Enhanced framework for fast kinodynamic planning and control on large curvature roads for autonomous vehicles

Abstract

Large curvature roads represent a common scenario in urban autonomous driving, posing significant challenges for planning and control which include: (1) The method based on the Frenet Coordinate Frame (FCF) tends to be unstable on roads with large curvatures, leading to issues such as obstacle deformation and incorrect heading angles; (2) The nonlinear kinematics of the vehicle should be taken into account when utilizing the Cartesian Coordinate Frame (CCF) as a substitute for the FCF; (3) Accounting for the transient entry of tire lateral forces into the nonlinear region during obstacle avoidance maneuvers on large curvature roads, necessitating the consideration of nonlinear vehicle dynamics. To ensure the safe maneuvering of autonomous vehicles on large curvature roads, we have devised a suite of efficient planning and control framework. This framework incorporate the nonlinear kinematics and dynamics (Kinodynamic) of vehicle within CCF, enabling swift maneuvers and autonomous obstacle avoidance on such roads. We ensure solution stability through a combination of low-dimensional global search and high-dimensional local optimization, while preserving the sparsity of the jacobian matrix during the optimization process for vehicle kinodynamic. By parameterizing all state and control trajectories using pseudospectral orthogonal collocation and converting obstacle avoidance constraints into convex corridor constraints, our approach significantly reduces computation time by up to 99.81%. Finally, employing the high-fidelity Carsim model simulation and real vehicle experiments, we can verify that this kinodynamic framework effectively handles challenging operational scenarios, including medium to high speeds, large curvature (nonlinear vehicle kinodynamic), and sharp turns (curvature discontinuity).