“It is well known that certain bodies accelerate learning,” Bongarde said. “This work shows that artificial intelligence can search for such bodies.” Bongard’s laboratory has developed robotic bodies suitable for specific tasks, such as applying a corpus callus-like coating to the feet to reduce wear. Bongarde said that Gupta and his colleagues expanded on this idea. “They show that the right body can also accelerate changes in the robot’s brain.”
In the end, Gupta said, this technology can reverse our view of building physical robots. Instead of starting with a fixed body configuration and then training the robot to complete a specific task, you can also use DERL to evolve the best body plan for that task and then build it.
Gupta’s unimals are part of a broad shift in how researchers view AI. Researchers began to put robots in virtual sandboxes, such as POET, OpenAI’s virtual hide and seek arena, and DeepMind’s virtual playground XLand, instead of training AI on specific tasks, such as playing Go or analyzing medical scans to learn how to constantly change Solve multiple tasks in the open training dojo. AI trained in this way is not to master a single challenge, but to learn general skills.
For Gupta, free exploration will be the key to the next generation of AI. “We need a truly open environment to create smart agents,” he said.
go to see more here in tech news