Inverse Strategy Inference Method for Differential Pursuit-Evasion Games with Incomplete Information

This paper addresses the problem of non-cooperative target control strategy inversion in space missions, particularly in the context of differential pursuit-evasion games with incomplete information. The study first establishes a differential game model for pursuit-evasion, followed by the inversion of control strategies using the Unscented Kalman Filter (UKF) and Expectation-Maximization (EM) algorithms. The UKF algorithm is used as a state estimator to handle the dynamic evolution of system states, while the EM algorithm is employed as an optimization method for estimating the control strategies based on observed data. A comparative analysis of the convergence performance of both algorithms is presented, where simulation results show that the EM algorithm achieves a convergence speed approximately 20% faster than the UKF. Additionally, robustness tests demonstrate that while observation errors significantly affect the accuracy of control strategy inversion, the UKF algorithm gradually diverges with the increase in observation errors, whereas the EM algorithm does not exhibit divergence. These results highlight the potential of the EM algorithm for effective control strategy inversion in differential pursuit-evasion games under incomplete information conditions.