Learning Mobile Robot Behaviours by Discovering Associations Between Input Vectors and Trajectory Velocities

摘要

In this paper, we describe a reinforcement robot learning method which enables a mobile robot to simultaneously acquire the ability to avoid objects, follow walls and control its velocity as a result of interacting with its environment. Our approach differs to conventional reinforcement learning approaches in that the robot learns associations between input vectors and trajectory velocities rather than learning one to one associations between input vectors and command responses. Thus, by interacting with the environment the robot becomes capable of predicting how fast it should travel along the various trajectories available to it at any instant and can exhibit effective object avoidance or wall following behaviour by simply being given an instruction to follow fastest arbitrary trajectory or follow fastest trajectory nearest to closest object respectively. We provide results demonstrating the effectiveness the our learning method by using a Yamabico mobile robot to acquire wall following and object avoidance behaviours. 1.