| Summary: | Modular self-reconfigurable robots (SRRs) have redundant degrees of freedom and various configurations. There are two hard problems imposed by SRR features: locomotion planning and the discovery of multiple locomotion patterns. Most of the current research focuses on solving the first problem, using evolutionary algorithms based on the philosophy of searching-for-the-best. The main problem is that the search can fall into a local optimum in the case of a complex non-linear problem. Another drawback is that the searched result lacks diversity in the behaviour space, which is inappropriate in addressing the problem of discovering multiple locomotion patterns. In this paper, we present a new strategy that evolves an SRR's controller by searching for behavioural diversity. Instead of converging on a single optimal solution, this strategy discovers a vast variety of different ways to realize robot locomotion. Optimal motion is sparse in the behaviour space, and this method can find it as a by-product through a diversity-keeping mechanism. A revised particle swarm optimization (PSO) algorithm, driven by behaviour sparseness, is implemented to evolve locomotion for a variety of configurations whose efficiency and flexibility is validated. The results show that this method can not only obtain an optimized robot controller, but also find various locomotion patterns.
|
|---|
