Reinforcement Learning-Based Strategy for Task Assignment in Multi-Satellite Games

This paper addresses the task assignment problem in multi-satellite pursuit-evasion games, involving model nonlinearities and couplings. Considering environmental changes, a mixed reinforcement learning method that combines off-policy and on-policy schemes is proposed. An off-policy approach is firstly developed to estimate the task execution costs and determine the optimal control strategies. To minimize the total task execution costs and maximize the number of matched evaders, a task assignment method based on a mapping function is introduced. The effectiveness of the proposed task assignment method is demonstrated through simulation results of a multi-satellite system.