Convex Approximation Based Avoidance Theory and Path Planning MPC for Driver-less Vehicles

Abstract

An improved path planning algorithm for the driver-less vehicle is proposed in this paper, and a soft-ware in loop system is set up to evaluate its performance under complex dynamic traffic scenarios. First, a convex approximation based avoidance theory is introduced, and a method to optimize the obstacle's reference point is proposed for enlarging approachable region. Based on the proposed algorithm, the theory of MPC (Model predictive control) and the curvilinear coordination system, and nine key weighting factors are considered thoroughly to achieve an optimal path, including the dimensions of ego and obstacle vehicles, path geometric constraints and ego vehicle's mechanical constraints, shortest path, lateral acceleration, path alignment, lane changing successively, vehicle to vehicle safety distance, left lane changing priority and the rate of front wheel angle change. Finally, the GWM H7 SUV is used as the driver-less prototype vehicle, and a Carsim + Simulink based soft-ware in loop system is set up,via using the dSPACE multi-cores platform, in order to test the proposed algorithm. The simulation test results demonstrate that not only a reasonable and smooth path is achieved to avoid the disturbances from the moving vehicles, but also an expected real-time performance is obtained.