Memory-Based Techniques for Task-Level Learning in Robots and Smart Machines

Abstract

We report on a preliminary investigation of tasklevel learning, an approach to learning from practice. We have programmed a robot to juggle a single ball in three dimensions by batting it up-wards with a large paddle. The robot uses a real-time binary vision system to track the ball and measure its performance. Task-level learning consists of building a model of performance errors at the task level during practice, and using that model to refine task-level commands. A polynomial surface was fit to the errors in where the ball went after each hit, and this task model is used to refine how the ball is hit. This application of task-level learning dramatically increased the number of consecutive hits the robot could execute before the ball was hit out of range of the paddle. The talk explores memory-based techniques for future implementations of tasklevel learning.