Path planning and trajectory generation - model predictive control

Abstract: Autonomous vehicle technology is a rapidly expanding field that will play an important role in society in the future. The ambition of autonomous technology is to improve safety for drivers, passengers and pedestrians, reduce traffic congestion and lower fuel consumption. To reach these goals advanced driver-assistance systems and autonomous driving aid the driver or takes full control over the vehicle. With the help of computers, a more optimal driving style is implemented. This thesis focuses on implementing lateral and longitudinal control for an autonomous vehicle with the help of model predictive control using artificial potential fields. The model predictive controller foresees the future for a finite time horizon using a mathematical model of the vehicle. Here a linearized and discretized kinematic bicycle model is used as the internal vehicle model in the controller. The model predictive controller avoids obstacles with artificial potential fields that are quadratically approximated using Taylor series with the help of Bézier curves. The control inputs are the vehicle’s side slip angle and the longitudinal acceleration. The simulations take place on roads in the United Kingdom with left-hand traffic in urban environments. The simulated use-cases are limited to roadside parking scenarios with and without traffic. The goal is to follow the lane center while navigating around obstacles in a predictable way, respecting traffic laws and avoiding collisions or hazardous situations. A simplistic decision-making module is used to determine the vehicle’s next course of action. After the subsequent maneuver is decided a velocity profile and a lane center reference are created for the ego vehicle to follow. The model predictive controller solves the optimization problem as a quadratic programming problem that minimizes a cost function while satisfying a set of constraints. The cost function minimizes the error for the velocity, the lateral lane position, the control inputs, the control input’s rate of change and the yaw angle. The artificial potential fields are also included in the cost function to guide the ego vehicle away from high cost regions. The algorithms were built and simulated using MATLAB. The quadratic programming problem was solved using MATLAB’s quadprog routine in the optimization toolbox, with a sample time of 0.1 seconds. The MPC developed good results with quick calculation times. The highest average calculation time was 0.0159 s with a maximum calculation time of 0.0549 s. The vehicle could perform overtaking maneuvers of two parked cars at 6 m/s in 8.4 s and in 17.2 s when overtaking from a standstill.