An Autonomous System for Extracting Fuzzy Behavioral Rules in Mobile Robotics

摘要

We describe an approach to automatic generation of local-cued reactive behavioral rules from the common features contained in a number of world models in the form of geometric spatial maps. It represents one stage of an ongoing effort to combine the best features of these two techniques in a complementary way. First we compare world modelling (or map-building) and reactive behavior as techniques for mobile robot controller design, identifying some of their major strengths and weaknesses, and briefly reviewing some examples of similar work in the context of our overall objectives. We then cover the following aspects of our overall architecture; the nature of the maps from which maze trajectories are derived, gathering sensorimotor data from an established trajectory, fuzzy clustering of the data to form rules, two variants of a reactive fuzzy-controller, and some experiments conducted in simulation to establish the viability of the approach. To date we are optimistic about the potential for these techniques in enabling an agent to build reactive behaviors autonomously. However there is much more work to be done, some of which we identify at the end of this paper.