Evolutionary Algorithm-based Attack Strategy with Swarm Robots in Denied Environments

Huan Liu,JunQi Zhang,Peng Zu,MengChu Zhou
DOI: https://doi.org/10.1109/tevc.2022.3194349
IF: 16.497
2022-01-01
IEEE Transactions on Evolutionary Computation
Abstract:An attack-defense confrontation problem arises from a robot swarm attacking a territory protected by another one. In denied environments, global positioning and communication are hardly available. It becomes difficult for a swarm to realize collaboration and handle confrontation against another. Commonly-used deep reinforcement learning relies on pre-training, which is time-consuming and has strong environmental dependence, especially in denied environments. To study attack strategies in denied environments, this work proposes a novel Evolutionary Algorithm-based Attack Strategy with Swarm Robots for the first time. Each robot obtains its situation information through perceiving its nearby peers and enemies. Such information is utilized to evaluate the benefits or threats of a robot’s next perceptible attack positions. Then, each robot uses the evolutionary algorithm to optimize its fitness function and searches for its optimal position. A collision avoidance strategy is integrated into the algorithm. Hence, a robot swarm realizes collaboration and handles confrontation as long as each robot can sense its surroundings. They utilize their own sensors to detect others locally without using global positioning and communication devices. The experimental analyses show that the evolutionary algorithm-based attack strategy has better scalability and more potential in solving large-scale confrontational problems than the deep reinforcement learning-based algorithms. Rationales of the proposed method is presented to verify the great capability of the proposed method.
computer science, artificial intelligence, theory & methods
What problem does this paper attempt to address?