AIhub coffee corner – AI at the Olympics

10 September 2021

share this:
AIhub coffee corner

The AIhub coffee corner captures the musings of AI experts over a 30-minute conversation. Inspired by the Olympics and Paralympics, this month we discuss sports and the role AI and robotics could play. There are two aspects to this. Firstly, building AI-based robots to play sports (as is being done with RoboCup). Secondly, using AI techniques for performance analysis and improvement.

Joining the discussion this time are: Sanmay Das (George Mason University), Tom Dietterich (Oregon State University), Steve Hanson (Rutgers University), Sabine Hauert (University of Bristol), Michael Littman (Brown University) and Oskar von Stryk (Technische Universität Darmstadt).

Tom Dietterich: My colleague Jonathan Hurst and his team recently had their legged robot run a 5k. This was basically at a fast walking pace – it took 53 minutes – but it was possible on a single battery charge, which shows how efficient their systems are. It didn’t break any of its joints or wear out its shoes, which were problems they had in earlier versions.

RoboCup has always been an amazing vision, but it would also be great to have competitions for pole vault and high jump with robots. I don’t know if it’s AI, but it’s an amazing mechanical and robotics challenge.

Sabine Hauert: And the robot runners actually have applications. One of my colleagues had a start-up that was designing running robots that would run in front of you on a jog and get you to go a bit faster.

Oskar von Stryk: There are multiple dimensions here. I’d also like to draw attention to the fact that many sports are using data collection and performance analysis, using big data, AI and machine learning to improve their training.

What you also mentioned is the embodied AI question. It is related to what makes intelligence. For example, if you compare the successes in AI with chess or Go playing machines against humans then they are fundamentally different to embodied AI robotics systems, like soccer for example. Go is more complicated than chess, as there are more possible moves, but the number of moves is still finite. In the real world (for example, in soccer) there are lots of uncertainties – you need sensing to understand where the ball is, where the other players are etc. In Go, when you play your own move, it is carried out exactly. In soccer, you can’t make an exact move, there is a high level of uncertainty. When the legs move, there is friction and many other effects, and these affect the kick. The quest is being able to do intelligent analysis, planning, perception and action with not much information in real time: that is what intelligence is.

Looking specifically at the Olympics, I don’t think it’s a single discipline problem. If you look at humans, the key to natural intelligence is versatility. A human, compared to animals, isn’t the fastest runner, or the best climber, or the best swimmer, but if you were to combine hundreds of disciplines, all of the sudden, the human is the winner.

Humans can do amazing things with limited technical ability in their bodies. For example, the speed of neural signals is much slower than the control frequency in today’s robots. Taking darts as an example, if you throw a dart you need it to leave the hand at a certain angle and with a certain velocity. To do this precisely you have to send the signal from your brain to your hand before you are in the area where you need to release the dart. Today’s robots would just go into the domain of angle and velocity and then release it very quickly. Humans must do this beforehand. And that’s the only way we can use the benefits of our highly compliant elastic body. Elasticity compliance is a big advantage with natural intelligence.

Steve Hanson: The slow reaction time of humans tells us that a lot of what we’re doing is simulation. We simulate and then we hope for the best. The dart is literally thrown before we know where it’s going. This means there are a lot of ballistic perceptual action cycles involved. The brain circuits are kind of known, but exactly the way this works is still very obscure.

Oskar: Look at table tennis at the Olympics – how fast they play! One player must react before the other has even hit the ball. Otherwise it would be impossible to reach it. This is amazing and we have nothing similar yet in AI robotics.

Sabine: Some of these high precision sports, such as diving, it seems like a robot would be very good at. You could remove some of the uncertainty you get, for example, in the real world football environment. I think it would be fun to watch a robot, but I think the beauty of sport is that amazement you’re describing, of a human being able to perform these very high precision tasks. That’s the thing that’s interesting; it’s not the score so much, it’s the achievements that make the Olympics the Olympics.

Oskar: We could make a table tennis robot that could keep up with a human, but it would have to have superhuman sensing and action abilities. The question is, can we do it with human-like sensing abilities?

Steve: There was a diver I saw who took too many steps, and her simulation, or whatever she was calculating, was off so she ended up (after two perfect dives) going feet-first into the water. These errors are probably very informative.

Tom: Oskar, were you imagining a robotics competition where we have to design a single robot that can do several different sports?

Oskar: I think, first of all, that versatility is very important. I also think that the key to intelligence is predictability: how can we make better predictions for the future from fewer data? And, how can we learn better? If you look at embodied AI, also it’s the versatility, not the specialisation.

Tom: It’s interesting though to see how different the bodies of the athletes are depending on their sports. So, humans are specialising, partly through selection and partly through training, to be good at one thing.

Sanmay Das: I was thinking we need a robo-decathlon. Due to the range of events in the decathlon (100 metres, long jump, shot put, high jump, 400 metres, 110 metres hurdles, discus, pole vault, javelin and 1500 metres), the person who wins this has the unofficial title of World’s Greatest Athlete, because they are like the all-around champion. Of course, it’s all track and field events.

Some events are deliberately combined to test. For example, the biathlon – the combination of skiing and shooting. I always thought this combination was utterly bizarre. But, it turns out that the reason it’s so hard is that when you’re skiing your heart rate and exhaustion level is so high that it is difficult to shoot accurately. It’s selecting for people who can be calm under very high stress.

These multi-event things are interesting and, I wonder, if we were designing multi-event robots, what would be the events we would be interested in?

Sabine: It would almost be nice to not have human events. I like the idea of robot competition for the sake of a robot competition. We’d be excited for the engineers behind the robots and what they’ve achieved. So, those 10 tasks – the robot decathlon – could look very different.

Oskar: I think it would be more interesting to combine physical and mental skills somehow. I remember a show in Germany that consisted of alternating rounds of boxing and a board game. To win you need to be not too bad in one and really good in the other, I would say.

Sabine: If you had a physical version of an Atari video game, where you’d have to calculate the optimal paths to get the coins, and then actually run and jump, maybe that would test the physical and mental aspects?

Tom: The engineering of the environment itself would be quite an accomplishment.

Sabine: We’d need to have augmented reality, with virtual coins failing – I don’t think we could actually be throwing stuff at the robots.

Michael Littman: I was watching the new speed climbing event, and they actually talk about the different parts of the course as “problems.” The climbers have to solve these problems, and also be able to hang upside down like a spider. So, this event really does mix the physical and mental. I agree that those kinds of events are the most interesting in the Olympics. That’s why I think RoboCup is so cool. It’s not just purely about being able to kick the ball harder than the other team. You have to actually react in the field.

In many events at the Olympics, a lot of the preparation is done by a team beforehand, and then it’s a case of executing it accurately. For things like shot put, we already have machines called cannons that are way better than we are. So, it really is where you have to think on-the-fly where it gets interesting.

Sabine: What about the use of AI to help humans (rather than AIs competing)?

Tom: There’s certainly a big industry in helping people with their golf swing, by taking video and analysing it. I imagine there must be people researching gait analysis for running.

Sabine: It would be interesting to see how they could specialise to the individual. The things that make these athletes fast must be very specialised to that runner. Maybe in terms of search optimization AI could suggest different options.

Michael: I know that in basketball there’s a lot of sports analytics to try to make the teams more effective. Michael Bowling, at the University of Alberta, has spent a lot of time building AI support for curling, a winter Olympic sport.

Oskar: I would add to what Michael said about the new Olympic climbing event. In preparation for that there has been a lot of data collection, databases of different paths, athletes wearing sensors, and these even measured grip at the contact areas. There is a lot of sensor-based analysis and interpretation to optimise.

Sabine: Should this be allowed, and is it fair across different teams, what AI they have access to, to enhance their performance?

Sanmay: I would say 100%. I grew up in India, and coming to the US the difference in facilities was huge. Coming to college in the US you suddenly have access to all these world class facilities whereas the best kid athletes in India growing up wouldn’t dream of having access to those. I actually think that the use of AI would probably go towards reducing inequities because it’s so much cheaper than the actual physical facilities. Obviously I’m not saying this is the biggest problem facing the world, or that this is the right thing to be spending time and resources on, but, compared to the other inequities in athletic preparation that exist around the world, I think this would level the playing field more than anything else.

Sabine: Of all the disciplines in the Olympics, which one do you think it would be fun to watch an AI doing?

Sanmay: I’m going to pick one that’s not in the Olympics, but it should be: squash. I would love, one day, to see an AI-robotic squash player. I think that’s probably the hardest one to get right, because it’s a combination of intense physical skills of all different kinds, and it’s a very tough game mentally speaking.

Michael: I’ll put in a vote for whitewater canoeing. It’s just such a dynamic environment. All of the team sports already have this property that they are fun to watch because of the unpredictability, but, of the individual sports, I really like the whitewater because it introduces so much chaos. The athletes have to be constantly reacting and innovating on-the-fly to be able to get through all the gates.

Sabine: I would pick skateboarding. If only because there could be some great blooper videos of robots flying in the air and tumbling.

Oskar: For the individual sports, I think that gymnastics is also very interesting. I find it amazing what humans are able to do.


AIhub Editor is dedicated to free high-quality information about AI.
AIhub Editor is dedicated to free high-quality information about AI.

            AIhub is supported by:

Related posts :

The Berkeley Crossword Solver

We recently built the Berkeley Crossword Solver (BCS), the first computer program to beat every human competitor in the world’s top crossword tournament.
27 June 2022, by

Deep attentive variational inference

The expressivity of current deep probabilistic models can be improved by selectively prioritizing statistical dependencies between latent variables that are potentially distant from each other.
24 June 2022, by

AI regulations are a global necessity, panelists say

150 attendees come together to discuss intellectual property, disability rights, and global regulation benchmarks and more.
22 June 2022, by

AI and machine learning are improving weather forecasts, but they won’t replace human experts

Machine learning can help with some of the challenges faced by weather forecasters.
20 June 2022, by

Rethinking human-in-the-loop for artificial augmented intelligence

How do we build and evaluate an AI system for real-world applications?
17 June 2022, by

2022 AI Song Contest entries now live

Listen to, and vote on, the entries in this year's AI song contest.
16 June 2022, by

©2021 - Association for the Understanding of Artificial Intelligence


©2021 - ROBOTS Association