CB-CEA: A Clustering-Based Co-evolutionary Algorithm for Generating Environment Patterns Effectively

摘要

This paper studies evolutionary learning of neural network navigators for mobile robots. To evolve navigators that generalize well, we should evaluate the navigators using as many environment patterns as possible during evolution. On the other hand, to evolve the navigators efficiently, we should use as few environment patterns as possible. It is difficult to know in advance what patterns can produce good navigators. To solve this problem, we have proposed a co-evolutionary algorithm that can evolve the navigators and the environment patterns together. However, results obtained so far are not good enough. This paper proposes a clustering-based co-evolutionary algorithm (CB-CEA) for generating environment patterns more effectively. The basic idea is to cluster the environment patterns, select a representative from each cluster, and use some of the best representatives for evaluating the navigators. The effectiveness of the proposed algorithm is verified through simulations.