Trajectory Planning for Autonomous Driving Based on Spatio-Temporal Corridor

Trajectory planning for autonomous driving is a typical three-dimensional constrained planning problem that takes into account spatio-temporal information. In this work, a trajectory planning method is proposed to solve this highly nonlinear and non-convex problem. First, at all timestamps where the obstacles appear, we decompose the two-dimensional planes consisting of the obstacles and the configuration space, and merge these decomposed planes to construct a series of trapezoidal bodies. Then a feasible spatio-temporal corridor is obtained by filtering the resulting trapezoidal bodies. Finally, the spatiotemporal corridor is projected longitudinally and laterally, then optimized respectively. When performing the longitudinal and lateral optimization, the optimization problem is transformed into a quadratic programming (QP) problem to be sovled. Our work uses the prediction information of obstacles to quickly construct a feasible spatio-temporal corridor. Then we transforms the three-dimensional problem into two two-dimensional QP problems to solve, which can greatly reduce the complexity of the problem and satisfy the conditions of real-time and safety while ensuring global optimality.