Trajectory planning and control of an autonomous race vehicle

Abstract: There is no doubt that autonomous vehicles are developing in a fast pace in several sectors and markets. For any fast-driving autonomous vehicle the capability of operating close to the limits of traction is essential since it increases manoeuvring capabilities. Increased manoeuvring capabilities allow for increased vehicle handling at higher velocities and in turn improved vehicle safety. Autonomous racing is a field where vehicles are forced to drive close to friction limits to reduce completion times. The same techniques used on a racetrack can further be integrated into other areas such as safety systems.All autonomous vehicles require perception, localization, mapping, trajectory planning and control. Each part is essential, although when considering autonomous racing planning and control are particularly important. Therefore, this thesis investigates two different trajectory planners and their racing capabilities, sensitivity against mass alteration and obstacle avoidance capabilities. The first trajectory planner is called the Feedforward-feedback (FF-FB) trajectory planner. It is built up with a three-step process consisting of path planning, velocity planning and control one at a time. The second approach is called the 2-layer Model predictive control (MPC). It is built up of a two-step process consisting of an optimization problem solving path planning and velocity planning as one and thereafter, control. The trajectory planners are built, integrated, and evaluated in a Robot Operating System (ROS) based simulation environment called Formula Student Simulator (FFSIM).Both trajectory planners proves that it is possible to handle a vehicle close to the limits of traction in a satisfactory manner. Their performance differentiates depending on the subject area. The FF-FB trajectory planner proved to achieve lower completion times on a racetrack considering known circumstances. On the other hand, the 2-layer MPC proved to achieve decent completion times on a racetrack without the same requirement on known circumstances. When an obstacle was introduced to a track the 2-layer MPC was able to detect at a significantly shorter detection distance compared to the FF-FB trajectory planner. Additionally, a mass alteration test shown that the 2-layer MPC is more sensitive to reduced mass in comparison to the FF-FB while the FF-FB trajectory planner is more sensitive to increased mass compared to the MPC.Since the performance differentiated, further development within a specific area could favour one over the other. If the only objective is a short completion time in a race for autonomous vehicles, the FF-FB could be advantageous given known circumstances, while considering use within a safety system, the MPC would probably be preferable.