GOAL SEEKING OF MOBILE ROBOTS USING DYNAMIC FUZZY Q-LEARNING

摘要

In this paper, goal seeking of mobile robots using the Dynamic Fuzzy Q-Learning method is presented. The proposed method allows a mobile robot to avoid obstacles and reach the goal efficiently. The salient characteristics of the proposed methods are: 1) Selforganizing fuzzy inference to calculate continuous valued actions and Q-functions; 2) Automatic generation of fuzzy rules and online recruitment of appropriate rules; 3) Eligibility trace method to learn faster and to alleviate the experimentation-sensitive problem caused by an arbitrarily selected bad training policy; 4) A better alternative to random search for initial state-space exploration especially when reward-giving states are sparse. Experimental results demonstrate that the robot is able to learn a near optimal policy within a few trials.