Towards a Robot Learning Architecture

Joseph O'Sullivan

I summarize research toward a robot learning architecture intended to enable a mobile robot to learn a wide range of find-and-fetch tasks. In particular, this paper summarizes recent research in the Learning Robots Laboratory at Carnegie Mellon University on aspects of robot learning, and our current work toward integrating and extending this within a single architecture. In previous work we developed systems that learn action models for robot manipulation, learn cost-effective strategies for using sensors to approach and classify objects, learn models of sonar sensors for map building, learn reactive control strategies via reinforcement learning and compilation of action models, and explore effectively. Our current efforts aim to coalesce these disjoint approaches into a single robot learning agent that learns to construct action models in a real-world environment, learns models of visual and sonar sensors for object recognition and learns efficient reactive control strategies via reinforcement learning techniques utilizing these models.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.