Real-time Motion Planning Based on MPC With Obstacle Constraint Convexification For Autonomous Ground Vehicles

This paper proposes a real-time local motion planning algorithm for autonomous vehicles, which is based on model predictive control (MPC) subject to static and dynamic obstacles. In the proposed planner, the constraints of the road boundaries and obstacles are approximated by low-dimensional linear convex hulls. The proposed MPC optimizes the planning and control goals within one optimization problem using a dynamical model. The first element of the computed optimal decision sequence at any time instant can be applied directly to the vehicle, and an additional tracking controller is not required. The performance of the proposed approach is evaluated in both static and dynamic obstacle scenarios within the high-fidelity CarSim simulation environment. The results show that the proposed algorithm can compute the control actions in real-time, and the vehicle can drive safely and smoothly amid static and dynamic obstacles. Moreover, the proposed approach outperforms the potential field method in terms of computational complexity and planning cost, and the MPC-based planner with kinematics in the high driving speed cases.