Data-Driven Inverse Cooperative Game Control via Off-Policy Q-Learning

In this article, the data-driven inverse cooperative differential game (ICDG) control problem is investigated. First, an excitation signal is selected to fully excite the system, and the system state and control input data is collected. Accordingly, the optimality condition of the cooperative differential game in the sense of Q-function is developed and the off-policy Q-learning technique is used to formulate the ICDG control as a problem of solving an algebraic equation. Second, the least-squares solution to the algebraic equation can be obtained provided that a rank condition is satisfied. Finally, a simulation example is provided, in which the cooperative driving behavior of two drivers is identified by using the proposed ICDG algorithm.