Fast Trajectory Planning in Cartesian rather than Frenet Frame: A Precise Solution for Autonomous Driving in Complex Urban Scenarios

Abstract

On-road trajectory planning is a direct reflection of an autonomous vehicle’s intelligence level when traveling on an urban road. The prevalent on-road trajectory planners include the spline-based, sample-and-search-based, and optimal-control-based methods. Path-velocity decomposition and Frenet frame have been widely adopted in the aforementioned methods, which, nonetheless, largely degrade the trajectory planning quality when the road curvature is large and/or the scenario is complex. This paper aims to plan precise and high-quality on-road trajectories, thus we choose to describe the concerned scheme as an optimal control problem, wherein the urban road scenario is described completely in the Cartesian frame rather than in the Frenet frame. The formulated optimal control problem should be numerically solved in real-time. To that end, a light-weighted iterative computation architecture is built. In each iteration, a tunnel construction strategy tractablely models the collision-avoidance constraints, and a constraint softening strategy helps to find an intermediate trajectory for constructing the tunnels in the next iteration. Efficacy of the proposed on-road trajectory planner is validated by simulations on a high-curvature urban road wherein the ego vehicle is surrounded by multiple social vehicles at various speeds.