Helping Robots Learn By Letting Them Fail

The [MIT Technology Review] has just released its annual list of the top innovators under the age of 35, and there are some interesting people on this list of the annoyingly accomplished at a young age. Like [Lerrel Pinto], an associate professor of computer science at NY University. His work focuses on teaching robots how to do things in the home by failing.

To do this, he set up an example home in his lab, amusingly filled with kids’ toy versions of appliances like ovens, and then asked a robot to perform tasks like warming his lunch. [Lerrel] had given the robot a primer, a machine learning model based on 30-second videos of his students performing some of the tasks using the same tools that the robot would have, such as robot hands, fish slices, etc. Then he left them to it, trying to perform the tasks for 24 hours a day and judging their own performance.

This technique is called reinforcement learning, where the machine learning system tries something, considers how close it got to the example, and then tries something slightly different. He published the results in a paper entitled “Teach a Robot to FISH“, based on the old saying give a man a fish, and you feed him for a day. Teach him to fish and you will feed him for life. This paper is well worth reading, mainly as it includes lots of examples of how the tasks were broken down and trained. There is also an interesting video of a talk that [Lerrel] did for a presentation to the CMU Robotics Initiative.

EDIT 9/20/23: fixed link to the Arvix paper and the YouTube video.

11 thoughts on “Helping Robots Learn By Letting Them Fail

  1. @Richard Baguley said: “Helping Robots Learn By Letting Them Fail”

    That’s what Elon has been doing for years with the never-ending ridiculously over-priced problem called: “Full-Self-Driving” (FSD).

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.