Research on Cnvergence of Multi-Robots Path Planning Based on Learning Classifier System

Learning classifier systems(LCS) are rule-based inductive learning systems that have been widely used in the field of reinforcement learning over the last few years, but seldom used in the multi-robots domain. In this paper a distributed learning classifier system, which combines reinforcement learning and genetic algorithm to create a set of rules on-line, is used to design optimal paths for multi-robots path planning. Due to premature convergence, local optimal solution, needing a larger storage space and other shortcomings of genetic algorithms，and targeted at the different effects of the static and dynamic environment, the authors design different fitness function in static and dynamic environment. They have derived and proved that the credit allocation algorithm is convergent and provides a theoretical guarantee for the path planning algorithms. Simulation results also show that the genetic algorithm and learning classifier system combination for multi-robots path planning is effective. Premature convergence, local optimal solution, needing a larger storage space and other shortcomings of the genetic algorithm have been significantly improved. The proposed new approach has increased multi-robots' ability to quickly find safe paths. So LCS has a very broad application prospects in the field of robotics and also is the future research directions.