional radio systems need to both create and encode (modulate) their signals, which uses more power. ALB only needs to encode data into the light already present, greatly reducing power consumption.

The system works by using special surfaces (such asliquid crystals shuttesr in 3D glasses) that quickly switch between being transluscent and opaque. This creates an “on-off” pattern in the light (i.e. amplitude modulation), which human eyes can’t detect but photodetectors can pick up and decode into data. This approach enables extremely low-power communication, making it ideal for remote, low-power sensors.

This Project

Traditional low-power communication schemes (such as BLE) need to wake up periodically to see if there’s any data to receive. This polling approach wastes energy if no data is being sent. With (ALB), we flip this around: receivers stay asleep until another sensor actually has information to share.

1. Wake-Up Step

   • A sensor that’s ready to send data can redirect sunlight toward a neighboring sensor.
   • By focusing or reflecting enough sunlight, it provides enough energy to “wake up” the receiving sensor.
   • This is similar to how RFID tags are powered by a reader’s signal, but here, sunlight serves as the power source instead of a radio wave.

2. Communication Step

   • At the same time, the transmitting sensor modulates this sunlight beam to embed data.
   • The receiving sensor, now awake and powered by the incoming light, decodes these light patterns.

Because sunlight is much stronger than RFID signals, ALB can operate over longer distances than RFId. Sensors don’t need to waste energy on periodic polling; they remain idle until another sensor initiates communication.

Location: Radio Systems (405.736, Lab)

Introduction & Motivation

Diabetes is a lifelong condition expected to affect 783 million adults by 2045. Existing blood glucose tests often depend on finger pricks (which are invasive and inconvenient) or Continuous Glucose Monitors that require microneedles and costly sensors ($250/month). These methods measure only glucose and ignore other vital indicators of diabetic complications, such as oxygenation and blood flow, which are crucial for spotting issues like neuropathy and retinopathy before they become severe.

Why Photoacoustic (PA) Sensing?

PA sensing is a novel way to measure multiple biomarkers, beyond just glucose, by combining the strengths of optical and ultrasound techniques. In PA, short pulses of light (~ns) enter the tissue, where they are absorbed by blood. This rapid absorption (when light is on) and release (when light is off) of energy generates ultrasonic waves, which travel back to a detector on the skin

Because the amount of light absorbed depends on specific biomarkers in the blood, these ultrasound signals reveal important data about them like oxygen levels and glucose concentration. Unlike purely optical methods, PA can probe deeper into tissue (several centimeters instead of just a few millimeters), while still providing higher specificity than traditional ultrasound, making it uniquely suited to early detection and comprehensive monitoring of diabetic complications.

This Project

Typical PA systems use high-power lasers that are bulky, expensive, and require specialized cooling and safety measures. In this project, we will explore whether LEDs can replace these lasers and still achieve reliable measurements. Such a shift introduces trade-offs, including how to generate enough light energy, how to handle potential low signal /high interference, and how to integrate everything into a small, comfortable wearable form factor. Solving these challenges could make PA technology far more accessible, paving the way for continuous health monitoring devices that track multiple biomarkers without the invasiveness or high cost of current solutions—an exciting prospect for students interested in biomedical engineering, electronics, and sensor innovation.

Location: Radio Systems (405.736, Lab)

In this project, you will develop and evaluate a system designed to recognise as many hand signs as possible in real-time and as accurately as possible. Your goal is to determine what is practically possible on a lightweight system such as a Raspberry Pi.

As a likely starting point, you will probably investigate installing and training a chosen generic machine learning framework on a Raspberry Pi  (Tensorflow lite, YOLO, MLPack, MediaPipe). Stronger students are expected to use problem domain-specific elements to try and elicit further performance of their system.

This project has plenty of room for creativity and attempts at novelty could lead to an academic publication should you make enough progress. It should be noted that there are many publications that claim to do sign language on a Raspberry Pi, but the vast majority of them can only recognise a small number of signs - so this is a clear aspect that can be challenged by your work.

Location: Robotics (405.652, Lab)

In recent years, significant research efforts have been dedicated to greater power transfer in Inductive Power Transfer (IPT) systems in a given footprint, specifically within the transportation wireless power system sector. Higher power transfer levels are essential in providing EV users with a quicker and non-disruptive charging experience. Such systems require some method of cooling, such as passive air cooling or liquid cooling. These methods display limitations in terms of power density, sustainability and size. In particular, the inverters and magnetics within the IPT systems are placed under high thermal stress. Therefore, other effective methods of cooling and regulating temperatures are required for a more reliable level of operation.

This project aims to explore a new method of cooling that combines a thermally and electrically conductive liquid metal coolant with magnetohydrodynamic (MHD) technology, which holds the potential to overcome the above-mentioned thermal challenges. The focus of the research is to compare cooling techniques that are currently in use with the new proposed design in terms of thermal management and efficiency. ANSYS will be utilised to create a thermal model, analyse the effectiveness of cooling, and design an electromagnet. Control circuitry will be developed on the electromagnet for flow-rate control and reading of the liquid metal coolant, and a simple cold-plate and radiator design will be implemented to aid in the cooling of the electronic system.

Location: Power Electronics (405.614, Lab)

Have you ever struggled with understanding lecturer’s speech in a class? Yes, then what if you had hearing impairment or were studying in a language that is NOT your first language?

Acoustics of rooms such as background noise and reverberation affect our hearing detrimentally by degrading the intelligibility of speech, making it harder to understand what speakers are saying. When the listener has hearing impairment or the language of the speech is not the listener’s first language (i.e. non-native listener), the detrimental effect will be much more severe, causing inequity in accessing crucial information between listeners. This inequity is particularly serious in education sector in the modern society where students with various backgrounds are learning together.

Studies have shown that the acoustics varies not only between rooms but also between different positions in the same room. It can also be improved by installing equipment such as public address systems (microphone and loudspeakers). This Part4 project is a continuation from the previous year which has been investigating how the acoustics of different learning spaces in the university affect the speech intelligibility of listeners with different language background (e.g. native vs non-native).

The project co supervisor will be Asscoiate Proffessor  Yusuke Hioka, who leads the Communication Acoustics Lab (CAL) at the Acoustics Research Centre

Location: Control & Signal Processing (405.722, Lab)

Speech is an incredible signal containing so much information: it not only contains communicative meaning, but it also contains information about the speaker.  Speech processing techniques can extract this information. One such application is being able to provide real-time feedback for pronunciation, which can be used in language learning applications ( see for example MPAi (https://ctjustinehui.github.io/MPAi_2024/app.html#/audiopermission) from the Curious Minds project “Learning the science of speech through te reo Māori” https://www.canva.com/design/DAGGkZnYEBo/4lT1haX4lenGwclLl2VdRA/view). One particularly powerful speech feature is formants, which are a measure of the vocal tract responses (if we think of the speech system as an acoustic chamber). Formants ( which are used in MPAi) are traditionally via methods that involve calculating or approximating the magnitude spectrum. In fact, the magnitude spectrogram, and it time varying extension – “the spectrogram” - have dominated speech technology over the last few decades. However, this trend has overlooked the fact that the phase spectrum is also a very powerful acoustic analysis tool. This project will involve creating a formant tracking algorithm via analysis of the phase spectrum, and then seeing how well in compares to existing formant trackers, including hand checked data.

Location: Control & Signal Processing (405.722, Lab)

This project aims to explore the feasibility of utilising blockchain technologies to create secure, transparent, and anonymous voting systems tailored for New Zealand’s election process. Elections play an important role in democracy, and the integrity of voting systems is paramount to ensure trust in the process. By leveraging blockchain, we explore solutions to address key challenges in the main topics of transparency, security, and voter anonymity.

The research will investigate various blockchain strategies, architectures, and tools to determine their applicability within New Zealand’s unique social-political and technological landscape.

Location: Radio Systems (405.736, Lab)

Modern software development frequently relies on unit tests that use mocks and stubs to simulate dependencies (e.g., Mockito in Java). By isolating the class under test (CUT) from external components, mocks and stubs help developers verify the CUT’s behavior in isolation. However, human-written test classes with mocks/stubs often encode valuable information about the dependencies—such as expected method calls, input parameters, and exception handling.

This project aims to leverage that embedded information to automatically generate test cases that exercise the dependencies in realistic scenarios. The central research questions are (i) How to extract and utilize mock/stub details to produce test cases that effectively evaluate the dependencies themselves?, and (ii) Does this approach generate effective and useful tests?  To succeed in this project, strong proficiency in Java and JUnit is essential. In addition, the project will explore the use of Large Language Models (e.g., GPT) for extracting the relevant information from stubs and mocks.

Location: HASEL (405.662, Lab)

Dysarthric speech is a motor speech disorder that makes speaking challenging for individuals with dysarthria. This condition affects muscle movements and often leads to distorted speech. It presents considerable obstacles for standard Automatic Speech Recognition (ASR) systems, which struggle to perform effectively under these constraints. These systems often fail to accommodate dysarthric speakers due to two key limitations: (1) collecting large speech datasets from individuals with dysarthria is impractical due to health-related fatigue and high inter-speaker variability and (2) underrepresented error patterns (e.g., consonant omissions, vowel distortions, irregular pitch), which reduce recognition accuracy. As a result, these systems present accessibility barriers, reducing the feasibility of their application for individuals with dysarthria.

This project aims to develop a novel data augmentation framework that synthesizes dysarthric speech by simulating disorder-specific errors in non-dysarthric datasets. By embedding common pathological patterns (e.g., consonant omissions, slurred phonemes) into healthy speech data, the method will generate training corpora to improve ASR robustness for dysarthric users.

This research can potentially advance voice technologies, enabling more effective speech recognition. Improved systems could significantly enhance communication and overall quality of life for individuals with dysarthria.

Location: HASEL (405.662, Lab)

The ability to unambiguously monitor and identify the location of maritime vessels is critically important in terms of illegal activity detections and maritime security for New Zealand.  Although tracking systems such as the Automatic Identification System (AIS) already exist, they are inherently insecure and transmissions can be intentionally altered to obscure the real positions of vessels.     To address these concerns, this project will investigate the feasibility of a receiver-driver non-cooperative validation framework for the maritime Automatic Identification System (AIS) that utilizes Time Difference of Arrival (TDOA) and Received Signal Strength Indicator (RSSI) analysis to independently verify AIS-reported geolocations, detect anomalies, and differentiate between systematic error and deliberate interference in order to enhance confidence in AIS data integrity.

Location: Radio Systems (405.736, Lab)

This project will focus on the design, development, and characterization of antenna systems for a Cubesat platform, enabling reliable telemetry, telecommand, and data transmission. The project will address the critical need for robust communication links in Cubesat missions, focusing on two distinct frequency bands: UHF for omnidirectional telemetry and telecommand, and S-band for directional, high-bandwidth data downlink.

The project will involve a comprehensive exploration of antenna design principles, trade-offs, and practical implementation considerations. Emphasis will be placed on achieving wideband performance in both UHF and S-band antennas to accommodate potential frequency variations and maximize data throughput. The student will investigate existing antenna designs, including patch antennas and other candidate structures, and develop innovative approaches to enhance their bandwidth and radiation characteristics. The availability of an in-house developed transceiver will be considered to inform design decisions.

Location: Radio Systems (405.736, Lab)

Project Overview:This project aims to develop a luminaire with tuneable white LED lighting powered wirelessly. Tuneable white lighting enables the adjustment between warm and cool light, which is essential for visual comfort and supporting the circadian rhythm.

Research Component:The project will explore principles for transmitting both power and control signals (for colour selection) to the luminaire wirelessly.

Location: Radio Systems (405.736, Lab)

Project Description:

Indoor farming depends heavily on artificial lighting to maintain optimal growing conditions, but this reliance can lead to light pollution, unwanted or excess light that results in energy waste, uneven luminance, and reduced system efficiency. This project explores light pollution within indoor farming environments, investigating how it impacts energy consumption, uniformity of illumination, plant growth, and overall profitability. The study focuses on evaluating key parameters such as illuminance distribution and uniformity. Possible directions for investigation include the use of adaptive technologies such as sensor-based feedback systems or dynamic lighting controls to manage light spillage. Additionally, the study aims to analyse how factors like spectral quality, surface reflectivity, contribute to energy efficiency and crop yield. By considering a range of potential solutions and their economic and environmental trade-offs, the project seeks to propose adaptable strategies for reducing light pollution. Exploring cutting-edge technologies or simple, scalable methods, depending on the findings. By providing a comprehensive understanding of light pollution in indoor farming, the project aims to contribute to the design of smarter, more efficient lighting systems that support sustainable agriculture and resource management.

Location: Radio Systems (405.736, Lab)

The internet is now crucial to all our lives, and whilst the importance of digital inclusion is well recognised in New Zealand, the level of service people have access to is not equitable. The difference in service between rural and urban New Zealand is well documented, with the rural services proving lacking. But even within an urban environment is it equitable across the whole city? In 2024 we examined digital equity in 38 Auckland public libraries and found that there was significant regional variability, with the western and northern libraries offering higher and more reliable speeds.

This year we which to continue the investigation, this time looking at our schools. This project is going investigate the quality of internet services across education via a series of experiments which will be designed to measure the service quality. To increase number of schools involved in the study, the project will not only perform the analysis of the internet services, but it will also involve devising a way to be able to conduct the measurements remotely.

Location: Control & Signal Processing (405.722, Lab)

The project explores fusion of a commercial software for building fixed industrial automation systems (IAS) and novel software paradigm that makes those systems dynamically reconfigurable. Dynamically reconfigurable systems support advanced features of IAS such as dynamic change of the products on the same production line, fault-tolerance and changes in physical configuration (adding, removing parts of the system). As the starting point, it uses Beckhoff commercial hardware and software platform for industrial control systems based on a network of industrial PCs and interfacing nodes connected through EtherCAT, an industrial proof networking system, and TwinCAT software framework for development of control systems.

The new software paradigm uses SystemJ concurrent programming language for the development of control that supports (dynamic) reconfigurability to tackle specific scenarios such as failure of physical parts of the system and fail-safe continuation of its operation, change of production process and products, etc.  The new paradigm will enable modelling the system operation and simulation before the deployment of software on a real platform with practically no changes of the developed controller model. As a testbed, prototype manufacturing system in Industrial Informatics lab can be used.

Location: Embedded Systems (405.760, Lab)

We are exploring the use of digital twins (DT) enhanced with Machine Learning (ML) in discrete industrial automation systems (IAS) control. The project explores how DTs can enhance control and efficiency of IAS and how ML can further influence control in systems that use complex sensors, such as cameras. We envisage IAS control based on interplay of three types of DTs: (1) DTs of workpiece which after completion of the “recipe” becomes a final product, (2) DTs of workstations that make physical changes and transformations on the workpiece as required by recipe for specific product, and (3) DTs of transportation system devices (such as conveyors, turntables, arm manipulators, temporary storage bays, and robots). Using DTs makes possible a completely new production paradigm and formalisation of IAS that produces final products by applying discrete operations as specified in the recipe. Complex sensors can be used to better characterise current status of production process and enhance performance of the overall system.

Location: Embedded Systems (405.760, Lab)

The project focuses on the development of a time-predictable computer architecture that incorporates traditional and custom processors, as well as application-specific processors on a chip. The platform will be suitable for combined control- and data-dominated embedded real-time systems that rely on ability to deal with time and safety critical reactive applications that also include significant use of digital signal processing.

The project will explore the architectural alternatives of satisfying the reactive and data processing requirements of the advanced embedded systems and result in the development of that execution platform based on time-predictable Network-on-Chip (NoC) that accommodates between 8 and 16 nodes (processor cores, application-specific processors, accelerated functions).

Location: Embedded Systems (405.760, Lab)

Plasma speakers, as a cutting-edge method of sound absorption, actively removing noise pollution at all frequencies. These systems, as demonstrated by Sonexos (https://sonexos.com/), have the potential to revolutionize audio systems through their unique design and innovative sound absorption mechanism. This research project aims to explore and optimize the design of amplifiers specifically for plasma speakers, with a focus on high voltage Class B and Class D amplifiers. The goal is to enhance the performance of plasma speakers by improving their efficiency, and power handling.

Location: Green Electronics (405.712, Lab)

As electric vehicle popularity grows, wireless power transfer systems are becoming more desirable. This is due to their convenience and ease of use compared to conventional plug-in charging systems. The current WPT systems have flaws in achieving high power efficiency and alignment correction while being compatible with different EV models.

Building on existing multi-coil designs, this project aims to use auxiliary coils to achieve synchronisation and improve alignment in WPT systems. The auxiliary coils are used to generate or detect synchronization signals to allow for control over the timing and efficiency of power transfer.

The project begins with a review of current synchronization methods in inductive power transfer systems, comparing techniques for transmitting synchronization signals. Once the optimal synchronization approach is selected, computer simulations will be modeled to evaluate system performance, followed by the design and testing of a prototype to assess real-world functionality. This study aims to advance WPT technology by incorporating auxiliary coils into WPT systems for synchronisation and alignment.

Location: Power Electronics (405.614, Lab)

Many people need help with their daily lives by reminding and suggesting ativities and information based on recent experiences. This project will use a wearable camera and microphone to record daily activities in real time, and make suggestions and answer questions using large language models, to help the wearer achieve their daily goals and activities. For example to remind people to do simple daily tasks, and remind people where to find objects in their environment.

Location: Robotics (405.652, Lab)

This project will take the first step in designing an advanced companion robot for interacting with and helping humans. This will involve a survey of previous and current robots for companionship, and of research in companion robot design, compiling a specification of significant necessary and desired features. Then proposing physical shape options for a companion robot, and making prototypes.

Location: Robotics (405.652, Lab)

This project is about the analysis and design of robots in architecture, including understanding previous research and implementation of robots in architecture, and design of potential future robotics in architecture. This includes the use of robot in creating architectural design, and the use or robots as components of architecture itself. An example prototype should be created.

Location: Robotics (405.652, Lab)

This project is to initiate a longer term effort in designing robots to help people in space. This initially requires a literature search and analysis of possibilities for robots to: 1) help in operations in space, such as maintenance and construction for a space station, and 2) to help on other moons and planets for creating and operating structures to enable humans to live in the environment, and 3) for robots to interact with humans and be useful companions in space environments, and 4) what other applications may be significant. After such analysis, a prototype design for at least one application should be made, and if possible built in the lab as a lab demo.

Location: Robotics (405.652, Lab)

This project is to develop a Motor / Dyno test platform for an inexpensive drone motor. The goal of the project is two fold, to model and characterise an inexpensive drone motor and to develop the mechanical hardware, power electronics and control system to do so. This project will give students a deep understanding of drone motors, the power electronics and control to drive them and also experience with the rigor of developing a reliable and repeatable test platform. A 3d printed prototype exists (not yet functioning) but it requires students with an interest and a bent towards mechanical systems, power electronics, control and scientific rigor to make it sing. This will be used to validate simulations using a known motor geometry to then have confidence in simulations of future superconducting electric machine for future passenger aircraft. It also has significant value in teaching undergraduate and postgraduate students and in hands on demonstrations.

Location: Green Electronics (405.712, Lab)

One view of software quality is to examine the cost of how a software product changes, or evolves, over time. Understanding software evolution is difficult because long-lived software products are usually quite large, and change in different ways, some of which are related to the quality of the product and so are interesting, and some that are not interesting.An issue that is often mentioned when discussing why some designs are costly to change is that the dependencies between different parts of the design are too high. This suggest that it may be useful to understand evolution by understanding how dependencies change.The goal of this project is to investigate how the evoluton of a software product can be understood by determining how dependencies in the product changes. There are already dependency datasets for multiple releases of several open-source Java products.

Location: Control & Signal Processing (405.722, Lab)

This project focuses on improving spatial visualisation skills, crucial for success in STEM fields, by developing a tool that automatically generates 3D models and corresponding orthographic/isometric projection questions. The aim is to enhance question generation and difficulty control of existing approaches by generating diverse 3D models. This project will also explore the integration of Large Multimodal Models (LMMs) to understand better their value in spatial visualisation tasks.

Joining the supervision team will be Burkhard Wuensche (Computer Science), Stephen Kavermann (Mechanical and Mechatronics Engineering), and Hazim Namik (Mechanical and Mechatronics Engineering).

Research Aspects:

* Revise the existing shape generation to produce more complex and varied 3D models, ensuring a wider range of practice material.

* Develop heuristics to deliberately create questions of specified difficulty levels, improving the tool's adaptability to different student needs.

* Explore the use of LMMs with respect to the spatial visualisation questions.

* Conduct testing to evaluate the tool's effects on students' spatial skills and confidence.

Technical Skills Needed:

* Proficiency in Python for scripting and algorithm development.

* Experience with CAD software would be a bonus, particularly FreeCAD, but this is not essential.

Location: MDLS (405-559)

It is generally considered that code should have "good style". But exactly what this means is unclear. There are several style guidelines that are not entirely consistent either in what they cover as being good style, or are even appear to disagree about specific requirements. There as been an on-going research project within the School of Computer Science to address this problem. One result has been a "code style model" (CSM) that is intended to explain any proposed "style" guideline or rule [1].The general goal of this project is to evaluate CSM. There are a number of directions this could take. For example, "code checkers" such as PMD, Checkstyle, or Findbugs have a large number of rules and it is not always clear which rules to use for a given context. CSM should help answer this by providing a means to classify the rules according to the code quality attributes they apply to. There have been various proposals to "assess" style, and in some cases bespoke tools to support the proposals. CSM should be able to be used to evaluate these proposals.Evaluate the rules in tools such as PMD, Checkstyle, Findbugs, according to CSM, and assessment guidelines such as described in the literature, and used in local courses (perhaps contact other institutions for what they do?). Ultimately evaluate CSM and also construct some kind of categorisation. E.g. "If I want to evaluate these code quality attributes, what rules should I consider from PMD?"[1] Diana Kirk, Andrew Luxton-Reilly and Ewan Tempero 'CSM: A Code Style Model for Computing Educators' ACM Transactions on Computer Education, 2025https://doi.org/10.1145/3716861

Location: Radio Systems (405.736, Lab)

In the current landscape of technical hiring, platforms like HackerRank have become standard tools for assessing candidates. However, it is widely acknowledged that these platforms focus on algorithmic challenges that do not accurately reflect a candidate's day-to-day capabilities on the job. This mismatch creates a scenario where candidates can "grind" these assessments without necessarily demonstrating the skills required for real-world problem solving. As AI tools like GPT become more capable, the problem of candidates using these tools to cheat on assessments is only growing.

Given the rise of AI and the increasing pool of candidates, there is a pressing need for more effective online assessments. The project explores ways to better assess a candidate’s on-the-job capabilities by looking beyond traditional DSA (Data Structures and Algorithms) questions. One avenue we aim to explore is interactive assessments, such as the mini-games used by companies like Optiver, which gauge competencies like multitasking, decision-making, and strategic thinking.

The research will focus on how companies currently evaluate candidates, the limitations of these methods, and alternative approaches that provide a more accurate reflection of a candidate’s performance in the workplace. Additionally, the project will involve the development of a web app prototype that offers an innovative alternative to current assessment tools, potentially incorporating real-world scenario simulations or holistic problem-solving tasks. The goal is to provide companies with a more efficient, reliable, and accurate way to filter candidates before they enter the resource-intensive technical interview stage.

Location: Radio Systems (405.736, Lab)

With so much online content, and the availability of recordings of in-person content, it is becoming harder to get students in lecture theatres. We can always make attendance compulsory, but then what is an efficient way of checking attendance for large numbers of students? It would be even better to offer in-person content which is engaging enough to encourage students to attend themselves - but then, how do we measure engagement?

This project focuses on these research questions:

1. How can we measure attendance for large numbers of students?
2. How can we measure / quantify engagement in the content / topics being delivered?
3. How can we encourage participation & engagement in classroom activities?

Location: Radio Systems (405.736, Lab)

With the advent of GenAI and other high-performance AI / Machine Learning models, there is the opportunity to explore applications of this technology in CS Education. Possible areas of applicability - many of which already having substantial bodies of reasearch - include:

• AI-based virtual tutors
• Personalized assessment generation
• Automated student interviews to check understanding
• AI-powered automarking, either fully automated or to assist human markers

We would like to explore this space with a view to creating AI tools for students and / or teaching staff - initially focused on "bootcamp-style" programming courses with the potential to expand in the future.

Location: Radio Systems (405.736, Lab)

The Auckland ICT Gaduate School offers an internship programme for its students as part of it Masters in Information Technology (MInfoTech) programme. To help students find internships, we advertise CVs and videos online to prospective employers, who then identify candidates of interest. We then help them set up interviews with those candidates. In addition, we organize "speed networking" events where students looking for internships can meet a number of companies in a single evening. To avoid wasting people's time, we make every effort to match companies and students according to skillsets / areas of interest, to ensure the majority of this networking at least hast the potential to bear fruit.

The processes above are currently very manual, with companies and students giving us keyword-based lists of interest areas, then having our staff perform matching. There is also no good way that employers can search our database according to their particular interests. We want to leverage the power of modern GenAI / NLP systems to offer a more "natural language" approach to gauging how suitable a particular candidate may be for a particular company, by processing candidates' CVs and videos, along with employers' natural-language descriptions of ideal candidates.

Location: Radio Systems (405.736, Lab)

A significant frustration in the technical job search process lies in the lack of feedback when applications are rejected at the CV and interview stages. Our project aims to use Generative AI to assist students by evaluating and providing actionable feedback on their CV content and interview question answers. The proposed solution involves training a Large Language Model (LLM) to consistently give users accurate feedback and assistance with their job applications.To understand what makes CVs and interview answers effective, we will research existing resources such as academic papers, existing literature and online guides to learn what creates a strong application. We will also seek advice from industry professionals and recruiters to identify common trends, preferences and standout elements. In addition, keyword alignment and content relevance to specific job descriptions, we will utilise all this information and data to design and train our AI model.To goal for the LLM is to:

• Consistently provide constructive feedback on CVs and interview question responses, identifying and suggesting possible areas for improvement.
• Offer industry insights into expectations for CVs and interview question answers.

Our project will explore whether Generative AI can effectively assist with student job searches, being able to consistently provide actionable feedback on their CVs and interview answers. We hope to be able to give users the tools to refine their applications and increase their chances of progressing in the hiring process.

Location: Radio Systems (405.736, Lab)

Early detection of neurodegenerative diseases, with deep-learning based image classification.

Location: MDLS (405-569)

A key "pain point" for underwater robots is the localisation problem (determining where the robot is located in relation to the environment). In such a context, Lidar and GPS technologies do not work very well. Several alternative technologies have been developed to try and solve the problem, including the DVL (Doppler Velocity Logger) and the USBL (ultra-short baseline) system. However, these solutions tend to be costly and are limited in their robustness and accuracy.

This project focuses on a problem with an ROV with a single down-facing camera in relatively clear water (so that the sea floor is visible from the camera). This project aims to determine to what extent localization information can be determined from the camera footage, given the current state of technology and your own innovation skills.

An example of a proposed solution can be foundhere.

NOTE: This is expected to be a software based project.

Location: Robotics (405.652, Lab)

Recent advances in Machine Learning (ML) have resulted in widespread deployment of ML solutions in several areas. ML solutions are usually characterised by their high computational complexity. Heterogeneous computing platforms are introduced to optimise performance and energy efficiency. They consist of a range of processing units, including Central Processing Unit (CPU), Graphics Processing Unit, Tensor Processing Unit (TPU), and Field Programmable Gate Array (FPGA). Each of the processing units on heterogeneous platforms is sourced by its own specific voltage rail. This project aims to investigate the energy-efficiency advantages provided by AMD AI engine technology available on Versal adaptive SoCs. The research will investigate the impact of undervolting on the accuracy of ML solutions implemented on the AMD Xilinx Versal ACAP. The results will then be compared to the results of similar scenarios implemented on Zynq MPSOoC.

Location: Embedded Systems (405.760, Lab)

Investigating anxiety is a continuing concern within virtual reality (VR) research as it is considered an area needing continued development and evaluation (Geraets et al., 2022). Traditionally, subjective tools, such as clinical interviews, self-report questionnaires, and neuropsychological tests, are used to assess and diagnose anxiety. However, interviews and questionnaires strongly rely on a patient’s memory and current mental state (Geraets et al., 2022). Therefore, there is a need for ecologically valid assessments that apply objective measures (Hermans et al., 2018). Emerging technologies, like immersive VR simulations, may provide an opportunity to address this need. As such, this project aims to investigate the detection of anxiety within VR environments. While some research has been carried out on stress detection, the mechanism by which anxiety is objectively detected within VR environments or contexts has not been established. Moreover, this study will explore whether VR scenarios designed with culturally relevant stressors elicit measurable anxiety responses in Pacific peoples. Finally, this project aims to ensure the validity, safety, and applicability of VR-based assessments for indigenous populations.

Part IV student’s role:

• Design and implement VR experiments that simulate real-life experiences to study anxiety. 

• Collect and analyse physiological data (e.g. IMU, EEG, PPG) to assess anxiety and physical responses within VR settings. 

• Contribute to the understanding of the underlying motor-cognitive mechanisms associated with anxiety induction in VR. 

Location: Radio Systems (405.736, Lab)

Capsule networks (CapsNets) are a class of neural networks that address one of the main issues and deficiencies of CNNs. In CNNs, extracted features in initial layers pass through pooling layers where critical information, including spatial relationships, could be missed, resulting in incorrect classifications. CapsNets can better model hierarchical relationships and have shown great performance in many fields. CapsNets have a different architecture in their layers and, hence, different computation complexity. This project investigates hardware accelerations for different layers of this class of neural networks for real-time applications to identify any potential performance advantages provided by AMD AI engine technology, which is available on AMD Versal devices.

Location: Embedded Systems (405.760, Lab)

Emotion recognition is a key component of human-computer interaction, enabling AI systems to understand user emotions for applications in digital health, adaptive systems, and mental well-being. This project aims to develop a multimodal emotion recognition framework that integrates visual Large Language Models (LLMs) and biosignal analysis to detect emotions in real time.

Using wearable biosensors (e.g., heart rate variability, skin conductance, and pupil dilation) and video-based emotion recognition powered by state-of-the-art LLM-based vision models, the system will analyze facial expressions, physiological signals, and contextual cues to classify emotional states accurately. The outcome will be an AI-driven emotion recognition pipeline that enhances applications such as mental health support agents, adaptive learning environments, and real-time affective computing.

Part 4 Student’s Role:

• Develop a multimodal AI pipeline integrating Visual LLMs and biosignal-based emotion detection.

• Collect and analyze physiological + video-based data for training deep learning models.

• Implement and evaluate emotion classification models for real-time inference.

Required Skills:

• Machine Learning & Deep Learning (PyTorch/TensorFlow, OpenCV)

• Large Language Models (LLMs) (Fine-tuning & Prompt Engineering)

• Signal Processing (Physiological Data, Feature Extraction)

• Python Programming (Data Preprocessing, Model Training, Evaluation)

Location: MDLS (405-559)

Physiological synchrony, the alignment of physiological signals between individuals, plays a crucial role in effective collaboration, communication, and social bonding. This synchrony is often influenced by shared goals, coordinated actions, and the dynamics of social interaction. With the rise of artificial intelligence (AI) and the ability to monitor physiological signals in real-time, new opportunities emerge for designing AI systems that can actively foster and enhance human-human interaction. This project investigates the potential of using a reinforcement learning (RL) agent to improve physiological synchrony between individuals engaged in a collaborative problem-solving task. By dynamically adjusting the task environment based on real-time physiological feedback (specifically, heart rate variability), the RL agent will learn strategies to promote synchronised physiological states, potentially leading to improved teamwork and task performance. The project will compare the effectiveness of this approach to a control condition without AI intervention, providing insights into the role of AI in shaping social dynamics and the potential for technology-mediated physiological synchrony.

Student Role

• Design and implement a desktop-based collaborative problem-solving task (e.g., a virtual maze navigation task) that requires coordination and communication between two participants.

• Develop a reinforcement learning (RL) agent using Python libraries (e.g., Stable Baselines3, TensorFlow Agents) that can interact with the task environment and modify it based on real-time physiological data.

• Design and conduct a user study comparing a control condition (no AI intervention) with an experimental condition (RL agent actively modulating the task environment).

• Interpret the results and contribute to the understanding of how AI agents can be used to enhance physiological synchrony and improve collaborative problem-solving.

Location: MDLS (405-559)

This project involves iteratively refining an intelligent body tracking to a Unity-based installation where users experience immersive sound and visual feedback with motion capture and VR headset. The immersive application will display an intelligent agent that mimics variations of the visitor's movement or dance.  The practical goal would be to integrate an intelligent movement avatar into a gallery installation experience. The research goal would be to understand how the agents affects movement, sensory perception, embodiment, and subjective experiences. Over the year, we would customise the agent based on stakeholder feedback.

Location: MDLS (405-559)

One of the amazing things about the speech signal is we can use acoustic analysis of the signal to provide us information on how to change our pronunciation. We can relate acoustic features to the position of our tongue and jaw. With this knowledge in mind, we have created a platform to help people with their pronunciation of te reo māori, it does not use machine learning, but rather speakers are provided visual feedback  on their pronunciation via acoustic analysis of their speech.  The challenge we are facing is what is the best way to display the information to the users? People don’t think a lot about moving the articulators, and also there is quite a bit of upskilling required in helping people understand the acoustic plots.  Our platform is in two parts a web booklet providing people with the back ground in to the physics of speech productionhttps://www.canva.com/design/DAGGkZnYEBo/4lT1haX4lenGwclLl2VdRA/view, and then the pronunciation feedback app, MPAi, which is introduced in various forms through out the book, for instancehttps://ctjustinehui.github.io/MPAi_2024/app.html#/audiopermission. There are better ways of displaying the information from the analysis though, and this project is investigating this. It will involve both gui design, and a series of HCI experiments ascertaining whether people’s production of Māori vowel has improved, with the new GUIs. The co-supervisor will be Dr Justine Hui from mechnical and mechatronics.

Location: Control & Signal Processing (405.722, Lab)

Speech intelligibility assessment is crucial for evaluating the communication abilities of individuals with dysarthria, a motor speech disorder commonly associated with neurological conditions such as cerebral palsy, Parkinson’s disease, and ALS. Traditional assessment methods rely on subjective evaluations by speech-language pathologists (SLPs), which can be time-consuming, inconsistent, and resource-intensive. This project aims to develop an automated and objective intelligibility assessment system using deep learning techniques.

The ultimate goal of this research is to develop an accurate, scalable, and clinically relevant tool for intelligibility assessment that can serve as a reliable complement to traditional methods. The insights from this research could be used in speech therapy applications, assistive communication technologies, and longitudinal monitoring of speech disorders.

Location: HASEL (405.662, Lab)

The INTRACOMP project is an international and interdisciplinary initiative dedicated to fostering Intercultural and Transcultural Competence (ITC). This project emphasises the importance of understanding, respecting, and engaging proficiently with individuals from diverse cultural backgrounds. INTRACOMP seeks to tackle societal challenges by weaving cultural awareness into educational and artistic spheres. As part of this effort, one of the work packages focuses on developing digital artefacts and platforms to form a "digital playground" supporting ITC. This involves designing interactive environments that promote cultural comprehension and inclusivity in arts education.

The anticipated output for this research project includes creating an interactive application using Large Language Models (LLMs) where users engage with various personas and developing a mechanism to assess cultural awareness. To achieve these, students must be familiar with web development technologies, including frontend technologies (HTML, CSS, JavaScript) and backend server development (e.g., Node.js).

Research Aspects:

* Develop an interactive application that utilises LLMs to simulate cultural interactions.

* Implement a mechanism to assess cultural awareness and inclusivity in the application.

* Evaluate the application's effectiveness in promoting cultural comprehension and inclusivity.

Technical Skills Needed:

* Proficiency in web development technologies, including HTML, CSS, JavaScript, and Node.js.

* Experience with Large Language Models (LLMs) would be beneficial but is not essential.

Location: MDLS (405-559)

Project Description:This project explores how variations in spectrogram parameters affect the performance and interpretability of machine learning models in speech technology, particularly in automatic speech recognition (ASR). Spectrograms are widely used in speech processing as they transform raw audio into a time-frequency representation, making it suitable for AI-driven models to extract features. However, in many end-to-end AI pipelines, spectrograms are used blindly, with little understanding of how parameter choices—such as window size, hop length, and frequency resolution—affect model learning and decision-making. This project aims to investigate these effects systematically, improving the interprettability of AI-driven speech applications.

Research Component:The research will involve experimental evaluations of different spectrogram configurations on speech datasets, assessing their impact on ML model performance. Since there is limited understanding of how spectrogram design choices influence model behavior, this study will analyse their role in feature extraction and performance of the ML models. A key focus will be improving the interpretability of AI models by identifying spectrogram settings that lead to more reliable predictions. This work will provide insights into best practices for speech processing in AI-driven applications.

Required Skills & Interests:

• Inquisitiveness to question common practices in AI tech development and critically evaluate the impact of design choices
• Proficiency in coding (Python preferred)
• Interest in signal processing and speech technology
• Understanding of AI/ML models (or willingness to learn)

This project is ideal for students eager to explore the intersection of speech processing and AI, with a focus on interpretability.

Location: Control & Signal Processing (405.722, Lab)

Project Description:Do you speak multiple languages? Have you ever wondered if emotions are expressed differently in different languages? How does this impact technology? If you have ever switched between languages and felt that emotions “hit differently,” this project is for you! Using speech analysis and machine learning, we will explore how pitch, rhythm, and intensity vary across languages and whether AI models struggle to recognise emotions consistently across them.

Research Component:We will analyze emotional speech datasets in multiple languages, extracting acoustic features to compare how emotions are expressed. We will also attempt to record some new emotional speech databases - so knowing another language, and having some acting skills maybe a bonus. The project will also evaluate AI-driven emotion recognition models to see if they perform equally well across different languages—or if they struggle with some more than others. By understanding these patterns, we aim to make AI systems with emotion recognition more inclusive and accurate for diverse speakers worldwide.

Required Skills & Interests:

• Curiosity about how culture and language shape emotional expression
• A passion for questioning biases in AI and exploring why models behave the way they do
• Proficiency in coding (Python preferred)
• Interest in speech processing, linguistics, and human communication
• Basic understanding of AI/ML models (or willingness to learn)

Let us explore how language shapes emotion in speech—and whether AI is keeping up with the way humans truly express themselves.

Location: Control & Signal Processing (405.722, Lab)

Equanimity is a Virtual Reality simulation of a traditional compassion meditation that involves processing another person's emotional pain and suffering and generating a desire for happiness. It uses 3D visual animations of smoke and light to help the user link breath with intent, and the user is guided to pace their breathing along with the tempo of the visual animations. Ultimately, the goal is to help the user slow down their breathing improve their vagal tone and increase parasympathetic activation.

This project seeks a motivated intern to help with the post-processing and analysis of physiological data taken from two separate sensors (cardiovascular and respiratory). The analysis involves pairing events in VR with breath data to see if the user is accurately matching their breath to the 3D animations. Other measurements will be the change in the user’s respiration rate (hypothesized to decrease), heart rate (hypothesized to decrease), and heart rate variability (hypothesized to increase) over time. The student should have experience with Python, data processing, statistical analyses, and machine learning, and ideally, also interest and exposure to programming in C# and Python. Experience working with Unity is desirable but not essential.

Location: MDLS (405-569)

Replacing batteries is an effective approach for solving slow charging problems of electric vehicles (EVs). However, the mechanical constraints and safety concerns associated with direct electrical contacts pose significant challenges. This project aims to develop wireless battery or supercapacitor modules to enable quick and safe replacement of energy sources in EVs and other electrical devices.

The project involves designing wireless charging and discharging interfaces with battery management for safe operation and compatibility with vehicles. A scaled-down prototype will be developed and tested to demonstrate the feasibility of the concept.

Location: Power Electronics (405.614, Lab)

Wearable devices like smart glasses have recently gained popularity for their innovative features, but powering these devices is challenging due to the limited space for establishing safe and reliable electrical contacts.

This project is to develop a wireless charging box/case for charging low power wearable devices such as smart glasses and smart rings. The work includes a literature review of existing charging technologies, the design of a new inductive or capacitive power supply with energy storage, circuit simulation, and practical implementation.

Perspective students should have a very good background in electric circuit design, and at least one student should take the power electronics course. Successful candidates will work closely with postgraduate researchers in the research group.

Location: Power Electronics (405.614, Lab)

This project is to develop measurement techniques for monitoring and identifying the parameter changes of high frequency (MHz) wireless power transfer systems. In particular, two methods will need to be developed based on hardware and software implementations, and they will be compared thoroughly based on the accuracy, time delay, signal to noise ratio (SNR), etc.

The project involves design of sensing and filtering circuits, equivalent software implementation, practical testing, and finally performance evaluation for comparison.

It would be ideal that perspective students have combined hardware and software capability. For example, one student from EEE, and another from CSC or SE. Successful candidates will work closely with postgraduate researchers in the research group.

Location: Power Electronics (405.614, Lab)

Supercapacitors are essential for energy storage applications, yet their internal dynamics require deeper investigation for improved reliability and efficiency. This study will analyze key electrical properties through extensive experimental testing, including long-term cycling, stress testing, and temperature-dependent evaluations. A materials-focused approach will be integrated to examine how electrode composition, electrolyte properties, and degradation mechanisms influence performance. Equivalent circuit models will be validated, and numerical methods will be used to estimate state of charge (SOC) and state of health (SOH). Additionally, the study will explore balancing techniques, reverse polarity charging effects, and ultra-low-frequency behavior to enhance practical deployment.

Key Electrical Properties Under Investigation

• Self-discharge characteristics
• Leakage current behavior
• Internal resistance variations over time and temperature
• Ultra-low-frequency dependence
• SOC and SOH estimation using numerical methods
• Reverse polarity charging effects on SOH
• Equivalent circuit model validation
• Series and parallel balancing performance
• Long-term cycling stability
• Stress testing under extreme conditions
• Influence of electrode and electrolyte materials on performance
• Degradation mechanisms and failure analysis

Location: Green Electronics (405.712, Lab)

STEM is a hot topic among parents and communities where there is a lot of focus on promoting Science, Technology, Engineering, and Mathematics (STEAM) among the future generation. The predecessor of this project has developed a series of activities, which target high school students. These activities are based around computer hardware, allowing students to investigate the backend of computing devices and hardware programming. In this year's project we aim to investigate the effectiveness of these activities by taking them to high schools and collect data. Ethics approval has been granted for this project already.

Location: Green Electronics (405.712, Lab)

Productivity in software development has significantly improved through the use of APIs. However some APIs have been found to be difficult to use, and in particular developers often make critical mistakes when using them, leading to difficult-to-find faults. API usability research studies how developers interact with APIs with a view to helping them use APIs correctly, and to improve APIs so that they are easier to learn and for which developers are less likely to make mistakes.This project is to explore the current state of API Usability research and advance it in some way. There are a number of directions this project could take, including developing tools for assessing the usability of a tool, creating tools that help developers learn and/or use APIs more efficiently, or proposing new variants of APIs that are more usable than the original.

Location: MDLS (405-569)

This project aims to develop a VR-based driving simulation where an empathic virtual agent assists drivers by adapting to their cognitive and emotional states. The system will monitor biosignals (e.g., heart rate, gaze tracking) and collect subjective self-reports to assess cognitive load and emotional states. The virtual agent will provide real-time interventions, such as stress management guidance, adaptive difficulty, and break reminders. The goal is to evaluate the feasibility of this system for cognitive load management, with potential applications in rehabilitation settings.

Part 4 Student’s Role:

• Develop a VR driving environment using Unity and integrate an interactive virtual agent.

• Design an experiment using biosignal recording (PPG, eye tracking, etc.) and subjective self-report questionnaires to measure cognitive load.

• Conduct a user study to evaluate the effectiveness of the system in managing cognitive load

Required Skills:

• Essential: Interest in VR, problem-solving, experimental design, data analysis, and basic programming (python, C#).

• Desired: Familiarity with Unity.

Location: MDLS (405-569)

This research-oriented project focuses on exploring the collection of biometric data during gameplay in virtual reality (VR), specifically through a game like Beat Saber. The primary objective is to estimate and gather various physical dimensions of users, such as height, arm span, waist width, and potentially many other metrics. The technical challenge lies in accurately estimating these dimensions within a VR environment without direct physical measurement.

The process involves creating a system where all data is gathered by a third-party monitoring tool, which will only collect information available through the game, such as headset position and movement data. This tool will not access any private or personal data beyond what is required for gameplay. After data collection, we will run a data pipeline to estimate various biometric dimensions, such as height, arm span, and waist width. These estimations will be cross-verified by offering users a survey at the end of the session, where they can confirm the accuracy of the data points we estimated. This ensures the reliability of our methods while providing transparency to participants about how their data is being used. We may be working on a data pipeline using tools like Microsoft Fabric to generate reports in Power BI and use machine learning models to improve the accuracy of our estimations.   A key focus of this project is to address the growing concerns regarding privacy within public VR spaces. Third parties may exploit biometric data to profile users without their consent. By mimicking this behavior, our research aims to investigate the potential risks of biometric data theft and raise awareness about how easily such information can be captured in virtual environments. We aim to conduct this study transparently, providing participants with insight into the data we are collecting and the methods behind our estimations, in order to contribute to a broader conversation about privacy and security in VR spaces.

Location: MDLS (405-559)

Exergaming, or exercise gaming, combines physical activity with interactive video games, offering numerous benefits. It can make exercise more enjoyable and engaging, especially for those who find traditional workouts monotonous. This approach can improve cardiovascular health, enhance coordination and balance, and increase overall physical fitness and even emotional and mental health.

VR exergaming has the added benefit of immersion, which can make game play even more engaging and provide dissociative benefits, e.g., for people with poor body image and a high level of self-consciousness. In this project we will work with exerbike-based games, which can elicit high-intensity workouts. To make such games exciting, it is preferable to allow users freedom of how to traverse a virtual world. Unfortunately, direction changes, especially at high speed, can cause simulation sickness (also called cybersickness) due to cue conflicts (the brains gets conflicting signals from the visual system and inertia system).

In this project we want to investigate strategies to reduce simulation sickness and discomfort in exerbike-based VR exergames with a high degree of freedom. You will develop a simple exergame implementing different strategies to display the virtual environment during gaming (e.g., reduced field of view, heads-up display, blurring or contrast reduction during direction changes, different backgrounds etc.). You will perform a user study evaluating these strategies and their correlation with optical flow of the displayed scene.

Location: MDLS (405-569)

Globally, metropolitan cities are rushing towards Net Zero Emissions by 2030. Use of active mode is becoming easier with new micro-mobility modes such as scooters and e-bikes. Despite these changes, a large proportion of the population in car-centric cities do not cycle. Changes in infrastructure are proving to be not enough to change safety perception and the ease of cycling.

In this research we will develop a VR environment to investigate cyclist’s safety perception of road and cycleway infrastructure. In a previous project we investigated the effect of bicycle lane design and environmental conditions. In this project we will investigate the effect of soundscapes whether certain cycleway designs are more suitable than others for perceived safety. Example of possible auditory parameters to investigate are different types of noises (car, truck), amplitude, changes in amplitude, sudden sounds (misfiring cylinder), different background sounds etc.

Location: MDLS (405-559)

The US Department of Energy (DOE) recently unveiled a groundbreaking initiative, the "Pathway to commercial liftoff of Virtual Power Plants," highlighting the transformative potential of VPP deployment (https://liftoff.energy.gov/vpp/). By 2030, deploying 80-160 GW of VPPs could fortify the US grid's capacity to support rapid electrification while redirecting grid spending from peaker plants, leading to significant cost reductions.  Similar efforts are ongoing in Europe and other industrialized economies globally.

Some key research and implementation questions that arise are as under:

1. What services can VPPs provide, and what is their future role in the energy landscape?
2. What should be the cost-benefit analysis of VPP programs to justify their implementation?
3. What is the role of distribution utilities in integrating VPPs and optimizing grid operations?      
4. How can utilities effectively integrate VPPs into grid and resource planning strategies

Recently there has been development from IEEE on Grid-supporting and Grid-forming Microgrid architectures in IEEE publications.  The implementation is primarily based on inverter-interfaced PV, storage and other micro-generation resources.  There is a need to research into developing interconnection guidelines if these microgrid architecture become operational in NZ in the coming years which is the focus of this project around developing Virtual Power plants formulation.

In the last couple of years we have supported a start-up SolarZero towards developing solutions towards grid-supporting(Link).This project won theNZ Energy Innovation Award in 2023

Power Systems Group @UoA also organized aninternational eventwhere in one of the panels the development, standardization and uptake was discussed.

References

1.Measuring the solar potential of a city and its implications for energy policy

2.Smart grid research in New Zealand – A review from the GREEN Grid research programme

3.Vector’s residential photovoltaic trials : a report on experience

Location: Power Systems (405.628, Lab)

More than half of aged care residents are not eating enough and are missing out on essential vitamins and other nutrients despite being offered a variety of food. There are several indicators used to measure nutritional status. These include body composition, clinical signs of deficiency, physical function, biochemical compounds, metabolic processes or dietary intake.

In this project we will work with an expert in Nutrition & Dietetics at The University of Auckland to build a machine learning model that identifies the risk of undernutrition in older people.

Location: Control & Signal Processing (405.722, Lab)

Modern software systems increasingly rely on external libraries and other software components to provide reusable functionality. As these libraries evolve and release new versions containing enhancements, bug fixes, and security patches, dependent software systems must update their dependencies to stay secure and functional. However, updating dependencies can introduce software failures, particularly when new versions contain breaking changes. Recent research has identified transitive dependencies—dependencies of a system's direct dependencies—as a leading cause of breaking changes. Long and complex chains of transitive dependencies often lead to version conflicts, incompatibilities, and unforeseen failures.

This project will conduct an in-depth investigation into transitive dependencies, aiming to better understand their impact on software maintenance and to develop strategies for mitigating the associated risks. The project will address key questions, such as:

• How frequently do multiple dependencies in the chain declare dependencies on the same underlying libraries?
• What versions of shared dependencies are used, and how often do conflicts arise?
• How can potential version conflicts and transitive dependency issues be more effectively communicated to developers?
• How can developers be supported in making informed decisions about selecting dependency versions?

Location: HASEL (405.662, Lab)

This project investigates how software developers collaborate in code reviews by analysing conversations from public open-source Github projects. The project will explore what makes technical discussions effective and constructive by combining existing organisational psychology methods with natural language processing (NLP) techniques. The project will apply models and techniques developed by Multitudes, a New Zealand software startup, to open-source discussions to understand human dynamics and knowledge-sharing that occur during code review. The project will also leverage established ethical ML frameworks from academic literature to report more nuanced evaluation metrics of model performance.

Location: HASEL (405.662, Lab)

Lake Waikare in north Waikato, is heavily polluted. The water pollution is caused by the runoff from extensive farming in the lake’s catchment area. Efforts are now being undertaken to improve the quality of the lake, and it is important to have technologies to continuously monitor the water quality. Currently, sparse static water quality sensors are deployed in and around the lake. In this project, we aim to develop a mobile buoy which can carry a set of water quality sensors and allow monitoring the lake’s water quality at multiple points around the lake. The craft will need to be autonomous and capable of reporting data continuously to a cloud-based backend.

The project will explore the various technologies to achieve remote control, autonomous navigation, sensing, and communication which the buoy will use; along with its on-board algorithms to cover the entire lake collecting water quality data.

Location: Embedded Systems (405.760, Lab)

New Zealand freshwater resources are heavily polluted. Efforts are now made to try and monitor freshwater resources with water quality sensors at varying geographical locations. The sensor data coming from the variety of sensors needs to be presented in an easy-to-understand visual representation to capture the spatial and temporal information embedded in the data.

In this project, the team will explore the state-of-the art visualisation tools and techniques with a focus on 3D representation of the geographic elements overlayed with the multi-modal time series data captured from the sensors. This may include (but not limited to) existing weather station data, soil data, and water quality data.

Location: Embedded Systems (405.760, Lab)

There is an immediate need to develop technologies that can precisely capture the emergence and trajectory of peoples’ resilience in high-risk industries. Despite government efforts, high risk industries such as construction continue to lead work-related harm and fatalities. The integration of psychophysiological monitoring with virtual reality (VR) makes it possible to document an individual’s experiences as they interact in simulations of real-world contexts. However, conventional psychophysiological-VR systems are constrained to measuring limited features of behaviour in fixed VR environments. This project will develop a VR simulation of the standard site safe training that can be experienced and engaged with by professionals working in the construction industry. Psychophysiological measures will be captured from the professional within the VR environment to help identify a person’s risk resilience and potentially adapt the training to their current status.

Location: MDLS (405-569)

Legal documents are primarily distributed as PDF documents, making it complex for machines to access the content. To automatically check for compliance with legal requirements, a common approach is translating the legal documents into a logical representation, which allows execution by an automated theorem prover. While textual content can be successfully translated into such a logical representation, there is limited research investigating how to translate graphical content (e.g., figures, graphs, etc) into a logical representation.

The objective is to utilise modern deep learning models for natural language processing and computer vision to parse the graphical content and translate it into logic rules. One option could be using image-language models to generate descriptions of the graphical content. Then, the existing deep learning model or training data for translating textual content can be reused or modified.

This project lets you experiment with cutting-edge deep learning models, utilising them for a domain application.

Location: Radio Systems (405.736, Lab)

The documents submitted to a council for consenting comprise many hundreds of pages, including building plans and written information about the building and its materials. To automatically process previously submitted document sets it is necessary to separate out the plans from the other written material. These plans represent different aspects of a building’s design and once the plans are identified it is necessary to classify what aspect of the building design is represented in each plan (e.g., ventilation design, structural design, etc).

The objective is to utilise modern deep learning models for natural language processing and computer vision to parse the document sets, to classify plans versus non-plans, and then sub-classify the plans according to the building aspect they represent.

This project lets you experiment with cutting-edge deep learning models, utilising them for a domain application.

Location: Radio Systems (405.736, Lab)

On construction sites there will be a rubbish skip that various types of construction waste is thrown into. This could include Gib sheets, framing timber, wrapping plastic, plastic tube, steel rods, bricks, etc. When the rubbish bin is collected the waste management company would be best able to recycle material if they knew what was in each bin and how much of each material was available. This project looks to identify a range of materials and to calculate the proportion of them in each bin.

A labelled set of images from waste bins is available from the company Kauricone who are looking at ways the classification can be improved.

Location: Radio Systems (405.736, Lab)

With the advent of cloud storage and computing, data generated and/or belonging to individuals or communities is being stored and processed in overseas territories, where the laws around data security, privacy, etc are at odds to the home country of the data owners.

Because of these inconsistencies new methods and systems are being developed. The European GDPR is a well-known example of this, which protects the data belonging to European residents. In New Zealand, along with the Privacy Act, new initiates are being undertaken to protect the data generated by and belonging to the Māori, under the umbrella of “Māori Data Sovereignty”.

The project aims to create a micro-datacentre, which uses green energy (solar + wind), which can be readily deployed in Māori communities to allow them to manage their locally generated data to align with “Māori Data Sovereignty”. The team will also research the various standards and best practices for data sovereignty in the context of New Zealand Privacy Act and Indigenous (Māori) data sovereignty, in order to identify the necessary technical requirements. The team will design and implement suitable hardware and software systems to ensure that the micro-datacentre is easily applicable and data storage is secure and reliable.

The project’s scope will be limited to time series IoT data, to act as a proof-of-concept for the idea. IoT devices using low power networks like Sigfox and LoRaWAN, have their data routed through the network operators’ networks, which causes data to be relayed and possibly stored in data stores which are outside the control of the data owners. In order to ensure security, the project will also explore into techniques that can ensure end-to-end security of data generated on IoT devices with limited compute power and tight power consumption constrains.

Location: Embedded Systems (405.760, Lab)

Setting up the digital platforms, including the software architecture and databases is a manualprocess, which is both time consuming and error prone. Also, ensuring that the ethics approvedprotocol is adhered to has to be ensured using testing. Can this be mitigated?We take our inspiration from two different fields. First, using advances in natural language processing (NLP),especially through large language models (LLMs), we seek to convert the description of the clinical trialprotocol into a machine readable format. During this step, we will try to group clinical trials into specific categories, such as randomised controlled trials (RCTs) and observational studies etc.During the next phase, we will explore automation of the software architecture using the synchronousapproach. Specifically, we will explore the recently developed HipHop.js language, which capturesthe reactive aspects, including concurrency, preemption effectively. This approach, will thus enable effective automation of clinical trials. We will explore a few clinical trials such as the "chemotherapy related cognitive decline" trial at Tata Memorial Hospital.

Location: Embedded Systems (405.760, Lab)

With age, elderly prefer to move to old-age-homes, where they can get a round-the-clock care and support. Although this helps improving the needed healthcare attentions for the elderly, it’s not easy or possible to take care of all the old-age home occupants at all times. This is where modern technology can help fill in the gaps. Although smart watches are already prevalent and can easily address this need, they need to be charged frequently and are mostly the size of a large wristwatch. These two limitations make them inconvenient for old age people with mobility or dexterity issues as well those who are not very tech savvy.

In this project, we are aiming to develop a continuous health monitoring system, consisting of long-range RFID receivers and wrist worn bracelets. The bracelet will be a simple semi-passive RFID tag with the ability to do simple sensing. The project team will also explore harvesting RF energy to supply the bracelet’s sensing needs. Lastly, the project will explore the possibility of locating the occupants around the old-age facility, using RF tri-literation techniques.

Location: Embedded Systems (405.760, Lab)

NOTE: This is a student-led pre allocated project and the students have written the following abstract and submitted to me.

Autonomous driving systems require precise real-time decision-making to ensure safety and reliability in dynamic environments. Synchronous languages are widely utilized in safety-critical domains such as avionics but remain largely unexplored in the context of road vehicles, presenting an opportunity for further research in autonomous driving systems. This project aims to develop an embedded system that demonstrates safe autonomous navigation using a synchronous programming approach. The system will utilize a suite of sensors for obstacle detection and environment mapping and to enhance spatial awareness. The autonomous vehicle will be tested within an emulated cityscape, designed to replicate real-world driving conditions with various road layouts and obstacles. By leveraging synchronous languages, the project will explore deterministic execution for improved real-time responsiveness and fault tolerance. The outcome will be a prototype that validates the effectiveness of synchronous embedded systems in achieving safe and predictable autonomous driving behavior within structured urban environments

Location: Embedded Systems (405.760, Lab)

Telescopes observe celestial objects (or sources) in the sky. The telescopes are usually movable and can point at different positions in the sky. Very large scientific telescopes in astronomical observatories are highly specialised, very expensive and scarce. Hence, they must be used as efficiently as possible.In a typical operation mode, a telescope will be configured to observe a given list of sources, which correspond to positions in the sky. It remains pointed at a source of the list for a certain observation period, e.g. 10 minutes, after which the telescope is rotated and tilted to point to the next object on the list. Due to the fact that the movement between positions takes time and that necessary observation periods might vary, this observation of sources becomes a difficult optimisation problem.In this project you will develop scheduling and ordering algorithms to optimise the use of telescopes. This includes the development or improvement of existing tools that can process lists of source to be observed which then creates optimised observation sequences that can be used for the operation of telescopes.

Location: MDLS (405-559)

As autonomous vehicles become mainstream, their interactions in an uncertain environment, especially with human driven vehicles, needs careful consideration. An obvious but challenging question is related to ensuring the safety of autonomous vehicles, especially in uncertain environments. While this is a difficult and challenging question to address in this project, we propose a disruptive solution by leveraging a recently developed networking protocol called bittide, which is deterministic. This is unlike many other types of networks used in modern vehicles, which often have unbounded communication delays, that make the verification problem potentially intractable. Moreover, the networking diversity also makes the security analysis problem equally hard.

We propose a novel solution that leverages Google bittide’s logical synchrony and the correct-by-construction design paradigm, to show for the first time that deterministic distributed applications in the automotive domain can be developed in a seamless manner. In this project, we will work on an AV emulation platform using FPGAs and the bittide network for communication.

Location: Embedded Systems (405.760, Lab)

Models like word2vec and node2vec represent real-world objects (e.g., documents, graphs, and images) in high-dimensional vector spaces, where semantically similar objects are mapped to vectors close to each other in a metric space. Approximate Nearest Neighbor Search (ANNS) identifies data vectors that are close to a given query vector, playing a crucial role in similarity search and powering applications such as Retrieval-Augmented Generation (RAG) and more. Graph-based methods like HNSW and its variations such as diskann,freshdiskann have gained an increasing attention for their superior empirical performance. Our recent work, SeRF, tackles the Range-Filtered ANNS problem using graph-based methods. In this setting, each data vector is associated with an attribute value (e.g., timestamps or prices), and queries search for nearest neighbors constrained by a specified attribute range. To the best of our knowledge, SeRF is the first peer-reviewed solution for this problem, achieving an \Omega(n) reduction in memory footprint without compromising search accuracy compared to vanilla approaches. Additionally, we have two works under submission that investigate ANNS for hybrid queries incorporating structural constraints. Despite its advancements, graph-based ANNS remains in its early stages compared to the mature relational database toolset for query processing. This highlights research opportunities to further enhance the scalability, efficiency, and functionality of graph-based ANNS solutions. It is a pressing research direction with massive research questions in both theory and practice in ANNS given the high demand of AI-based data-intensive applications in coming years.

The first project would be in practice. A comprehensive suite of techniques has yet to be developed to systematically and efficiently support scalable ANNS query processing with potential point or range filters across a memory hierarchy. Key research directions include i) vector management in external memory, i.e., to developing I/O-efficient techniques for querying and updating ANNS data stored in external memory, ii) multi-model search indexes and iii) real-time, adaptive ANNS systems, i.e., process dynamic, streaming data, that learn and optimize query performance on-the-fly, adapting to evolving datasets and changing user requirements.

Location: Radio Systems (405.736, Lab)

Models like word2vec and node2vec represent real-world objects (e.g., documents, graphs, and images) in high-dimensional vector spaces, where semantically similar objects are mapped to vectors close to each other in a metric space. Approximate Nearest Neighbor Search (ANNS) identifies data vectors that are close to a given query vector, playing a crucial role in similarity search and powering applications such as Retrieval-Augmented Generation (RAG) and more. Graph-based methods like HNSW~\cite{hnsw} and its variations~\cite{diskann,freshdiskann} have gained an increasing attention for their superior empirical performance. Our recent work, SeRF~\cite{ZuoQZLD24}, tackles the Range-Filtered ANNS problem using graph-based methods. In this setting, each data vector is associated with an attribute value (e.g., timestamps or prices), and queries search for nearest neighbors constrained by a specified attribute range. To the best of our knowledge, SeRF is the first peer-reviewed solution for this problem, achieving an $\Omega(n)$ reduction in memory footprint without compromising search accuracy compared to vanilla approaches. Additionally, we have two works under submission that investigate ANNS for hybrid queries incorporating structural constraints. Despite its advancements, graph-based ANNS remains in its early stages compared to the mature relational database toolset for query processing. This highlights research opportunities to further enhance the scalability, efficiency, and functionality of graph-based ANNS solutions. It is a pressing research direction with massive research questions in both theory and practice in ANNS given the high demand of AI-based data-intensive applications in coming years.

Theoretical understanding of why HNSW performs so well remains limited, leaving its adaptation to memory hierarchies shrouded in uncertainty. While some progress has been made by Indyk and Xu, their work represents only an initial step. This gap presents significant opportunities for further exploration, particularly in developing foundational insights that can guide the efficient adaptation of graph-based ANNS methods to modern memory architectures.

Location: MDLS (405-569)

In real-world software development, requirements are often conveyed in natural language, making ambiguity and misinterpretation common. Identifying gaps and clarifying missing details is a critical skill for developers, yet introductory programming courses rarely emphasize this practice. Instead, students typically work with well-defined problem statements designed for automated grading, which provide immediate feedback but limit opportunities to practice clarifying specifications.

While automated grading can scale well and is widely used in programming courses, the arrival of large language models (LLMs) has presented a significant challenge. Typical introductory-level programming tasks can now be reliably solved by LLMs. This means that students can simply provide a programming problem statement—exactly as defined by their instructor—to an LLM as a prompt to generate working code. This can reduce student engagement with programming tasks, and may lead students to question the value of writing code for well-specified tasks.

One way to address this issue is through programming problems with deliberately vague or incomplete specifications—problems that require students to ask clarifying questions before arriving at a solution. This project aims to define the essential characteristics of these ‘probeable problems’ and develop a novel platform to host and deploy them. The platform will enable students to engage with problems that resist trivial solutions from LLMs while fostering critical thinking.  Additionally, the system will collect fine-grained interaction data to analyze how students navigate ambiguity and refine problem specifications.

Location: Radio Systems (405.736, Lab)

Background

The rapid integration of large-scale solar farms into New Zealand’s power grid has introduced new challenges, particularly in managing harmonic distortions. Research indicates that the most efficient method for addressing harmonic distortion is not always at the source (i.e., the solar farms). Moreover, installing multiple harmonic filters across various sites can complicate harmonic management, creating additional challenges for grid operators.

To address these issues and nurture future engineering talent, Lodestone proposes engaging two Year 4 students from the University of Auckland to collaborate on a harmonic study. The study will focus on solar farm integration and explore harmonic solutions from a network-wide perspective.

The outcomes of this study will provide valuable insights for both electricity distribution businesses (EDBs) and developers, enabling them to better manage harmonic risks during the development and planning stages of solar farm projects.

Objectives

The primary objectives of this project are:

· To analyse harmonic impacts associated with large-scale solar farms connected to the New Zealand power grid.

·  To develop cost effective mitigation strategies for harmonic distortions.

·       To review current market rules, including the existing liability structure and recommended changes in liability allocation between generators and networks as a general public good.

Scope of Work

The students will work on the following tasks:

·       Data Analysis: Collect and analyse harmonic data from existing solar farm projects and the EDB’s network.

·       Harmonic Modelling: Use software tools (e.g., PowerFactory) to simulate harmonic distortions under various grid scenarios.

·       Mitigation Strategies: Explore and evaluate measures to reduce harmonic distortions, such as filtering techniques or grid code adjustments (e.g., harmonic allocation method).

·       Policy Change: Review the existing liability structure and recommended changes in liability allocation between generators and networks.

·       Reporting: Prepare a comprehensive report summarising findings and recommendations.

Benefits

For the Students:

·       Hands-on experience with real-world power grid challenges.

·       Mentorship from industry professionals.

·       Valuable exposure to renewable energy integration.

For Lodestone and the EDBs:

·       Fresh perspectives and innovative approaches to harmonic studies.

·       Contribution to the development of future engineering talent.

Location: Power Systems (405.628, Lab)

Background and Context:

Coinciding with the release of ChatGPT followed by other Language Models in 2023-24,a 6-year long working group of CIGRE-B5released its final brochureTB 926which presents the main treats of an informal or near-formal definition of a simplified Domain-Specific Language (DSL), informally called PacDSL, intended for the formal specification of functional requirements of Protection, Automation and Control of electrical power systems.

The definition covers the basic aspects of the lexicon, syntax, semantics, and pragmatics of a typical DSL for this objective, that is agnostic to technology and proprietary solutions, aiming the needs of utilities, academics, manufacturers, and consultants. The brochure intends to document the need for this formalization, based on a survey among utility stakeholders, and to exemplify some possible constructs and Use Cases for this proposed language. It mainly shows its insertion in the engineering cycle during the requirement specification to be deployed on IEC 61850 and other implementation standards.

Power systems Group @ UoA have been pioneers in the development, use and implementation of IEC 61850 from 2004 onwards.   This is now well established technology within Digital Substations globally with several NZ distribution utilities like Vector, Orion and Unison having had established various aspects of this standards into their networks.    In 2023, Transpower has also embarked on all its future newer substations developed and installed through Digital Substations.   This has necessitated knowledge and practice of IEC 61850 across all of New Zealand/Australia based power system consultancies to have working knowledge of this emergent technology

UoA Power Systems Group (PSG) have carried out several IEC 61850 based projects with New Zealand distribution  utilities in the past and these entities will be supporting the development of the prototype solutions. In particular Transpower and Vector protection engineer's will be providing their inputs along with active support from PSG PhD's and postdoc's actively researching in Smart Grid projects. (e.g.  Automatic Bus Transfer scheme). Two-decades of IEC 61850 has been captured in thisCIGRE GREEN Book.This e-book is available from UoA library and Chapter 8 identifies how functional schemes could be developed for various schemes.

The proposed project will prototype PacDSL LLM's for protective relaying using IEDs and IEC 61850 commercial testing facilities available in the power systems lab.

Location: Power Systems (405.628, Lab)

Resilience and, in more specific terms, robustness is more relevant now than ever, with the increase of extreme weather events in frequency and intensity due to the effects of climate change. This is especially relevant here in New Zealand, where we are very susceptible to extreme weather events, for example, Cyclone Gabrielle.It is important for the grid to be robust, a part of resilience which means the grid's ability to resist these extreme weather events. For example, the least amount of customers on the grid possibly losing power during a severe weather event (high-wind, flooding, cyclone etc.)

This project is associated with developing efficient tools using AI and LLM for esilience impact assessment methods for regional electricity networks, using high-resolution high-impact weather scenarios, to help develop post impact recovery management and through some known case studies explore if any additional extra pre-preparatory information could have been extracted for future use.   A 2023 IEEE PES report that the supervisor was involved in might provide someuseful AI/ML practical applications in Power System Protection and control background referenceto this project.

One major source of electricity distribution network outage is due to wind-gust . As part of an ongoing research program and PhD projects, we have access to maximum wind gust analysis for Auckland, Waikato and Northland region across New Zealand on a granular scale.   There is a need to focus attention on one geography like Auckland and develop co-metrics along with local distribution network company to find out the correlation of extended outages caused due to extreme wind gus scenarios of the past and then through some advanced learning and estimation techniquest like machine-learning, predict some potential outages based on future wind-data gust models.

Some work has already been done, which will be good starting reference for this project

(1)Resilience based Criticality Analysis for Seismic Performance Assessment of Underground Cables(2)Resilience Framework and Optimal Scheduling for DERs Factoring Uncertainties

(3)Extreme Weather Risk Framework for Power System Location-based Resilience Assessment

(Won Best Paper Award)

The project team can also consider looking at other impact weather hazards like extreme flooding due to "atmospheric river" and accordingly arrange the framework.

Another potential case-study worth looking into could be forest fire. One of our recent interesting paper around forest fires and impact on solar generation is available from

Solar Power Generation in Riau-Indonesia, Potential and its Challenge Due to Forest Fire

The project team will analyse Hazard Data like wind and precipitation, potentially from government entries like Metservice, Niwa and Geonet. And compare it to outage data which we can get from Transpower for transmission outage data and the local distribution companies like Vector, PowerCo, Unison etc for more localised outage information. We will perform a Geospatial analysis on this data, and with the help of machine learning, we can determine which areas were the most prone to these outages and determine if the increased presence of renewable energy allowed certain regions to fare better during extreme weather events.

Location: Power Systems (405.628, Lab)

Digital substations represent a significant advancement in the modernization of power systems infrastructure, leveraging technologies to enhance efficiency, reliability, and monitoring capabilities. Unlike conventional substations architectures that rely on point-to-point wiring, digital substations employ standardized communication protocols to facilitate seamless integration of intelligent electronic devices (IEDs) and substation automation systems. At the forefront of this transition is the International Electrotechnical Commission (IEC) 61850 standard, which provides a comprehensive framework for the design, engineering, and operation of digital substations. IEC 61850 defines a common language and data model for communication between IEDs, control systems, and utility SCADA (Supervisory Control and Data Acquisition) networks, promoting interoperability, flexibility, and scalability across diverse substation architectures. The adoption of IEC 61850-compliant solutions, is a key strategy for utilities to unlock the full potential of smart grid technologies.

However, as the threat landscape in the digital age evolves, safeguarding critical infrastructure such as electric utilities become paramount. To defend against growing cybersecurity threats, many electric utilities are beginning to integrate intrusion detection systems (IDS) to monitor electrical substation automation communications networks. These systems examine Ethernet network traffic to detect unusual or malicious behaviour that could indicate a cyber-attack and thus, must integrate with systems that employ IEC 61850.

IDS solutions employ two primary techniques for cyber-attack detection: signature-based and anomaly-based detection. Signature-based detection compares network traffic against predefined pattern rules indicative of malicious activity. In contrast, anomaly detection learns normal network behaviour over time and triggers alerts when deviations from this baseline occur.

While anomaly detection holds promise due to its adaptability, concerns have been raised regarding its suitability for digital substations (Cybersecurity in Substations – Attack Vectors on Substations and their Countermeasures - OMICRON (omicronenergy.com). Unlike traditional IT environments, digital substations exhibit unique communication patterns, characterized by long periods of stability interspersed with bursts of activity during primary electric events such as switching or power system faults. This dynamic behaviour may inadvertently trigger false alarms in anomaly-based IDS, potentially overwhelming cybersecurity analysts with unnecessary alerts.

This project was undertaken for the first time during 2024 and focus was on IDS.  Through the 2025 project, we would like to expand it to testing it on actual New Zealand substations and developing simulation-emulation in the power system laboratory @UoA.  Students will be working closely with international IED vendors (NZ suppliers).   We have established a new collaboration with Doble Engineering solutions who have recently introduced IEC 61850 testing kit.  This project will explore incorporating that (particularly timing and cybersecurity stack) within our existing test environment that has Omicron, OPAL-RT and the newly installed RTDS.   Students will also need to explore new toolchain kits utlizing TYPHOON and integrating simulation tools like PLECS, PSCAD, DigSilent

Location: Power Systems (405.628, Lab)

Web technologies are foundational and continue to be widespread, with front-end development skills in high demand. This project pursues the automatic assessment of web aspects dynamically, by executing them as would occur within a typical browser using the Selenium WebDriver framework. In this project you will  write custom code to assess and interact with web components through browser-specific drivers, expanding functionality to assess visual Gestalt Principles and interactivity in programmatic fashion. Selenium enables the remote control of a browser and mimics user actions on the browser including button click, drag, and drop selection, checkboxes, key presses, taps, and scrolling. The use of this tool is educational to support increased understanding of accessibility guidelines and visual design skills.

Location: MDLS (405-569)

Over the course of his internship, Albert Sun—our Part IV Software Engineering student—has made impressive progress on our open-source IoT platform. Building on this work, we now aim to develop a new demonstration platform that showcases our IPT system’s power capabilities while integrating advanced IoT control, sensor inputs, and data management functionalities. The goal is to create a live, interactive model that demonstrates how the IPT system operates in real-world scenarios.

Location: Green Electronics (405.712, Lab)

Monitoring the health of beehives is important for several reasons. Although there are several factors affecting hive health, this project will focus on detecting motion of bees. Whilst it is intuitive to use imaging methods, this is only viable in hives designed for cameras. This is not viable for widespread deployment.

The existing method uses electrostatic sensors because bees acquire a charge after a flight. Our understanding is that the data from existing sensors are quite difficult to use and this warrants a closer examination.

This project will build/redeploy the existing design to gain a better understanding and then consider any modifications. The project will require both digital, analogue hardware and embedded software skills.

Location: Control & Signal Processing (405.722, Lab)

This is cross departmental project.

This project aims to develop and qualify a robust power system element as part of Te Pūnaha Ātea – Space Institute's (TPA-SI) ongoing efforts to create a highly integrated, in-house avionics suite for Cubesat missions. This suite, comprising power, communications, and on-board data handling (OBDH), will enable lower-cost and faster delivery of future TPA-SI missions. A simplified version of the suite will also be applicable to the Auckland Programme for Space Systems (APSS), promoting design and system alignment between the two programs.

This project will build upon existing designs for avionics derived from the PyCubed open-source project and ongoing student work to create a fit-for-purpose power system. The system will encompass solar panel power harvesting, battery management, and power distribution, all integrated within a compact form factor suitable for integration into the overall avionics suite.

Working in collaboration with other students and staff researchers, the student will develop and test development models of the power element, both in isolation and as part of the integrated Cubesat system. A key aspect of the project is to develop and perform appropriate qualification tests for all key components, with a particular focus on the batteries to ensure the system can meet requirements for launch and operation.

Project Scope and Objectives:

·       Literature Review: Conduct a review of relevant literature on Cubesat power system options and existing designs, including open-source projects and best practices.

·       Requirements Capture: Capture, review, and flow down power system requirements for both TPA-SI and APSS Cubesat platforms, ensuring alignment with mission objectives.

·       Battery Build, Test and Qualification: Build a battery system that meets design and qualification requirements. Perform environmental testing to validate the system's robustness in space-like conditions.

·       Circuit Design and Build: Design and build the necessary power conditioning, regulation, and distribution circuitry, ensuring efficient power conversion and protection.

·       Power Element Testing: Conduct thorough testing of the power system element, including performance characterization, efficiency measurements, and fault tolerance testing.

·       Integrated System Contribution: Contribute to the testing and integration of the power system within the overall Cubesat avionics suite.

Location: Control & Signal Processing (405.722, Lab)

Monitoring the health of beehives is important for several reasons. Although there are several factors affecting hive health, this project will focus on detecting of mainly CO2 levels and perhaps others like humidity, temperature levels too. Sensors are commercially available and these need to be evaluated with rigour.

A custom sensor will also be also be built and compared with the commercially available ones. An important aspect is building a simple system that permits the creation of test rig for predicable CO2 levels and this will also require calibration using CO2 measurement laboratory grade instruments.

The project will require both digital, analogue hardware and embedded software skills.

Location: Control & Signal Processing (405.722, Lab)

We have developed an alternative model (resonant model, RM) for bio-physical cells that are suitable for integer arithmetic and hence suitable for FPGAs. The each cell is built using several digital resonators and these are highly amenable for parallelization. We would like to create (simulate and implement) a design that achieves this.

At present we can fit about 30 cells (without prallelization) on a Cyclone V. If successful we should be able to fit 5x to 10x cells. The project will involve understanding the RM and then implement a cell/tissue using VHDL (ModelSim). The project will develop a design for parallelization and implement these on an FPGA.

The students will be expected to be skilled in VHDL and Matlab. Please meet with the supervisor before bidding for this project.

Location: Control & Signal Processing (405.722, Lab)

Hard real-time, safety critical programs (e.g., pacemakers, nuclear controllers, fly-by-wire aircraft systems, automative control, etc) are an integral part of deployed embedded systems. Usually these systems are developed in the C programming language adhering to the MISRA or DO-178B (aircraft controller) standards. However, C is known to be full of so called foot-guns, e.g., out of bounds errors, memory allocation errors, null pointer dereference, dangling pointers, double free, etc. Hence, developing hard real-time safety critical programs is usually difficult in C.

In this project we want to develop a hard-real time operating system in the Rust programming language. Rust brings with it certain memory access safety guarantees. We want to develop a hardware abstraction layer and associated real-time operating systems (at a minimum a real-time thread scheduler) in Rust to exploit hard-real time architecture: FLEXPRET (https://github.com/pretis/flexpret) developed at UC Berkeley.

Location: Embedded Systems (405.760, Lab)

Did you know that men over 85 have the highest suicide rates in New Zealand? Suicide among older adults has a devastating impact on families and communities, making early intervention critical. This project aims to develop an AI-driven adaptive learning platform that equips frontline workers with the knowledge and skills to identify and support older adults at risk of suicide.We are looking for students to contribute to the development of this interactive e-learning platform, which will be used by healthcare professionals, support workers, volunteers, and aged care staff. The tool will integrate Large Language Models (LLMs) and AI-driven adaptive learning technologies to provide real-world crisis scenarios, dynamic feedback, and personalized learning pathways.Why Join This Project?This is an excellent opportunity to gain hands-on experience in:•       E-learning development using interactive and AI-enhanced adaptive learning technologies•       User experience and data-driven design for public health education•       Software development with a real-world impact in mental health and aged care•       AI-driven crisis response training that integrates cutting-age LLM technologiesYou will be supervised by both a computer science academic and a medical expert in suicide prevention, ensuring guidance in both technical development and mental health content. By contributing to this project, you will help create a transformative AI-powered training tool that empowers crisis responders and helps save lives.

Location: Radio Systems (405.736, Lab)

Most sensors fall into either resistive or capacitive sensors for which several established interfaces are readily available. For resistive sensors, the Wheatstone bridge (and variants) are the state of the art since the late 1800s! While for capacitive sensors, several microcontrollers offer easy to use building blocks integrated into their silicon. Recently, a direct interface circuit (DIC) has been reported as a good alternative for lossy capacitive sensors.

This project will validate the claims in the literature and it will also evaluate alternative analogue or digital techniques and hence report on the merits and weaknesses of the various methods.

This project will require skills in building electronic circuits, instrumentation and writing software in Matlab/Python for signal analysis and instrument control.

Location: Control & Signal Processing (405.722, Lab)

The project addresses discrete manufacturing system control that can immensely improve by using artificial intelligence (AI) and machine learning (ML) models in real-time context. The AI models such as DeepSeek and other competing models can make realistic inroads into manufacturing realm. The project is explorative and aims at making systematic survey of major AI models used in manufacturing context and classifying types of applications in which they have been used in research and/or commercial domain. We focus on applicability of these models on edge network level where there are constrained hardware resources associated with sensors (e.g. cameras) and machines used in manufacturing and transportation of workpieces. Some experiments with DeepSeek or other elected model(s) such as Qwen, and Intel hardware/software platforms and AI software, which specifically target OpenVINO will be developed.

Location: Embedded Systems (405.760, Lab)

Internet of Things (IoT) layer typically comprises of huge number of sensor and actuator wireless end nodes that connect to the rest of Internet through a suitable communication network that enables reliable transfer of data to/from end nodes. In this project we consider a IoT layer that includes Bluetooth (BLE) enabled end nodes and a wireless (Wi-Fi) mesh network as the communication network.

Wi-Fi mesh is resilient and dependable network capable of self-organisation and reconfiguration that can select any mesh node to serve as a gateway to the rest of Internet. The assumption will be that the mesh nodes have permanent power supply, but a study of what would be needed to operate it certain amount of time on batteries will be done within project.

As both Wi-Fi mesh and end nodes are wirelessly connected computers, in real-life applications (such as building automation, industrial automation etc) it will require up-date of software of each individual node in case of release new version of software or failure of any node.  The up-date should be done through OTA mechanism without disturbance of IoT layer operation, which is a challenging task. This will be investigated in the project and a novel OTA mechanism will be experimentally recommended.

As the mesh node a ESP32 Wi-fi and BLE enable nodes will be used, as the end nodes in-house or commercial BLE-enabled nodes will be used.

Location: Embedded Systems (405.760, Lab)

With the increasing use of renewable energy sources like solar and wind, managing power grids efficiently is becoming more challenging. Since these sources are highly reliable on uncontrollable environmental factors— such as solar panels and wind turbines depend on the weather—it’s hard to balance energy supply and demand. Traditional grid systems rely on centralized control, which isn’t great at adapting to sudden changes in energy production and consumption.

This project aims to build a multi-agent reinforcement learning (MARL) system to make smart energy management decisions. In this system, different parts of the grid—like solar farms, wind turbines, battery storage units, hydroelectric dams, and consumers—act as independent agents that learn and improve over time. The idea is to have these agents communicate and collaborate to optimize energy distribution and storage in real time.

One of the biggest challenges in renewable energy is storage. Since power generation isn’t constant, we need a smart way to decide when to store extra energy, when to release it, and how to minimize waste. Our system will train AI agents to handle these decisions efficiently, keeping the grid stable even when energy production fluctuates.

To do this, we’ll use reinforcement learning techniques like Proximal Policy Optimization (PPO) or Deep Q-Networks (DQN). We’ll train our system in a simulated smart grid and compare its performance to traditional grid management methods.

By the end of this project, we hope to develop an adaptive, decentralized energy management system that makes power grids more efficient, reduces energy waste, and helps integrate more renewable energy into the grid. This could be a big step toward making power systems smarter and more sustainable for the future.

Location: Embedded Systems (405.760, Lab)

A high-power ( >11 kW) Inductive Power Transfer (IPT) research laboratory poses a challenging Electromagnetic Interference (EMI) environment due to high-power dc-ac/ac-dc switching converters and large unshielded magnetics; both of which result in a galvanically isolated power transfer system. During IPT laboratory experiments it is critical for the power supplies, power electronics controllers, and oscilloscope measurement equipment to have a robust communication channel with a central computer. These communications are commonly achieved via ethernet cable LAN and USB serial UART, both of which have been proven to be effective for low-Power IPT experiments. However, as the power transfer level of the IPT system under test increases, these wired connections tend to experience latency issues, packet loss, and connection dropouts. This creates a problematic scenario where the main computer can no longer regulate the power transfer process or read data from measurement equipment, and the experiment must be manually shut down and restarted. This project proposes an investigation into appropriate methods of wireless communication that can maintain robust and reliable connections between devices in a high-power IPT testing system. A secondary research objective will be investigating practical methods of reducing EMI emissions from the power electronics within the system.

Location: Power Electronics (405.614, Lab)

Batteries are widely used as energy storages/buffers for driving mobile electronic devices. However, the charging speed of a battery often needs to be strictly limited, and it has safety and environmental concerns. This project is to investigate the feasibility of using supercapacitors as an alternative energy storage solution.  Charging circuits for powering wearable biomedical devices such as tracking tags will be designed. Both wired and wireless charging methods will be explored at different power levels. The project involves power electronics charging circuit design, simulation, practical testing, and comparison between different charging technologies.

Location: Power Electronics (405.614, Lab)