| dc.description.abstract | This study proposes a highway driving strategy for autonomous vehicles.
First, a model predictive control (MPC)-based trajectory planner is built based on a kinematic model. A series of candidate strategies are created and form a strategy space. With the model and prediction of surrounding vehicles’ movements, the MPC-based planner, according to the candidate strategies, generates feasible trajectories. Next, a decision-making payoff function is applied to select the best trajectory. The payoff function consists of four terms, including lane-changing incentive, cost of controls, cost of risk, and cost of a late lane-changing decision. This decision-making payoff function will select the best trajectory, but this trajectory only provides longitudinal acceleration information.
To maneuver a vehicle, the controller should involve lateral movement. We proposed a yaw rate profile approach as a strategy space for lateral controls. Given longitudinal acceleration, each yaw rate profile will lead the vehicles to a different lateral position, and the one that drives the vehicle to the center of the target lane is the best yaw rate profile. While the vehicle is changing to the target lane, the best yaw rate profile keeps updating. However, because the method to update does not consider the initial error so it fails in some cases. To cope with this issue, an MPC-based path tracking controller is introduced to minimize the error while making the vehicle operating within certain constraints.
Two simulations are created. The first simulation is to test the decision-making payoff function; with a larger weight designed for lane-changing incentive, the autonomous vehicle is more aggressive and more willing to take risks to achieve the lane with higher average speed. The second simulation is designed to show that with the MPC-based path tracking controller, the autonomous vehicle is able to overcome the problems caused by the errors and successfully changes to the target lane. | en |
