Robotics

New method uses crowdsourced feedback to train robots

To teach an AI agent a new task, like how to open a kitchen cabinet, researchers often use reinforcement learning—a trial-and-error process where the agent is rewarded for taking actions that get it closer to the goal.