Continuous curvature path planning for semi-autonomous vehicle maneuvers using RRT

This paper proposes a sampling based planning technique for planning maneuvering paths for semi-autonomous vehicles, where the autonomous driving system may be taking over the driver operation. We use Rapidly-exploring Random Tree Star (RRT*) and propose a two-stage sampling strategy and a particular cost function to adjust RRT* to semi-autonomous driving, where, besides the standard goals for autonomous driving such as collision avoidance and lane maintenance, the deviations from the estimated path planned by the driver are accounted for. We also propose an algorithm to remove the redundant waypoints of the path returned by RRT*, and, by applying a smoothing technique, our algorithm returns a G² continuous path that is suitable for semi-autonomous vehicles. In order to deal with sudden changes in the environment, we apply a replanning procedure to enable our algorithm to rapidly react to the changes in a real-time manner, without full recomputation of the RRT* solution. Numerical simulations demonstrate the effectiveness of the proposed method.