Optimal smooth polynomial lane change generation for autonomous racing

Abstract

Online path re-planning is a crucial task for the control of autonomous vehicles, especially when driving at the handling limits in a dynamic environment. In this article, we propose a fast path planning algorithm meant for trajectory change scenarios, that is tailored to generate smooth reference paths. The proposed solution is based on a convex combination of the current and target paths, and guarantees heading and curvature continuity. Additionally, we complete the algorithm with a real-time compliant nonlinear program to generate nearly-time-optimal maneuvers. The proposed approach is tested on a driving simulator and compared against state-of-the-art solutions, which typically rely on numerical differentiation of path coordinates, often introducing inaccuracy and yielding inefficient or imprecise solutions.