A Practical Sampling-based Motion Planning Method for Autonomous Driving in Unstructured Environments

Abstract

Motion planning is a fundamental technique for autonomous driving that guides the ego vehicle to the destination. Demands of driving safety and real-time calculation make it tend to adopt a layered strategy in the motion planning. This paper proposes an efficient method for the reference trajectory generation at a global level in unstructured driving environments. First, sampling-based path construction and optimization are applied to incrementally search for a feasible spatial reference path. Then, a trapezoidal speed profile, further smoothed by a cubic B-spline, is carefully designed along the path considering speed and acceleration limits. Finally, a simplified tracking method is implemented along the speed-path-coupled reference to obtain a feasible control sequence and further smooth the trajectory. Simulation results illustrate the effectiveness of the proposed method that it can efficiently obtain a collision-free trajectory, including waypoints information as well as speed and steering inputs profiles.