Researchers at ETH Zurich in Switzerland recently taught the one-armed, four-legged ANYmal search and rescue robot to play badminton against a human.
Fans of the sport may be relieved to learn that ANYmal's skills are unlikely to oust humans from the court any time soon, but training the robot to track and strike the shuttlecock was a complex task and the results are impressive.
“We used methods like reinforcement learning, which is basically where you improve the behaviour of the robot through trial and error,” explains the lead roboticist on the project, Dr Yuntao Ma.
“For badminton, perception is one challenge, agile control is another, and the third is to coordinate these two factors,” he says.
This is far from the first time robots have been trained to play sports. They’ve performed dance numbers and gymnastics routines, run marathons and skied, played ping pong and learned to juggle. There’s even a football RoboCup.
It's not all fun and games
“Sports require skill,” says Dr Raffaello D’Andrea, a professor at ETH Zurich who specialises in robotics and artificial intelligence (AI).
“If you want to create robots that have dexterity or the ability to cope with the physical environment, you can use sports as a proxy for learning those tasks.”
Sports also capture people’s imagination and serve as an engaging way to demonstrate new technological advances.
Watching a robot successfully return a shuttlecock is more exciting and relatable than a mundane task that requires the same type of action.
“If you play badminton, you know how hard it is to move and hit the birdie to someone else. It gives you a closer connection to the complexity of the skill,” says D’Andrea.
These skills can then eventually be put to work in environments that are too ‘dull, dirty, or dangerous’ for humans, such as autonomous drones that have been trained to take warehouse inventories or the robots used in bomb disposal.
Ma believes that the key to developing truly versatile robots lies in finding ways to bring disparate research together.
The recent rapid advances in AI and large language models look set to enable far more collaboration between different robotics communities.
“In traditional robotics, researchers usually focus on one task, like locomotion or navigation, but [large] language models offer ways to merge all these different skills. I think a more generic robot will be the next big thing,” Ma says.
It may be a while, however, before we see robot all-rounders coming through, even though the cost of robots and their constituent parts has decreased dramatically in recent years, making it easier for researchers to build and test out different capabilities.
While this affordability has, in turn, opened up the retail market, robots are still chiefly selling to business customers who know how to program them for specific tasks.
“There’s a big difference between writing a research paper and putting out a video online [versus] doing a pilot with a client and rolling out a solution. There are tonnes of videos on the web, but there are very few rollouts,” says D’Andrea.
Those solutions are far more likely to land in industry than the sports field, but anyone who even occasionally watches a match knows how much other technologies have already become part of the sporting world.
For instance, computer vision technologies like electronic line-calling (ELC) are replacing line judges in games such as tennis and cricket, but it’s debatable whether this improves the spectator experience. For the casual punter, much of the joy of sports is in its human unpredictability.
Read more:
- 5 mind-blowing inventions turning humans into real-life cyborgs
- 10 of the world's worst-ever inventions
- What is copypasta, and can it stop my data being used to train AI?
What would be the point?
If a robot tennis player eventually takes Centre Court at Wimbledon, it might score as many match points as novelty points but, after a while, would anyone want to watch it?
“People are entertained by watching sports because they connect to the athletes. Anything that replaces these athletes with robots is not going to be a success,” says D’Andrea.
The same is likely true for anything where human emotion plays a central part. For more than two decades, the Japanese government has invested billions of yen in developing robots for use in residential care homes for the elderly.
While well-intentioned, uptake has been limited, and the evidence so far suggests the robots create more work for carers than they alleviate.
There’s a difference between pushing the boundaries of technology and finding ways to test the limits of what robots can do, and applying those technologies in the wrong places.
So, for now, sportspeople can rest easy. “I think we have to be careful that we don’t go too far for the sake of efficiency”, says D’Andrea. “I would ask the question: why do we want to replace the people?”
Read more: