A Learning Mobile Robot: Theory, Simulation and Practice

摘要

. This paper presents an implementation of the sins multi-strategy learning controller for mobile robot navigation. This controller uses low-level reactive control that is modulated on-line by a learning system based on case-based reasoning and reinforcement learning. The case-based reasoning part captures regularities in the environment. The reinforcement learning part gradually improves the acquired knowledge. Evaluation of the controller is presented in a real and in a simulated mobile robot. 1 Introduction How to specify behaviour in a robot has come a long way since the low-level languages of assembly robotics (Lozano-Perez, 1982). The classical AI approach to control 1 proved too slow and too fragile for the real world but showed that representations of the environment, however difficult to maintain, produce interesting behaviour. In nouvelle AI, e.g. (Brooks, 1985; Brooks, 1991a; Brooks, 1991b), agents merely react to the current environmental situation posed, limited ...