How Australian Sports Commission has used Artificial Intelligence to help achieve Olympic Games gold

It may help us win gold medals in Paris.

Dr Stuart Morgan has spent the last decade developing AI applications for high performance sport at the Australian Sports Commission, and he says it’s already being used more widely than you’d think.

“I think most sports are using AI somewhere, it’s pretty ubiquitous,” Dr Morgan told CODE Sports.

This week he will address the National Sport and Physical Activity Convention in Melbourne on AI, and in the countdown to the Paris Olympics reflect on the way it contributed to the Australian swim team’s success at the last Olympics.

In Tokyo, the ASC deployed its ‘Sparta 2’ software with our swimmers for the first time.

The Sparta 2 is an AI-based program that tracks every swimmer in a race on stroke rate, stroke length, velocity in the pool, turn times and breathing, for the first time at the pool in Tokyo.

Usually this kind of data would take days to compile and be analysed for the next meet or event. Using this AI model, the data was instead coming through in real automated time and it was something only Australia had.

“It was a game changer for our coaches because suddenly they had access to metrics about their own swimmer’s performance and all the other swimmers in the race between events, between rounds,” he says.

The Australians could now check the data of everybody in a heat that morning and understand where improvements could be made, and where opponents weaknesses could be targeted, before a final that night.

“Nobody else in the world had it and we had lots of other swimming nations looking over our shoulders with curiosity.”

Morgan expects other countries to have developed their own technology three years on, however the system and its interpretations in Australia has also evolved further.

“The race dynamics data is a new way of thinking about races, so I think we’re getting better at understanding what that information means for preparing for a race and I think our coaches will have come up with clever ways of exploiting that.”

AI’s use in Australian sport dates all the way back to 2009 when Australian Hockey coach Ric Charlesworth was interested in calculating how the tactics he was using with his team influenced the work rate of his team’s opponents.

Would a certain defensive or offensive strategy result in tiring opponents quicker?

“There was no way at that point of measuring it,” Dr Morgan says, and given they did not have access to opposition GPS figures, it seemed an elusive task.

“So we set about building an early version of a computer vision system that would track the movement of players on both teams; measure how far they ran, when they were sprinting, how fast they were sprinting, how much territory they needed to cover and what the general shape of the play was.”

This predated AI as we now know it, yet it showed an application that AI could eventually be deployed through sport.

“It wasn’t until about 2015 that the maths that underpin the large neural networks that are what we call AI started to work. That was a real watershed moment in Artificial intelligence.”

All of the sudden much more complicated problems could be solved and Morgan and his colleagues started building models for sport.

Player tracking was the first achievement, then that evolved into human pose estimation which predicts outcomes based on the task and body position.

The Sparta 2 for swimming is the latest development in a constant evolution over the decade.

Morgan and his team have also built sport specific systems for pole vault and javelin, but as AI has become more accessible sports are finding solutions from a variety of sources. The ASC now provides a cloud-based system called Pipelines which allows sports to take things in their own specific direction.

“We’ve led the way in a lot of this work, and a lot of the systems that we’ve built were not commercially available when we first launched them,” he says.

“But artificial intelligence is becoming so ubiquitous and so accessible, it’s not as hard now to build a player tracking system, for instance, compared to 2009 when we first started working in this space.”

Dr Morgan’s next challenge will focus heavily on our own indigenous sport; Australian Rules.

Morgan will be leaving the ASC next month to take the role of Head of Machine Learning and Artificial Intelligence at Champion Data, the stats company closely associated with AFL football.

“We’re looking at ways to make our data collection even more accurate and more comprehensive,” he says about his new role.

“The virtuous cycle is that by gathering richer data and deeper data it’s possible then to make stronger predictions or go more deeply into analysis and prediction and understanding the nature of competition in the AFL and other sports.”

Whether it be inside club coaching teams, or for fans that pore over statistics, AI will play a part in the conversation of how AFL football is played in the near future.

Behind closed doors it already is in many areas of sport.

More Coverage

The lessons learned from ‘largest governance reform in sport’

Shannon Gill

The evolving crowd ‘connection’ helping Blues reclaim AFL power

Shannon Gill

“The technology has evolved more quickly than our thinking around the ways that we can use and explore AI, so the technology is really exploding at the moment.

“It’s one of the biggest shifts in the technical landscape and I would dare say in human history, but we haven’t really caught up to some of these opportunities yet.”

Dr Stuart Morgan is one of the speakers at the National Sport and Physical Activity Convention, to be held at the Melbourne Convention and Exhibition Centre, 27-28 June.