Inverse Model Predictive Control Based on Trust-Region for Trajectory Imitation

Model predictive control (MPC) has been widely adopted in practical applications compared to finite-horizon optimal control due to its superior performance. While significant progress has been made in finite-horizon inverse optimal control, these approaches prove inadequate for addressing inverse MPC (IMPC) challenges. This paper presents a novel trust-region based IMPC algorithm that effectively minimizes the deviation between the generated real and demonstrated trajectories, achieving accurate trajectory imitation. The trustregion IMPC algorithm employs MPC's receding horizon properties to compute the key gradient and hessian terms of the real trajectory with respect to cost parameters. The algorithm is evaluated through a simulation experiment, demonstrating its effectiveness in trajectory imitation.