A Decoupled Trajectory Planning Framework Based on the Integration of Lattice Searching and Convex Optimization

This paper presents a decoupled trajectory planning framework based on the integration of lattice searching and convex optimization for autonomous driving in structured environments. For a 3D trajectory planning problem with timestamps information, due to the presence of multiple kinds of constraints, the feasible domain is non-convex, so it is easy to fall into local optimum for trajectory planning. And the solution space of this problem is so enormous that it is difficult to identify an optimal solution in polynomial time. To address this non-convex problem, and to improve the convergence speed of an optimization process, the approach based on lattice searching is adopted in consideration of the ability to discretize driving environments and reduce the solution space. And the resulting path generated by lattice searching typically lies in the neighborhood of the global optimum. But this solution is neither spatiotemporally smooth nor globally optimal, so it is generally called the rough solution. For this reason, a subsequent nonlinear optimization process is introduced to refine the rough trajectory (combined by path and speed). The proposed framework is implemented and evaluated through simulations in various challenging scenarios in this paper. The simulation results verify that the trajectory planner can generate high-quality trajectories, and the execution time is also acceptable.