Transforming Technologies (Past Projects)

Humans and other animals have evolved their sense of hearing over millennia to gather information about the world around them. Sound is converted to electrical signals by the ear mechanism after which it is processed in the brain. This process is analogous to digitally recording audio and processing it with a computer. Over the past decade multi track audio recording equipment has become more and more affordable, whilst the capability of computers to process this audio has increased exponentially. This project focuses on developing smart sound capture stations which accurately record audio data with accurate timing information and transmit it to a central location. This data can then be processed to reveal information about the surrounding world. We will be exploring a number of applications of our devices including, air traffic monitoring, locating a sound source in a sound field and detection of an object’s velocity based on Doppler shifts.

Watch video

Group member

• William Meegan

This project extends an existing machine learning system which measures hip implant displacement with significantly enhanced speed and precision than an unassisted radiologist, and introduces a graphical user interface. While the automated system is often proficient, there are edge-cases where it may fail spectacularly. This user interface allows the radiologist to intervene throughout the automated process, and quickly correct errors without significant compromise to the system's accuracy. For example, in the case of dual hip implants, where the orientation-detection sub-system could falter, the radiologist may intervene and manually correct this stage of the process.

Watch video

Group member

• Nathan Crowe

Our Project: Helping Robots Understand and Explore the Internet

Aim: Our project is like teaching a robot to explore the internet, just like when you ask a friend for directions. We want to make it easier for robots to understand what they see on websites and take action, like finding information or things.

Methodology: We use three important things to make this happen. First, we have a smart computer program that can look at web pages and understand what's on them. Second, we use another program that can look at pictures and tell us what's in them. And third, we have a special computer model called ChatGPT that can have conversations and answer questions.

Why it's Important: This helps people who can't see well because they can ask the robot to find things on the internet. It also helps businesses show their products to more people. Our project makes robots smarter, so they can be better helpers to humans!

Watch video

Group member

• Duoqi Zheng

In the quest to deepen our understanding of the universe, physicists are faced with the challenge of gathering precise data well below the subatomic level. Yet understanding this world not only advances science, but drives technology, informs medicine, and inspires innovation. The complex interactions of light can, in theory, be leveraged to measure distances tinier than even the building blocks of matter, but require unique technology to do so.  This project had four objectives: to explore and understand the physics underpinning the project, to design apparatus to explore and test identified theories, to extract the wavelength information from the speckle patterns using innovative techniques, and to centralise the system for a streamlined, repeatable measurement process. The result of this project was to increase the technology readiness level of these temperamental experiments through the integration of smart systems and emerging automation, ensuring robustness and usability.

Watch video

Group member

• Alexander Trowbridge

Aircraft safety is of paramount importance in the aviation industry, and certification processes rely heavily on rigorous, full-scale structural testing to ensure the airworthiness of aircraft. Thermal imaging presents a powerful tool to enhance structural testing procedures. This project investigates using thermal imaging through a technique called Thermoelastic Stress Analysis to interpret non-visible characteristics of structures. This technique facilitates the detection and assessment of hidden structures and damage, which could otherwise remain inaccessible.

To support this investigation, finite element analysis was used to construct thermal-structural coupled models to simulate thermal responses. These responses were interpreted for a stress response and evaluated to develop various detection algorithms. Experimental testing successfully validated the developed methods, demonstrating their effectiveness in identifying hidden features and assessing structural integrity. The developed methods hold great potential for integration into structural certification procedures for aircraft.

Watch video

Group members

• Matthew Pascale
• Christina Vincent

Differential equations have been used for centuries to model many different phenomena around us, including the climate, the spread of diseases, and biological ecosystems. By solving these equations, we can make predictions of the future. However, these models are always subject to unavoidable errors that arises from many different places, including computational limitations, incorrect physics, and observational imprecision. By embracing a probabilistic framework, we extend these models to stochastic differential equations that include this uncertainty as an essential component. This approach leads better predictions, but is challenging as solutions are much more difficult to obtain and numerical simulation is computationally demanding. My project aims to account for this uncertainty in a fast-to-compute manner, by taking advantage of what the system tells us when there isn't any uncertainty. I showcase the prowess of my work by forecasting the dynamic movements of the Gulf Stream, an important part of the North Atlantic Ocean.

Watch video

Group member

• Liam Blake

Imagine you could type a sentence on your computer and make a cartoon character dance! Sounds fun, right? That's what our project is all about. We're working on a special AI model (Temporal Diffusion Model) that can take words and turn them into dance moves for animated characters. But it's not as easy as it sounds. Sometimes the characters move in funny or strange ways that don't look right. So, we're like detectives, figuring out what makes the characters move weirdly and how to fix it. We're doing lots of tests and making changes to get it just right. Our goal is to make a mini-version of this program that can create short, fun dance videos for apps where people like to watch quick clips. In the end, we made a tool that helps anyone create awesome animated dances just by typing!

Watch video

Group member

• Akide Liu

Currently, the Australian farming industry spends $3.3 billion annually on weed control, primarily relying on chemical methods, known as pesticides. In addition to this financial burden, pesticides have been linked to chronic illnesses such as Alzheimer's, Parkinson's and Diabetes. Furthermore, weeds are developing increasing resistance towards pesticides, diminishing their effectiveness. Due to these concerns, industry sponsor Flux Robotics, is working to provide farmers with a pesticide free alternative. This project has focused on the design of the robotic arm, weeding tool and weed tracking control scheme for Flux's Autonomous Weed Removing Robot. Flux's existing robotic arm has been redesigned for improved capability and withstanding harsh environments. Prototypes of weeding tools were designed iteratively to improve weeding ability, while minimising design complexity and assembly cost. Finally, the weed tracking control scheme has been designed to establish a model for identifying and tracking weeds while the robot is in motion.

Watch video

Group members

• Thomas Howard
• Gabrielle Annese
• Madeleine Piercy 

Sentiment analysis is the process of assessing the sentiment of text, most commonly on a positive-negative scale. This typically involves either using a dictionary or lexicon, where words or phrases are matched with sentiments, or more advanced machine learning methods. For historical documents, we face the problem that modern sentiment dictionaries may not be applicable, due to shifts in sentiment associated with words over place, time and context. In my project I will use expert knowledge to manually annotate a set of documents, and then use these annotations to construct a domain-specific dictionary. This method is intended to help historians assess the sentiment of large sets of similar documents.

Watch video

Group member

• William Pincombe

In structural engineering, monitoring structure health is vital to prevent safety issues and premature deterioration. Detecting early structural damage involves tracking cracks for repairs and predicting service life. Automation using tools like Digital Image Correlation (DIC) aids accurate crack detection. However, current DIC algorithms only handle small deformations. This study focuses on developing algorithms and numerical codes for detecting and calculating crack opening and orientation. DIC data are post processed in DIC software at to generate displacement fields as inputs for the new algorithms. The new algorithms allow obtaining both crack orientation and crack opening from the displacements. Successful deformation data output enhances understanding of cracks, strain, and accurate replication of large, closely spaced cracks. This research improves structural safety by effectively assessing and addressing structural damage.

Watch video

Group members

• Aaron Rigg
• Digne-Christa Gateka
• Sochakra Da

Complex surfaces are manifolds of complex dimension 2 (that is, real dimension 4). In general, these objects behave very differently to their real counter-parts and have many remarkable properties. The Enriques-Kodaira classification organises the complex surfaces of compact type into ten distinct classes (using an appropriate notion of "sameness") and is an outstanding result in the theory of complex geometry. The first part of this project consists in gaining sufficient background in order to understand the classification. This involves understanding many standard results in differential and complex geometry. In the latter half of the project (which is yet to commence), further aspects of these surfaces will be explored. Even though the classification of these surfaces is well understood, there are still plenty of opportunities for extending interesting results in this contemporary area of mathematics.

Watch video

Group members

• Radee Tchorbadjiev

This project endeavours to develop a team-based robotic system designed to resolve simulated fires. The primary objective is to establish an infrastructure-less visual information communication network between a team of robots, which includes Unmanned Ground Vehicles (UGVs) and Unmanned Aerial Vehicles (UAVs), enabling them to communicate seamlessly. The network does not require an internet access point, rendering it especially valuable in emergency scenarios. The project's methodology involves the creating a fully-rendered simulation mimicking a fire emergency situation. This is implemented within a testbed featuring four distinct zones and LED holders representing simulated "active fires."

The outcome of this project is to underscore the effectiveness of collaborative multi-robot systems when confronted with emergency situations. Through the successful establishment of a communication network that enables robots to interact and make autonomous decisions based on predefined operational policies, this project aspires to contribute to the objective of ensuring enhanced safety and more efficient responses during emergencies.

Watch video

Group members

• Avani Karandikar
• Max Marriott

Have you ever wondered why people say rocket science is hard? Rocket dynamics, or how a rocket behaves, are complicated and inherently unstable. By investigating systems with similar dynamics, students can learn how to control rockets and even more complicated systems. The goal for this project was to control a double inverted pendulum using advanced control techniques and computer vision. This project was designed to use only equipment from the university, so it could be replicated in the lab for other subjects. A design for the double pendulum system was produced, where a rotary aluminium link was attached to existing hardware. A combination of software, coding, and mathematical derivation was used to create controllers for the system. A virtual model of the double pendulum system was created to test the controllers before using the hardware and computer vision.

Watch video

Group member

• Mitchell Raw

In a world increasingly reliant on interconnected devices and systems (IoT), our project aims to simplify the process of creating IoT testbeds. These testbeds act as crucial tools for researchers and professionals to gather data and conduct experiments in the realm of cyber-physical systems.

In our project, we're working to streamline the creation of these testbeds. This encompasses tasks such as configuring computer hardware, setting up network connections, and deploying software infrastructure for monitoring, managing, and tasking the testbed.

Recognizing the repetitive nature of setting up such testbeds, we are exploring the development of a software platform that automates the steps involved in the previous tasks as much as possible. This way, when provided with a set of computers and the architectural blueprint, the platform could autonomously handle the intricate process of assembling and configuring the IoT testbed, saving time and effort for researchers and practitioners.

Watch video

Group member

• Ethan Selway
• Spencer Edwards
• Zane Pattinson 

In an era where trust in the media is of utmost importance, television remains Australia's most relied-upon news source. With 90% of viewers valuing unbiased reporting, detecting the neutrality of news providers is more crucial than ever. But how do we quantify the amount of bias within the media?

Let us take a journey through the Australian television news landscape and into the constantly changing world of topics and sentiments within televised news. We aim to build a comprehensive bias measure to decode the truth behind the headlines and empower the public with the tools they need to make informed choices about their news consumption.

Watch video

Group member

• Irulan Murphy

Visible Light Positioning (VLP) is a promising technology that enables future indoor lighting systems to provide both lighting and positioning services with high accuracy. As light sources are increasingly replaced by light-emitting diodes (LEDs), VLP systems can use LEDs as signal sources to transmit positioning information. The receivers typically use either photodetectors or cameras. Our project will compare the hardware variations among the existing VLP systems and investigate the hardware requirements to explore the state of the art and future research directions in VLP. The trade-off between hardware requirements and performance such as positioning accuracy, response time and system limitations will be investigated. In the second part of our project, an innovative solution in VLP will be proposed and its performance will be analysed through simulations and experiments.

Watch video

Group members

• Yingying Luo
• Yucheng Xie

Two and a half millennia ago the Ancient Greek philosopher Xenophon wrote “To be a successful farmer one must first know the nature of soil,” a statement that has continued relevance for farmers of the twenty-first century. A current limit on the wide-spread implementation of carbon sequestration practices is access to fast, accurate and reliable soil organic carbon measurements. Our project aimed to meet the needs of modern farming by developing an efficient, in place soil sampling system for detecting organic carbon content using Near Infrared (NIR) sensor technology. This year we designed and manufactured a working prototype that attaches to the University of Adelaide’s autonomous tractor and is capable of autonomously measuring, storing, and outputting data about the organic carbon content of soil at depths of up to 500mm.

Watch video

Group members

• Kristen Coles
• Alleigh Hamnett

Cave systems serve as scientific time capsules due to their isolation from the world and human activity. Scientists endeavour to research cave environments to gain a greater understanding of their formation and discover remarkable scientific artefacts such as fossils. The Cave Exploration (CaveX) robot project, which began in 2021 and is currently in its third iteration, aims to provide a more robust way to map complex cave environments. The prototype has been developed to provide mapping of cave regions which are difficult for humans to enter and has a small Light Detection and Ranging (LiDAR) sensor for generating detailed 3-Dimensional maps of the cave structure. This year's iteration of the project has implemented autonomous functionality which eliminates the need for constant user control and increases its utility in a cave environment. To facilitate autonomous functionality obstacle detection, path planning, and gait control algorithms have been implemented.

Watch video

Group members

• Riley Groome
• Tyler Groome
• Luka Moran

Sour beer is a type of beer characterised by its signature acidic taste, achieved by the addition of Gram-positive bacteria producing lactic acid as a metabolic by-product, to the batch. Traditionally, wild microbes were exposed to the wort for several months allowing bacterial fermentation to occur, souring the beer in the process.  A new technique for sour beer production, kettle souring, has provided an alternative to this timely process; whereby lactic acid-producing bacteria, primarily either Lactobacillus or Pediococcus, are carefully added to the wort prior to yeast fermentation. Whilst kettle souring can be achieved in roughly two weeks, currently only Lactobacillus is a recommended species for the process, offering a highly limited taste profile. Our project aims to categorise flavours produced by Lactobacillus compared to Pediococcus, ultimately indicating whether Pediococcus can be used as an alternative species for kettle souring, subsequently expanding the flavour profile of kettle sours.

Watch video

Group members

• Stella Lock
• Hiba Salha

Suppose you have the set of points satisfying some equation in some space. We can, for example, count the number of points in each line, plane or even higher dimensional subspaces. It makes sense that the equation determine these counts, or the "combinatorial profile" of these points. We can then ask the question:

"If we have a set of points, with the combinatorial profile as if they solve a given type of equation, is there an equation for which this set of points is the solution?"

In particular, the type of equations we investigate are "quadrics", and the space we look at is "projective finite space".

The thesis is a collection of results within this research area and the proof and conjecture of results with respect to characterising quadrics by their combinatorial profiles.

Watch video

Group member

• George Savvoudis

In the world of satellite communications where we rely on signals travelling between Earth and space, there's an exciting innovation called the Transmitarray (TA) antenna. They stand out from traditional antennas, thanks to their ability to provide high gain, high radiation efficiency, and the flexibility to change the direction they point towards.

Unlike the traditional phased array antennas, TAs have proven to be a cost-effective alternative with a low-profile design. The main goal of our project was to explore different TA designs and techniques to create a TA antenna that produces circularly polarized signals in the X-band frequency range, which is a desired capability within satellite communication. Using the electromagnetic simulation software – ‘CST Studio Suite’ simulations of the complete TA antenna configuration were undertaken, producing promising results. A prototype was manufactured and tested in order to verify the simulation, the findings have proven our TA designs suitability and performance capabilities within the satellite communications realm.

Watch video

Group members

• Jack Hughes
• Shashinda Madurapperuma

Researching marine ecosystems without risking researcher safety or disturbing ecosystems is a challenge that requires careful consideration and innovative solutions. To address this challenge, our project aims to develop an Autonomous Underwater Vehicle (AUV) capable of ocean navigation, drawing inspiration from cuttlefish and their versatile movement abilities. Our project uses bio-inspiration methods to find naturally occurring solutions from the cuttlefish species, and applies the systems engineering process by gathering requirements, creating designs, constructing, and testing. We have made notable progress in designing, constructing, and testing two distinct propulsion system prototypes: one that mimics the cuttlefish's water jet propulsion, and another that harnesses fins for underwater mobility. Additionally, previous water leakage of the fin mechanism has been addressed by a redesign. These achievements mark significant steps toward the realisation of an ideal cuttlefish AUV to advance research tools and deepen society's understanding of marine ecosystems.

Watch video

Group members

• Jake Costi
• Ella Marie Di Stasio
• Tom Paul Ducker
• Jason Lor
• Gregory Lott

Surf Life Saving South Australia (SLSSA) are facing an ever-growing problem of managing resources, coordinating teams, and utilising assets. Our project aims to gather the required information to enable access through a centralised dynamic application, useable via laptop, tablet, and phone. The SLSSA were looking for a way of coordinating shift plans between teams, resources and assets before volunteers are on duty, ensuring optimal beach safety.

This meant coordinating with SLSSA volunteers at multiple levels throughout the organisation and gaining a thorough understanding of their life saving operations. Throughout the project we had the opportunity to visit the central hub at West Beach and observe the how their teams currently communicate and coordinate beach safety protocols.

Through combining the use of cutting-edge technology and continuing to grow our knowledge about SLSSA, we have been able to develop our application and can't wait to see it in action throughout the organisation.

Watch video

Group members

• Mitchell Follett
• Sarah Telford
• Sophie Porter

In my research, we focus on a class of spaces known as Lorentzian homogeneous spaces. These spaces exhibit a unique type of symmetry, akin to how a basketball looks the same no matter how you turn it. Our particular interest lies in understanding the concept of isotropy and its impact on the overall geometry of these spaces, specifically the holonomy which is closely related to the curvature.

To grasp the idea of isotropy, imagine yourself standing at a specific point within one of these spaces and observing your surroundings. Isotropy can be thought of as the set of transformations that you can apply to the space while keeping that point fixed and unchanged.

Watch video

Group member

• Steven Greenwood

We are taking an innovative leap into autonomous vehicles by creating a remote-controlled scale car that can race head-to-head against other vehicles. Vision-Based Autonomous Racing has created a 1:16 scale autonomous car specification to be used with an autonomous pipeline that will be capable of safely racing against multiple vehicles using only cameras and inertial sensors. We are continuously improving these modules for better performance, using new machine learning research, and hoping to allow other teams to do so in the future. Having completed two functional cars and a foundational autonomous racing pipeline, we are moving toward improving the racing capabilities for a better understanding of how far vision-based autonomous racing can go. This blend of racing excitement, competition and technical innovation offers a unique, future-oriented perspective to autonomous vehicle research and student engagement.

Watch video

Group members

• James Donlan
• Robbert Symonds
• Levi Smith

As Australia's top wine bottling contractor, Vinpac International is taking the lead in industrial efficiency. We've developed a cutting-edge system designed to minimize operational expenses and reduce their carbon footprint. Imagine a factory where conveyor belts proactively pause during job changes and stoppages. Leveraging advanced computer vision technology, our solution seamlessly integrates with revolutionary Ignition SCADA systems to automatically detect carton movements and manage conveyor activity in real-time. The result? Conveyor belts that pause and restart automatically during job changes, saving significant operational costs, lowering their carbon footprint and minimizing energy waste across their 24/5 factory runtime. The partnership between The University of Adelaide and Vinpac International has set a new standard for sustainability and cost-efficiency in the bottling industry.

Watch video

Group members

• Leo Jones
• Maan Patel

It is getting easier and easier for anyone to voice their opinion through the use of social media. Twitter is one of many popular platforms where users can do so. Sentiment analysis is a mathematical method that allows us to identify the emotional tone (sentiment) behind a body of text.

However, many factors affect the sentiment of text. During the analysis of Twitter tweets, the impact of gender on sentiment will be explored. Most sentiment analysis gives a single sentiment to each word. In this project we will consider assigning multiple sentiments to words based on information about the author such as their gender and income.

Watch video

Group member

• Oska Dubsky-Smith 

ChatGPT is an AI chatbot currently revolutionising the world and transforming many different industries. ChatGPT can assist in accomplishing a wide range of tasks, from something as straightforward as generating ideas for birthday gifts for loved ones, to tackling technical challenges such as coding to support programming development. In anticipation of this revolution, our project investigated how this tool can be harnessed to innovate and improve education within the field of engineering. To accomplish this, we first studied the capabilities of ChatGPT, finding its strengths and weaknesses. We then looked at how engineering is taught, and the skills students need to become great engineers. After this we explored how ChatGPT has been applied in other areas of education and what methods could be applied specifically to engineering. Our project has demonstrated that ChatGPT can be a powerful tool to assist in engineering education, especially when given detailed and tailored prompts.

Watch video

Group members

• Ian Casey
• Zachary Luke
• Joseph El Shafei
• William Petterson

A fundamental object in complex geometry is a Riemann surface, which is a one-dimensional connected complex surface in which we can undergo complex analysis. Fundamental examples of Riemann surfaces include the complex plane, tori and even a sphere (called the Riemann sphere). This project in pure mathematics aims to explore an important class of maps between Riemann surfaces, called holomorphic maps, and specifically the maps between them -- so maps of maps! Holomorphic maps have numerous powerful properties, which results in them not only having many applications in mathematics, but also in other areas of STEM (for example, applications in physics). Relevant to the notion of maps of maps are entire curves, which we will think of as a map that smoothly transforms one holomorphic map into another (so a map of maps). This project primarily looks to explore these entire curves utilising techniques from Riemann surface theory and general topology.

Watch video

Group member

• Tyson Rowe

This project focuses on studying micro and nanostructures through mathematical and computer-based methods, without experiments. Its main goal is to develop theories for these tiny structures and apply them in real-world areas like chemistry, medicine, and electronics.

The project seeks to unravel the behaviours and properties of these minuscule structures, helping scientists and engineers design better tools for various applications. It has two primary objectives: understanding the rules governing how these structures move and behave, and verifying these rules by comparing them with existing models and data.

The project aims to establish a strong knowledge foundation with accurate equations, contributing to the progress of micro and nanostructures. It's like solving a captivating puzzle in the realm of the tiniest structures, paving the way for innovation in science and technology.

Watch video

Group member

• Geng Yang

Laminated structures are the backbone of modern structures, advanced or simple. You will find them in the airplane used to take you from country to country, and you might even find them on your wall! Laminated structures are also commonly referred to as sandwich structures because they have an outer layer, and a core. 

Given the prevalence of these structures, our research focused on the behaviour of a laminated beam to see what it will do under certain conditions. We also aimed to develop a method to predict its behaviour using mathematical calculations, and a mathematical modelling software. After comparing the methods of prediction to each other, and the real model, the most efficient process can be chosen as the best way to determine the mechanics of a laminated beam.

Watch video

Group members

• Abdelqader Baba
• Nasir Ahmet
• Taeik Kim
• Latif Ruzehaji

Much of pure mathematics consists of the study of a class of objects with similar structure and the structure-preserving maps (morphisms) between them. Category theory captures this notion by abstracting away the low-level details and focusing on more general and fundamental aspects of objects and morphisms. Of particular interest to this research project is the category of simplicial sets, which facilitates a combinatorial model of topological spaces. The data of a simplicial set can be loosely thought of as both the `building blocks' and `instructions' required to build a topological space. To provide an everyday analogy, consider a LEGO set with bricks and instructions. The LEGO bricks themselves are analogous to the building blocks of a topological space—some may be thought of as points, some as lines, others as surfaces and so on—and the accompanying instructions describe how they are connected. There is a fundamental link between simplicial sets and higher category theory, which is presented in this display.

Watch video

Group member

• Matthew Drown

The record for the highest jumping robot is 33 meters. A study published in 2022 by Nature defines the research for the record-setting design. The design utilizes a compression and tension hybrid lightweight model to set the record. The aim of the project is to design a record-breaking model while also exploring the potential of a more cost-effective method. The design gaps identified are material selection of the energy storage components, drag reduction, and lighter power storage alternatives. Multiple prototypes are constructed and the most suitable design is identified through thorough testing. A final model is built and tested to confirm the results. A cost-effective analysis is also conducted on the final model as well as other adequate models constructed. These findings provide fundamental research for the continuation of the project.

Watch video

Group members

• Benjamin Morris
• Kyle Ochoa

Morphing wing technology has been a part of flight since the first aircraft. However, limitations of previous technology and the G force experienced by pilots has restricted development. With the rise of unmanned aerial vehicles (UAVs), this technology has seen a rise in interest and research. Morphing wing aircraft have the potential of becoming incredibly aerodynamically efficient, which can lead to decreased fuel consumption, improved travel times and reduced environmental impacts. Research into the effects of adding morphing wings to an entire UAV model is limited, specifically, the effects of wing twist on aircraft efficiency and manoeuvrability. Out project aims to provide additional data on wing twist performance while also furthering the use of composites and modern morphing techniques in their design.

Watch video

Group members

• Ryan Campbell
• Thomas Taylor
• Samuel Rowe
• Samuel Markesinis

Installing major updates to your computer can run the risk of breaking features you once relied on. Windows releases a new update every month, if you can’t suddenly use your printer anymore how can you be sure it wasn’t Microsoft’s fault? During system updates, it is possible that certain programs and features stop working, this project will figure out if that has happened and tell you which files were responsible. Our project will show you in real time the health of your system and how many programs have been broken because of an update. In short, the aim of the project is to evaluate the health of your computer during, and after, a major system update.

Watch video

Group members

• Chris Richardson
• Mohmad Aqmal Pulle
• Nathan Giang
• Stefan Andrea Parenti

Visual object tracking is pivotal for breakthroughs in healthcare, driving, and safety. Yet, current systems struggle to recognise unfamiliar objects, lacking the "open vocabulary" ability that can help one generalise. More specifically, we believe the association between natural language and vision is particularly effective because the richness of language gives humans flexibility and equips them to describe objects clearly. This motivated us to investigate open-vocabulary-based approaches for the tracking task.
Our project explores powerful vision-language detection models that use an image as well as a natural language prompt for producing detections, hence displaying open-vocabulary capabilities. These detections are further used by multi-object association algorithms, which can link the detections over multiple video frames to track the desired object, effectively solving Single Object Tracking.

Watch video

Group member

• Jash Vira 

Imagine we have a magical tool called "Super Glasses". This tool can view images and read text to recognize things. But right now, it simply matches images with text without truly understanding their meaning. We want the "Super Glasses" to be smarter!

To achieve this goal, we employed some special techniques. Firstly, we used "Support Vector Machines" to help it classify things better. Then, we incorporated "CLIP" and "ImageNet", two large databases of images and text, to enrich its knowledge. Lastly, we applied the "Stable Diffusion" technique to ensure its learning is more stable.

Through these methods, we hope that the "Super Glasses" can not only accurately recognize things but also adapt to new situations without biases. This way, it can assist us even better!

Watch video

Group member

• Jiaxin Wang

Although we may not be aware, we contribute to event logs all the time in daily life. Whether this is through your computer crashing, when logging into your favourite social network platform, or doing some online shopping. The system keeps track of all this user information in the form of event logs. Process mining is simply a tool used to extract information from these event logs and produce models from which we can further analyse the system. Our project investigates the flow of this information using a process mining modelling technique called a Petri net. Using this model, we can determine the average time a user takes to complete an operation (such as finalising an online shopping purchase).

Watch video

Group member

• Ethan Johnson

According to the Open Source Robotics Corporation, autonomous systems will gain traction as a standard in transforming industries such as manufacturing, healthcare and agriculture. Anticipating the ongoing surge in demand for intelligent robots, our project focuses on fostering teamwork within the realm of robotics. Our primary aim was to empower a collective of robot agents to seamlessly accomplish team-oriented tasks with efficiency. By using the information around them, these robots collaborate to perform their tasks. In pursuit of this objective, we harnessed ground robots equipped with specialised manipulator arms. This strategic setup enabled each robot to adeptly undertake distinct tasks aligned with their designated roles. To showcase the project's aims, we constructed a testbed scenario featuring a pixel art stand. This scenario utilises the robots' capabilities by employing the manipulator arm in action.

Watch video

Group member

• Jack Lee
• Jai McElroy

On the 18th of June 2023 an OceanGate-operated submersible made a routine voyage to the world-famous Titanic shipwreck, never to return. A proposed cause of the tragedy was the lack of non-destructive evaluation of the vessel's hull, leading to the unmonitored degradation of the material integrity and subsequent catastrophic implosion. The goal for our project was to design and manufacture a 3D imaging system using terahertz frequency (one trillion cycles per second) radiation to evaluate fibreglass composite materials used to make marine vessels. By observing how terahertz radiation interacts with matter, we can look inside materials to assess for damage. We developed software and hardware to move a sample in three dimensions and measure the reflected radiation to construct 2D and 3D images. The imaging system will be used to advance research into using terahertz imaging for evaluating the internal characteristics of compromised marine composites.

Watch video

Group members

• Fraser MacPhee
• Elijah Schutz

Given the urgent and escalating environmental impacts of climate change, the need for a reduction in greenhouse gas emissions is paramount. High quality research is essential to develop and refine alternative efficient energy sources through production and utilisation strategies. Laser diagnostic experimental facilities are critical to enable this research. The University of Adelaide has established itself as a world leader in the application of laser diagnostic techniques to improve the understanding and development of energy systems. To further augment these capabilities, a new laser diagnostics facility is being established on campus. This honours project is dedicated to optimising the management systems of this facility, ensuring maximum efficiency and effectiveness in laboratory operations. This endeavour will significantly increase research output, empowering our researchers to make significant contributions towards resolving the impending environmental crisis.

Watch video

Group members

• Blake Cranna
• Thomas Johnson
• Jacinta Fedele
• Caitlin Stallan

In today's digital age, one of the biggest challenges is deciding how to share important health information while safeguarding privacy and anonymity. Imagine having crucial health data to share but wanting to do so without revealing your identity.
Our project is all about simplifying the decision-making process when it comes to health data privacy and anonymity. We've developed a tool that acts as a trusted guide, like a knowledgeable friend, helping you make smart choices about sharing your health information without disclosing your identity.

The outcome of our project is substantial - it leads to better decision-making concerning health data sharing, enhanced privacy protection, and the preservation of anonymity. We've made it simpler for individuals and healthcare organizations to make these critical choices, creating a safer and more secure digital healthcare environment. 

Watch video

Group members

• Vinay Kumar
• Maximilien Schuller

In the agricultural process, the monitoring of crop growth and post-harvest storage plays an important role in ensuring the quality and consistency of the crop. Typically, this process is done through a manual visual inspection of the crops which is relatively inefficient and inaccurate when done on a large scale. This can lead to problems in identifying critical issues like diseases and spoilages in the crop which are often times difficult to detect with the naked eye. 

Therefore, an alternative solution proposed by the group would be to leverage multi-spectral technology and build a camera from scratch which can detect a combination of visible and infrared light. This camera would consist of multiple camera modules and a small computer which can rapidly and efficiently access the conditions of agricultural products. Additionally, the camera would contain several user-friendly features such as exposure adjustments, auto-focus, and a simple graphical user interface (GUI). The outcome would lead to the user being able to obtain a multi-spectral image.

Watch video

Group members

• Yangyang Cui
• Jia Ming Lai

Autonomous vehicles are an important area of research that aims to improve the safety of transportation and the mobility of people. With the rapid advancements in AI technology, there have been significant developments in the field. However, we still don't have fully self-driving cars that are commercially available and it is a challenge that many major automobile companies are currently working on. To make this vision a reality it would not only require progress in tech but also legislative backing as well as trust from the community. One way this could be achieved is through autonomous racing. Racing is one of the most popular sports with millions of people watching it across the world. It provides a playground for testing as well as exhibiting the current capabilities of AI in a complex environment. Our project aims to investigate the optimization of autonomous vehicles and their implementation in a racing environment.

Watch video

Group members

• Anirudh Kandari
• Zheyuan Zhu

This project endeavours to leverage emerging artificial intelligence (AI) methodologies, particularly natural language processing (NLP) techniques, to establish a comprehensive database of carbon material electrocatalysts for oxygen reduction reaction (ORR). Given the critical role of ORR in driving renewable energy technologies such as water splitting and fuel cells, this project seeks to train NLP models using research publications. The primary aim is to extract insights strategies including material functionalization swiftly and dependably, heteroatom incorporation, hybrid structures, and nano-structuring to enhance the electrocatalytic performance. Through integrating AI technology in catalysis and materials science, the outputs of this project provide a promising research paradigm in the interdisciplinary of data science and chemical engineering.

Watch video

Group members

• Mohammad Alfadhli
• Mohammed Alotaibi

Text summarisation is a tool that almost everyone uses in their daily lives. It can summarise 10 news articles or scientific articles into 10 short sentences or shorten student's essay to fit into the word limit. Text summarisation models have been supported by exponentially growing amount of data online, making it aware of a wide range of social events. However, as it is trained with an enormous amount of data, it gets "confused" sometimes and can become unfaithful to text input, which we define as hallucination. Even with large language model like GPT-4, it hallucinates 30% of the time. Our project investigates the balance between faithful and factual in abstractive summarisation models, focusing on medical domain, where both factual and faithful are important. With the understanding of factuality and faithfulness of summarisation models, we can build a safe and reliable summarisation models.

Watch video

Group member

• Mong Yuan Sim

In the world of healthcare, we're working on a special project to make medical tests faster and more affordable. Medical tests help doctors figure out what's wrong with people's health, but they can be slow and costly. We're using a clever tool called the 'Limited Query Algorithm' to speed things up. It's like a shortcut for solving health puzzles. We're also using the power of computers to assist us. Right now, we're organizing data and preparing it, like cleaning up a messy room. Then, we'll use our computer tricks to see if we can make diagnosis better and cheaper. We're making sure our ideas are safe and helpful for doctors and patients. We're determined to make medical tests better for everyone!

Group member

• Minghan Zhou

Impedance is a fundamental electrical property found in all types of electronics. Students can explore deeply into circuit designs and electrical devices with the assistance of an impedance analyser. However, a single commercial impedance analyser can exceed 30,000 AUD in cost, making it impractical for the university to acquire one for each lab bench. Our project is dedicated to creating a cost-effective impedance analyser tailored for university laboratories and research settings. We are utilizing Analog Discovery, a versatile waveform generation and measurement device produced by Digilent, as the foundation for our analyser. Furthermore, we are using MATLAB and LabVIEW, two robust software development tools, to drive the impedance unit. We have successfully developed an impedance analyser that costs only a few hundred dollars, seamlessly integrated with the software platform we've created for user-friendly operation.

Watch video

Group member

• Man Lai Chan

This project aims to explore the application of AI technology in multi role chat scenarios by studying the development of NLP (Natural Language Processing) technology in summary and question answering tasks. At the same time, a visual AI multi-character chat application based on NLP will be developed by combining web technology. This project verifies the contribution and performance of artificial intelligence technology in multi chat scenarios by randomly sampling real data generated by the application and conducting qualitative and quantitative analysis of the sampling results. During the sampling process, participants will be fully informed of the research content and risks, and their privacy will be protected through information encryption. The final verification results indicate that AI performs excellently in Q&A and summary tasks in multi role chat scenarios. And combining network technology can solve some practical problems encountered by different groups in production, life, and learning.

Watch video

Group member

• Xiangcheng Chang

In diagnosing abnormalities or injury in pelvis, X-ray radiography plays a crucial role. However, manual annotation of anatomical landmarks in X-ray images is time-consuming, and there might be differences in measurement for different radiologists especially with different levels of experience. The aim of the project is to apply deep learning techniques to perform the automated hip landmarks detection. While implementing a deep learning algorithm, limitation of X-ray training dataset is challenging due to the lack of large and diverse dataset in medial field. To overcome this challenge, semi-supervised learning that uses less dataset is applied as the deep learning techniques. Therefore, this automated system will perform anatomical landmarks detection using limited dataset, with potential applications in the detection and monitoring of pathology.

Watch video

Group member

• Khine Thin Zar Htun

In a scenario where the environment has a very high possibility of uncertainty and unpredictability, the ability to track object or target in the area with minimal risk as well as response to unanticipated events is a challenging task. Creating a vehicle with features to tackle the problems is a crucial solution. The goal of this project is to develop an Artificial General Intelligence (AGI) to drive an autonomous vehicle as well as perform specific tasks given in a complex unknown environment. Individual systems which will be combined has been developed to perform target tracking and path planning. Simulation has been made to explore the individual systems as well as the combined system performance towards the sensor and the vehicle. The outcome of this project is an autonomous vehicle that possesses the ability to autonomously track targets, dynamically plan paths, and make reasoned decisions in unforeseen scenarios.

Watch video

Group member

• Anas Ahmad
• Sean Yong

In today's tech world, we have a lot of software available in the market that provides a lot of features and functionality. This race of providing more and more features to users makes the main functionality get ignored. This can be prevented by analysing the user flow with the help of calculating limiting probabilities of a Markov chain. However, this process is expensive in time and most companies avoid this process to save cost. This project aims to reduce the time complexity and also use the optimum amount of computation power available by using efficient algorithms. With this idea, we will be able to analyse the feature through event logs and be able to deduce whether the feature is worthy enough to keep or not.

Watch video

Group member

• Rahul Ghetia

In today's world of intelligent systems, we want to enable computers to learn and perform well in changing environments, just like humans. However, in real-life applications, it is often unrealistic to let the computer and the environment interact and learn in real time, so offline reinforcement learning is needed to deal with this challenge.

This project aims to verify the performance of offline reinforcement learning under Atari. I used a method called CQL (Conservative Q-Learning) to help the computer learn from pre-collected data without interacting with the game environment in real time. With this approach, we can allow computers to learn from past experience and improve performance without the need for real-time feedback.

The results show that the CQL algorithm performs well on Atari games, suggesting that offline reinforcement learning can be useful in these situations.

Watch video

Group member

• Haoyi Li

The vast expanse of our oceans remains one of the least explored terrains on Earth, concealing numerous mysteries and potential discoveries. In the realm of marine exploration, accessing deeper and intricate parts of the oceanic world presents a significant challenge. To address this, the deployment of technologically advanced drones becomes indispensable. Our project focuses on the development of a miniature underwater drone, specifically tailored for exploration in clear waters. Equipped with a Raspberry Pi 3a+ and a dedicated camera module, it employs servo and DC motors for refined navigation, a TF mini lidar for accurate depth perception, and a LiPo battery to ensure sustained exploration. Upon successful integration and testing, the drone demonstrated its capability to capture high-definition imagery from clear underwater environments. Through this initiative, we aim to facilitate a deeper understanding of pristine aquatic realms, paving the way for enhanced scientific research and environmental conservation.

Watch video

Group members

• Muhammad Alif Aiman Ahmad Fadzil
• Mohammad Nazif Mohamad Sobri
• Yang Li

In modern agriculture, efficient plum harvesting is a challenge, with ripe plums often hidden amidst complex images. The accuracy, speed, and robustness of plum detection directly impact the efficiency and quality of plum-picking robots, which is a critical factor in today's farming landscape. This project addresses the real-world challenge of efficiently detecting objects, particularly plums, in complex images. The primary goal is to provide farmers with a reliable tool for identifying ripe plums amidst challenging conditions. By combining manual labelling and deep learning-based object detection and segmentation algorithm, we can automatically identify and delineate plum trees and their fruits in images, facilitating yield estimation and selective harvesting. It's like giving our farmers superpowers to enhance their plum cultivation! The outcome of our project is a highly efficient object detection model tailored to the agricultural context. This model equips farmers with a valuable tool to detect plums swiftly and accurately, even in challenging scenarios.

Watch video

Group member

• Surabhi Ramesh

In the bustling financial world, traders are always looking for ways to maximize profits while minimizing risks. However, human emotions and the ability to process large amounts of data are limited, which can lead to missed opportunities or potential losses. I utilize advanced computer algorithms to scan, analyse and respond to market data in real time. By eliminating human bias and emotion, I implement an objective, data-driven approach to trading. Not only is this approach faster, but it is also designed to be smarter and can consider multiple scenarios and make appropriate decisions.

Watch video

Group member

• Shuaiju Lyu

This project is like a detective mission, but for numbers and patterns! We wanted to explore how messy or organised things are in the world. imagine you have a bunch of colourful candies, and you want to know how mixed up they are. That's what we tried to do with number and patterns. 

To do this, we used something called Recurrent Neural Networks (RNNs). Think of them as super smart detectives that can spot hidden patterns in the data. The best part is that RNNs can help us figure out how information changes over time. 
So we are finding out how messy or ordered things are, and this can help scientists, engineers understand the world better. 

Watch video

Group member

• Saurabh Prashar

Aim: The aim is to detect and count apples using machine learning and construct 3D segment for the images for apples.

Methodology: The project detects apple using machine learning tolls like YOLOv8. The segmentation and detection helps the model to do 3D reconstruction with SAM ( Segment anything model ) which further reduces the problem related to occlusion to detect apples with more accuracy and precision.

Result: I have successfully generated 3D construction of an image using SAM by detecting all the apples from our own trained YOLOv8 model.

Watch video

Group member

• Sushant Gupta

The project's main goal is to make it easier and more accurate for doctors to spot problems with people's hips by using advanced computer technology. In the world of medicine, doctors often look at X-ray pictures of patients' hips to find out what might be wrong. But these pictures can be a bit like solving a tricky puzzle because it's not always easy to see everything clearly.

To tackle this puzzle, we used computer smarts and special algorithms. We gave a computer program lots and lots of hip X-ray pictures and told it exactly where to find important marks on the hip bones. Then, the computer learned and got really good at finding those marks all by itself.

The result of our project is a fancy tool that helps doctors quickly and accurately figure out hip problems by helping them understand X-ray images better. It's a bit like having a super smart assistant who works with doctors to make sure they don't miss any important details in the pictures. This helps patients get the right treatment sooner.

Watch video

Group member

• Linet Maria Cherian

How tedious it must be for farmers or farmland owners to keep track of each and every plant of a crop type and gather information about the crop itself? How easy would it be for the parties involved in maintaining large farmlands if, from small tasks such as counting the number of plants belonging to a particular crop to more specific information regarding each plant could be automated? This can be done using Computer Vision (CV) algorithms which are special tools that help computers understand and recognize things in pictures and videos, such as objects or movements, using clever math and patterns. By investigating different CV algorithms that can be used to not only detect but track each and every plant that belongs to a specific crop, this project presents a CV model to detect and distinctly track plants in a farmland that is captured through a video feed.

Watch video

Group member

• Savini Abayaratne 

Think of the stock market as a giant puzzle where the goal is to pick the best combination of stocks to earn the most money. This combination is called 'Portfolio', for which many investors rely on traditional investment strategies to design theirs. Our project analyses specific numbers related to stock prices, such as their highs, lows, opening and closing prices, and sales volumes. With this data, our advanced computer program - Machine Learning, suggests a portfolio designed to outperform older strategies and other tech-based models. This model uses transformer architecture to predict the probability of stocks rising in the future, and then, with a scoring system, the best ones are chosen and the model keeps improving its choices using a learning method. Moreover, our tool is designed not to make impulsive decisions in a volatile market. With this approach, investors can get better returns for their investments in the stock market.

Watch video

Group member

• Revanth Phani Sai Medukonduru

Background: Think of Stochastic Boolean Function Evaluation (SBFE) as a method to answer yes/no questions without spending much. Like choosing which medical test to take first for detecting a disease or which features of a program to check to see if it's harmful or a completed but flipped jigsaw puzzle and your goal is to flip the pieces one by one to figure out which animal is on the other side.

Aim: We want to use this idea to make machine classify, a way computers learn from data, cheaper. Right now, teaching computers with a lot of data via deep learning can be pricey.

Methodology: We'll use SBFE methods that are good at answering questions without spending much. The focus will be on learning which data points (or features) are the most important to check first.

Outcome: We hope to create ways that make it much cheaper to teach computers to tell the difference between safe and harmful programs.

Watch video

Group member

• Tien Phu Ngo 

Our project is using technology to create a smart farm. Since the earth population will increase each year, we want to help farmers to grow their crops better and save the Earth by using fewer resources such as fertilizer, which is like food for plants. 

We are using computers to identify crops such as lettuce and capture their uniqueness, just like each human is slightly different and determine how many crops are present on the farm. we need this because usually robots will double count crops when the camera sees it twice. Just like humans, robots can walk forward, backward and turn to the sides so sometimes they see the same plant more than once.

With our computer magic, it will help the robot count more accurately and when you have so many crops, the accuracy of the count makes a lot of difference in the resources.

Watch video

Group members

• Erick Hartawan

E-scooter sharing scheme (ESS) has become a popular option of first/last mile trips in the urbanised areas. The dynamic demand for ESS varies significantly over time and locations, making it challenging for ESS operators to deploy e-scooters to the right locations in advance. The goal of this project is to develop a machine-learning based spatiotemporal model to predict the e-scooter demand in Adelaide. A simulation model has been developed to simulate ESS demand in a 10x10 grid on land use type, time of day, and day of week and e-scooter trip interactions. Data generated from the simulation model is fed into a machine learning model for training and evaluation. The machine learning model will be valuable for ESS operators to analyse the rebalancing strategies to fulfil the user demands in time and at locations.

Watch video

Group member

• Jianguo Liu