Fast Bidirectional Motion Planning for Self-Driving General N-Trailers Vehicle Maneuvering in Narrow Space

Abstract

Self-driving General N-trailers (GNT) vehicles are one of the future solutions to build intelligent factory due to its flexibility and large load. Maneuvering of GNT vehicle to its destination requires accurate and robust motion planning. But the narrow operating environment causes nonlinear nonconvex constraints which are challenging. Furthermore, the nonholonomic constraints in GNT kinematics elevate the complexity in state space. Therefore, motion planning of GNT vehicle maneuvering in narrow space within a reasonable time and high success rate is a critical problem. This paper proposes a fast bidirectional motion planning algorithm to generate trajectories for GNT vehicles to maneuver in a narrow space. A coarse-to-fine motion planning paradigm has been proposed to balance the robustness and time. In the coarse step, an initial guess is generated through a bidirectional-sampled closed-loop Rapidly-exploring Random Tree, and a spatial-temporal safety corridor has been constructed to convert nonlinear nonconvex constraints to linear convex constraints. In the fine step, an optimal control problem is defined accordingly and solved to obtain feasible trajectory. Four different scenarios have been conducted with forward and reverse GNT vehicle maneuvering in a narrow environment. The results show that the proposed method outperforms state-of-the-art sampling-based and optimization-based motion planning methods.