BURST
The battle of man versus machine at the University of B.C. on Wednesday ended in a draw, with humanity both humbled and enthralled by the result.
UBC’s Boye Ahlborn Engineering Physics Project Lab has created an AI-powered robotic air hockey table, which saw its ultimate test Wednesday afternoon when it took on two former world champions: the father and son duo of Tim and Jacob Weissman.
Jacob Weissman won his match 7-5, while Tim — who had predicted he would “destroy” the machine — lost 7-2. He came away impressed by what he had seen.
“It was exhilarating and frustrating at the same time,” said Tim, who flew up from Houston, Tex., with Jacob specifically to cross mallets with the UBC robot.
“It was amazing to me that it was actually blocking every shot that I was shooting. It was like trying to score against a brick wall. The defence is amazing. I feel like I could block it enough. I could defend it enough. But on offence, I just had a hard time scoring. It’s amazing.”
The table has been worked on, in various iterations, by 25 students and seven different teams over the past five years. The AI that powers the robot was self-trained in computer simulations, and had reached the stage of development where it was better than any of the students who had worked on it and needed a real test. None of the UBC players had scored on it in the months leading up to Wednesday’s game.
“It’s incredible to see an actual professional play on it,” said Mauro Ferraz, one of the recently graduated students involved in the table’s creation. “A few people (here) were able to score goals here and there, but overall, it was not competitive enough to actually demonstrate the capabilities of the table. So actually seeing professional players on this — and struggling against it — is incredible and very rewarding.”
Added classmate Ian Hartley: “When we took on this project a couple years ago, I didn’t think that it was going to get to this point at all. My mind is still kind of blown that it actually was able to be competitive against the pro players. It was pretty cool.”
The table will continue to be improved upon, with mechanical refinements and the evolution of its machine learning models, but that will be done by future classes. The data they gathered from Wednesday’s games will eventually be incorporated into that learning algorithm.
Tim Weissman, who had said ahead of the game he was silently rooting for the robot, can see some areas he would improve — from a human standpoint.
“I think they need to have it develop the ability to stop the puck and set up a specific shot on offence,” he said. “To drift the puck in a controlled manner, rather than just waiting and seeing where the puck is drifting and then taking a shot wherever that happens to be.”
The combination of AI and the table’s robotics is a glimpse into what Boye Ahlborn lab director Dylan Gunn sees as the near-future. Most of the big tech companies, like OpenAI, started in robotics but then focused more on Large Language Models (LLMs) like ChatGPT or Claude, he said, as it was too complex to mesh the two technologies.
But now robotics are catching up in the real world, like the recent World Humanoid Robot Games where, among other revelations, Usain Bolt’s 100-metre record was beaten by a robot.
“This is the least performant (AIs) will ever be. They are only going to get better,” said Gunn. “Now the AI methods are powerful enough that they are coming back to the physical world … All of the evolution that’s happened in the past however many years with ChatGPT, I think is now going to come into physical systems, humanoid robots.
“We should expect in the next three-to-five years to start seeing this kind of thing much more out in the open: Very capable robotic systems that are able to work in the real world and hopefully start impacting all kinds of things in a positive way.”