Minimum curvature trajectory planning and control for an autonomous race car

This paper shows a software stack capable of planning a minimum curvature trajectory for an autonomous race car on the basis of an occupancy grid map and introduces a controller design that allows to follow the trajectory at the handling limits. The minimum curvature path is generated using a quadratic optimisation problem (QP) formulation. The key contributions of this paper are the extension of the QP for an improved accuracy of the curvature approximation, the introduction of curvature constraints and the iterative invocation of the QP to significantly reduce linearisation errors in corners. On the basis of the resulting raceline, a velocity profile is calculated using a forward-backward-solver that considers the velocity dependent longitudinal and lateral acceleration limits of the car. The advantages and disadvantages of the proposed trajectory planning approach are discussed critically with respect to practical experience from various racetracks. The software stack showed to be robust in a real world environment as it ran successfully on the Roborace DevBot during the Berlin Formula E event in May 2018. The lap time achieved was within a tenth of a second of a human driver and the car reached about 150km/h and 80% of its acceleration limits.     «

This paper shows a software stack capable of planning a minimum curvature trajectory for an autonomous race car on the basis of an occupancy grid map and introduces a controller design that allows to follow the trajectory at the handling limits. The minimum curvature path is generated using a quadratic optimisation problem (QP) formulation. The key contributions of this paper are the extension of the QP for an improved accuracy of the curvature approximation, the introduction of curvature constr...     »