Making Advanced AI Tools Accessible to Sports Practitioners
Themes: AI for holistic athlete performance and wellbeing / Data driven technological innovations for health and performance
Student type: PhD – Submit an Expression of Interest
Host University: QUT
Sports research objective/question: Understand how practitioners (sports scientist, bio-mechanists, etc) use computer vision/machine learning to quantity athlete performance.
- Develop new computer vision/machine learning methods to enable measurement of sports performance.
- Research programme would make use of AIS’s existing cloud-based “pipelines” product, which allows sporting practitioners to develop computer vision (CV) applications in a no-code environment.
Discovering Novel Strategies with Reinforcement Learning
Themes: AI for holistic athlete performance and wellbeing / Personalised performance and team performance
Student: Mengmeng Wu
Host University: QUT
Sports research objective/question: How can reinforcement learning (RL) be used to discover new sporting strategies?
- Research programme could consider a specific sport for which a rich simulation environment already exists (such as track cycling), or use a simplified simulation environment (i.e. for soccer or hockey, which captures rules and athlete limits) to discover strategies.
Supporting Sporting Knowledge Discovery with Large Language Models
Themes: AI for holistic athlete performance and wellbeing / Diverse data collection, integration and governance
Student: Candidate is starting soon
Host University: QUT
Sports research objective/question: How can sports practitioners use LLMs to support their work?
Develop methods to video and/or numeric data interpretable/discoverable through Large Language Model (LLM) queries. Research programme would seek to extend the use of LLMs to the sports domain, and in particular to retrieve multi-modal data (i.e. video clips, physiological data) from AIS/sports organisation data warehouses and aid/support the visualisation of this data.
Profiling athlete’s throwing mechanics from video using high precision computer vision
Themes: Personalised Performance and Team Performance
Student type: PhD – Submit an Expression of Interest
Host University: QUT
Sports research objective/question: Video machine learning for tracking athletic throw events – can we profile an athlete’s throwing mechanics from video and determine how/where the throw velocity is generated?
- Developing high precision models to auto track throwing implements from a fixed video camera
- How to apply outputs from computer vision to support coaches
AI and ML Techniques for Predicting Future Olympic and Paralympic Medal-Winning Performances: Developing reliable and accurate models to predict athlete performance and career progression
Theme: Personalised Performance and Team Performance
Student type: PhD – Submit an Expression of Interest
Host University: QUT
Sports research objective/question: The proposed research seeks to advance the understanding of Olympic and Paralympic medal-winning performances by employing Artificial Intelligence (AI) and Machine Learning (ML) techniques. The study aims to develop sophisticated predictive models that consider the diverse categories of sports in the Olympic/Paralympic Games, including racing, judging, and head-to-head sports. By categorising sports based on their inherent characteristics and employing appropriate AI and ML approaches, this research will provide insights into the career progression of successful athletes in various sports and allow for the development of more robust frameworks with which to assess long term athlete development and performance progression.
Swimming race profiling, prediction and simulation using multivariate sensor data
Theme: Personalised Performance and Team Performance
Student: Mackenzie Parker
Host University: QUT
Sports research objectives/questions:
- Develop individualised race profiles to predict race time distribution given different pacing profiles
- Map individual physiology to profiles
- Model profiles in the context of the career trajectory of swimmers and how they might perform at the next race
Scanning and off-ball effort in Rugby
Theme: Personalised performance and team performance
Student type: PhD – Submit an Expression of Interest
Host University: QUT
At any one point in a rugby game, only one player can be in possession of the ball, but their actions depend on what their 14 teammates and 15 opponents are doing. A line break, for instance, is not down only to the player with the ball. It may have been created by the lines other attacking players were running, or it could be prevented by the defence repositioning quickly after a ruck. Tracking this effort, or the amount of information a player takes on board before making a decision, is challenging.
- Can we develop AI solutions to track off-ball effort?
- Can we monitor how players scan the field?
- Can we use this information to better assess the impact of each player on the game?
Modelling cumulative physical load during performance and relationships to injury
Theme: Complex Systems Modelling
Student type: PhD – Candidate has been identified
Host University: QUT
Sports research objectives/questions: In sports like Rugby, training is monitored carefully by sports scientists such that risks to injury can be reduced. In game time, performance is less constrained, heightening the risks that players might endure physical load at higher levels of risk. There is little understanding of how available data might be used to provide better information on individual player loads during a game.
- How to mitigate risk of injury in a game by identifying a real-time measure of cumulative physical load?
- What are the key contributing factors in physical load?
- To what extent does a physiological baseline combined with cumulative physical load contribute to adverse outcomes?
Characterising and Quantifying Player Performance in Wheelchair Basketball
Themes: AI for holistic athlete performance and wellbeing / Diverse Data Collection, Integration and Governance / Personalised performance and team performance
Student: Penny Keats
Host University: QUT
Sports research objectives/questions:
Within wheelchair basketball, players are categorised based on their level of disability and are scored from [1..4.5], where 4.5 is the highest level of impairment. A team can field a maximum of “14 points”, forcing teams to field lineups with a mix of impairment levels.
- Can advanced metrics (beyond simple box-score stats) be developed to better compare players within and across disability classifications; to better quantify the value of and expected points from shots; and to create defensive metrics?
- How can player performance be disentangled from team performance to quantify player value?
Hybrid Computational Fluid Dynamics (CFD) and The Digital Athlete (TDA) Modelling
Themes: Holistic Athlete Performance and Wellbeing / Personalised Performance and Team Performance
Student: Tiri Pestrivas
Host University: QUT
Sports research objectives/questions:
-
- Focusing on aerodynamics impacted sports such as cycling, triathlon and wheelchair racing, can the Digital Twin and SuperSportFlow CFD be integrated for unified athlete performance and aerodynamic modelling? Research would assess the impact of environmental factors (heat, humidity, wind direction, etc.).
- How can a hybrid model be used to support optimisation of athlete equipment selection and race strategy? Research would assess impact of postures (e.g., tuck when going downhill) and related physical considerations (e.g., centre of mass location for stability)
- How can the model be extended to incorporate effects of multiple athletes such as in team pursuit?
Analysis of training and racing in sprint kayak of cohort data over time: incorporating high fidelity sensor data
Theme: Personalised Performance and Team Performance
Student type: Masters – Submit an Expression of Interest
Host University: QUT
Sports research objectives/questions:
- Understanding the effects of environmental impacts on training and race performance over time
- Understanding the race profile characteristics of athletes over an athlete’s career
Biology-based mental health correlations and predictions in athletes
Theme: AI for holistic athlete performance and wellbeing
Student type: 2x Honours or 1x Masters – Submit an Expression of Interest
Host University: QUT
Sports research objectives/questions: Does athlete mental health correlate with the athlete’s unique biological/physiological makeup throughout the season?
- Quantify phenotype using AI to predict personalised markers of stress and correlate them to validated tool in mental health. (Pre-season, start, mid, end of season)
- Collaborating with other phenotype and genetic assessments to provide deeper insights
Prediction of relative hormonal environment for all common menstrual cycle ‘types’ (natural and pharmaceutically modified)
Themes: AI for holistic athlete performance and wellbeing / Personalised performance
Student type: Honours – Submit an Expression of Interest
Host Universities: QUT & UNSW
Sports research objectives/questions:
- Develop an open source predictive model for individual daily female hormonal profile based on known variants, including both endogenous (natural cycle) and exogenous (medicated cycle) cycle types.
- Develop an AI model to understand the relationship between the predicted hormonal environment and daily tracking variables: physical/psychological/emotional.
Novel 360 degree viewing widget of menstrual cycle (hormone profile vs wellbeing/recovery/performance variables)
Themes: AI for holistic athlete performance and wellbeing / Personalised performance
Student type: Honours – Submit an Expression of Interest
Host Universities: QUT & UNSW
Sports research objectives/questions: How can the cyclical nature of women’s training and wellbeing data be better visualised to enhance ongoing understanding and learning?
- Normalise female cycle data for the representation of repeated patterns.
- Develop an open source novel 360 degree visualisation tool with easy to use viewing windows for past, present and future for optimal user experience and learning of personal traits which may be hormonally influenced in response to the individual’s hormonal profile.
- By improving the visualisation of predictive modelling (what is likely to occur in the next viewing window) is it possible to both enhance the user experience and informed decision making regarding, recovery, performance and menstrual cycle management.
Novel Bayesian modelling and advanced inference to infer hydration state and dynamics
Theme: Data-driven technological innovations for health and performance
Student type: PhD – Submit an Expression of Interest
Host University: QUT
Sports research objectives/questions: WearOptimo has developed a wearable sensor to sense the hydration status of individuals. The sensor uses micro-needles to penetrate the skin and sense impedance, which in turn allows hydration state (hydrated/dehydrated) to be estimated. In addition, the device has a temperature sensor and IMU integrated. The device is currently undergoing clinical trials to assess its ability to measure hydration state.
This project would use data from WearOptimo’s hydration sensor and develop novel Bayesian methods to model hydration state.
- How can hydration state and its change over time be inferred?
- How can the temporal change of hydration be modelled, considering variations in individuals and
environments?
Novel Neural Network Architectures for Robust and Explainable Biosensing
Theme: Data-driven technological innovations for health and performance
Student type: Candidate starting soon
Host University: QUT
Sports research objectives/questions: How can hydration be accurately measured and forecast?
WearOptimo has developed a wearable sensor to sense the hydration status of individuals. The sensor uses micro-needles to penetrate the skin and sense impedance, which in turn allows hydration state (hydrated/dehydrated) to be estimated. In addition, the device has a temperature sensor and IMU integrated. The device is currently undergoing clinical trials to assess its ability to measure hydration state. This project would use data from WearOptimo’s hydration sensor and develop novel neural network and deep learning based techniques to classify a subject’s hydration state, and forecast future hydration states.