Development of a Model-based Control Strategy for Autonomous Vehicle Collision Avoidance

Abstract

Human inattention is the leading cause of traffic accidents in many regions around the world. Autonomous vehicle technologies are rapidly emerging with the aim to remove the human factor in key driving procedures, such as perception, decision-making, path planning, and control. These technologies are subject to technological, ethical, and social scrutiny; therefore, extensive work is required to instill confidence in the reliability of these automated driving features. One key responsibility of automated driving is in planning and tracking a trajectory to avoid collisions with obstacles, such as other vehicles. One of the foremost challenges in the formulation of a feasible path is considering the dynamics and constraints of the vehicle and the environment.

Model predictive control (MPC) is one of the most common control techniques for its ability to handle constraints. For this reason, MPC has been widely studied for path planning and tracking for autonomous vehicles and mobile robots. MPC relies upon an accurate vehicle dynamics model which enables accurate state predictions, thereby resulting in effective control actions to achieve the desired objective. It is challenging, however, to capture all of the details and uncertainties of the dynamics associated with a vehicle. In particular, modeling tire dynamics requires detailed nonlinear models to fully reflect the vehicle behavior. One common technique for motion planning using MPC is to employ artificial potential fields (PFs) which generate an artificial repulsive force from obstacles or road boundaries to influence the controller to track the vehicle along a safe trajectory. Some state-of-the-art PF-based techniques include the PF intensity directly in the MPC objective function, thereby considering the vehicle constraints and dynamics as part of the path planning.

In this thesis, an enhanced PF-based motion controller is presented. The control design uses MPC with a detailed dynamics model; the model considers the combined-slip effect on tire forces, nonlinearities, and actuator dynamics. Therefore, it offers an improvement upon prior studies which rely upon simplified dynamics models. Moreover, the PF intensity is included in the objective function, like prior studies, although the PF approximation is further simplified by only considering the lateral component of the repulsive force as part of the latera controller. A separate, novel longitudinal control policy uses the longitudinal component of the PF gradient to regulate the speed setpoint when approaching an obstacle in the same lane; subsequently, proportional-integral-derivative (PID) controllers command axle torque and brake pressure to track the reference speed. The developed controller and dynamics model are validated in both simulation and physical vehicle tests. To emulate the various driving scenarios where avoidance or stopping is required, a virtual driving environment is employed: simulated obstacles are placed in the roadway, the detections of which are sent to the controller. The controller performance is demonstrated in various evasive maneuvers, and in different road conditions.