Sign In

Communications of the ACM

ACM TechNews

Crowdsourced Feedback Helps Train Robots

View as: Print Mobile App Share:
In the future, this method could help a robot learn to perform specific tasks in a users home quickly, without the owner needing to show the robot physical examples of each task.

This new approach allows feedback to be gathered asynchronously, so nonexpert users around the world can contribute to teaching the agent.

Credit: Christine Daniloff, MIT/iStock

A reinforcement learning approach developed by researchers at the Massachusetts Institute of Technology (MIT), Harvard University, and the University of Washington trains robots using crowdsourced feedback from nonexpert users.

MIT's Marcel Torne said that with the Human Guided Exploration (HuGE) method, "The reward function guides the agent to what it should explore, instead of telling it exactly what it should do to complete the task."

The researchers divided the process into two parts, using a goal selector algorithm updated continuously with crowdsourced human feedback and another algorithm that enables the artificial intelligence agent to explore in a self-supervised manner guided by the goal selector.

In both simulated and real-world tests, HuGE enabled agents to complete goals more quickly than other methods.

From MIT News
View Full Article


Abstracts Copyright © 2023 SmithBucklin, Washington, D.C., USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account