Torch: Strategy evolution in swarm robots using heterogeneous–homogeneous coevolution method

Abstract

Because swarm robots have been applied widely in various fields, the evolution capability of their strategy have become of primary interest; therefore, the evolution method of swarm robots’ strategy has attracted attention in both industry and academia, especially for complex applications owing to their varied task scenes. Large amounts of researches have been conducted to realize strategy evolution in swarm robotics systems. However, there are few studies on the strategy evolution sufficiently examining the simultaneous improvement of evolutionary and strategy performance, which are two key demands of swarm robots. Besides, the strategy that evolved under the global information is difficult to fully adapt to the distributed task scenarios. To address these issues, this study presents a heterogeneous–homogeneous swarm coevolution method known as TORCH to improve the evolution capability of swarm robots. The method uses a swarm coevolution mechanism to accelerate the evolution. For the first time, we employ a behaviour expression tree in TORCH which expands the strategy search space of the evolved strategies. TORCH makes the swarm robots’ strategies evolve under local information conditions; hence, the evolutionary strategies are more adaptable to the distributed task scenarios. Extensive experiments have been conducted to verify the proposed TORCH, including a comparison with three methods based on the homogeneous swarm evolution method and parameter expression. The results demonstrate the superiority of the TORCH in terms of evolutionary efficiency improvement and strategy performance enhancement.