Learning sensor-based navigation
A mobile robot is presented which uses reinforcement learning to acquire reactive navigation skills. The basic skills needed for reaching a goal whilst avoiding obstacles are encoded as sensation-action associations in a modular neural net. The robot has no a priori knowledge of either the environment or the effect of its actions, and learns on-line using only raw sensory data collected as it moves. The experimental results show that a few trials suffice for the robot to navigate efficiently in a real environment of moderate complexity.
Bibliographic Reference: Article: Making Robots Smart: Behavioral Learning Combines Sensing and Action (1998) pp. 87-109
Record Number: 199810913 / Last updated on: 1998-08-07
Original language: en
Available languages: en