基于模糊强化学习和模型预测控制的追逃博弈

Addressing the strategy formulation and robust control issues in the pursuit-evasion game of agents in 3D space, this paper proposes a hierarchical pursuit-evasion game framework based on fuzzy reinforcement learning and model predictive control(MPC). The proposed framework integrates the Apollonius circle in 3D space with the fuzzy actor-critic learning(FACL) algorithm to obtain the agents' motion information, which is then used as the reference input for the MPC algorithm to design the controller for quadrotor unmanned aerial vehicles. By decoupling the underactuated system model of the quadrotor, altitude, translation, and attitude controllers that consider the integral term of the error system are designed. The reference information provided by the FACL algorithm effectively enhances the control efficiency of the MPC algorithm. Simulation and experimental results demonstrate that the designed hierarchical framework can effectively solve the pursuit-evasion game problem in 3D space.