Human children generally learn to tie their shoes by age 5 or 6. Robots, on the other hand, have been working on the problem for decades. In a new paper, Google DeepMind researchers showcase a method for teaching robots to perform a range of dexterous tasks, including tying a shoe, hanging a shirt, and even fix fellow robots.
DeepMind employed a new learning platform, ALOHA Unleashed, paired with its simulation program, DemoStart, to teach robots by watching humans. The research primarily showcases how system can learn to perform complex, dexterous tasks from a visual demonstration. It’s easy to see, however, how this could be put into practice, particularly in assisting those with accessibility issues.