Research on Multi-agent Formation Control Method Based on Curriculum Learning and Improved PD Algorithm

This paper proposes a control framework combining the MADDPG algorithm from multi-agent reinforcement learning and the improved PD algorithm, applied to the problem of formation flight control for aerial vehicles in a two-dimensional space. By using curriculum learning, the target path is segmented to gradually guide the agents from short-range navigation to a long-range flight, enhancing training stability and sample utilization efficiency. Additionally, a multi-dimensional reward function is constructed to improve control performance. In the simulation experiments, the system demonstrates both good formation maintaining and target tracking capabilities, validating the advantages of the proposed method in terms of stability and convergence.