Title:MPPI-IPDDP: Hybrid Method of Collision-Free Smooth Trajectory Generation for Autonomous Robots

Abstract:This study presents a hybrid trajectory optimization method that generates a collision-free smooth trajectory for autonomous mobile robots. The hybrid method combines sampling-based model predictive path integral (MPPI) control and gradient-based interior-point differential dynamic programming (IPDDP) exploiting their advantages of exploration and smoothing. The proposed method, called MPPI-IPDDP, consists of three steps. The first step generates a coarse trajectory by MPPI control, the second step constructs a collision-free convex corridor, and the third step smooths the coarse trajectory by IPDDP using the collision-free convex corridor computed in the second step. For demonstration, the proposed algorithm was applied to trajectory optimization for differential-driving wheeled mobile robots and point-mass quadrotors. A supplementary video of the simulations can be found at this https URL.