Reinforcement learning based fission-fusion for heterogeneous UAV swarm under dynamic interference environments

Abstract

The control of unmanned aerial vehicle (UAV) swarms represents a complex field of study, chiefly due to the conflicting behaviors observed among individual UAVs and the impact of external movement disturbances on the swarm.However, the fission-fusion dynamics of UAV swarms in response to unknown dynamic disturbances have received comparatively less attention than their behavior in static flight. An unknown dynamic interference environments fission-fusion for heterogeneous UAV swarm via reinforcement (DEFHRL) learning algorithm is presented, which effectively addresses the challenge of fission-fusion control within unknown dynamic interference environments. Firstly, we develop a heterogeneous swarm self-organized fission-fusion control framework that enables multi-swarm fission-fusion maneuvers for UAV swarms. Next, we introduce a topological fission selection algorithm that facilitates control of fission selection under minimal interaction loads, effectively enabling controllable swarm sizes. Finally, we introduce a subgroup adversarial algorithm, grounded in reinforcement learning, designed to conduct dynamic adversarial engagements against unknown interference with minimal resource expenditure. Simulation experiments show that UAV swarms, when operating in dynamic heterogeneous environments, can successfully execute self-organized fission-fusion maneuvers, effectively safeguarding the primary swarm against heterogeneous interference.