Task-Level Robot Learning
MASSACHUSETTS INST OF TECH CAMBRIDGE ARTIFICIAL INTELLIGENCE LAB
Pagination or Media Count:
We are investigating how to program robots so that they learn from experience. Our goal is to develop principled methods of learning that can improve a robots performance of a wide range of dynamic tasks. Our interest is in complex tasks such as throwing, catching, batting, yo-yoing, and juggling. We have developed one method of learning, task-level learning, that successfully improves a robots performance of both a ball-throwing and a juggling task. With task-level learning, a robot practices a task, monitors its own performance, and uses that experience to adjust its task-level commands. For example, we have programmed a robot to juggle a single ball in three dimensions. The robot practices the juggling task by batting a ball into the air with a large paddle. The robot uses a real-time binary vision system to track the ball and measure its own performance. Task-level learning consists of building a model of the performance errors at the task level during practice. The robot compensates for the performance errors by using that model to refine the task-level commands. When using task-level learning, the number of hits that the robot can execute before the ball is hit out of range dramatically improves.
- Computer Programming and Software