Evolutionary Design of Controllers for Autonomous Line Tracking Mobile Robots Using Population Based Incremental Learning

This paper concerns the design of sensor topology and reaction controllers for autonomous mobile robots following a line. Artificial evolution [1,2] is used as the design methodology. Here, it is shown PBIL [3] is powerful enough to evolve autonomous creatures exhibiting complex behavior. Designing optimal sensor topology, symmetric and asymmetric control mechanisms, reactive controllers with feedback, and suitable criteria for evaluating evolved robots are among the topics discussed in this paper. Intuitive insight into the nature of the problem proved to be a crucial determinant of the success of evolution. Defining some measure of stability turned out to be very useful. The idea can certainly be extended to other benchmark problems. PBIL is demonstrated to be much more effective than GA in exploiting the salient features in both fitness criteria and controller architectures. All the simulations were carried out by the EVO-ROB system, an open architecture, component based software framework especially designed for ER experiments