Adaptive Pure-Pursuit Controller Based on Particle Swarm Optimization (PSO-Pure-Pursuit)

One of the most crucial tasks of autonomous vehicles (AVs) is the Path tracking module. It defines the ability of the vehicle to pursuit the generated free collision path. Furthermore, this task has to be done maintaining the stability of the vehicle's heading to ensure the passengers' comfort during the trip. Geometrical controllers are known and easy to be applied in robotics and autonomous driving. It is considered a type of lateral controller that neglects dynamic forces on the vehicles and assumes that no-slip condition holds at the wheels. A geometrical bicycle model defines the error and generates the optimal steering command that achieves path tracking. The main idea of the original pure pursuit (PP) is that a reference point is selected on the path with a look-ahead distance from the vehicle, and the optimal steering angle can be calculated, as it will be the angle needed to intersect with this point. The point continues moving forward as the vehicle turns to pursue the path to follow the generated curve, reducing the steering angle and slowly bringing the vehicle towards the path. The paper in hand uses particle swarm optimization (PSO) to calculate the optimal look-ahead distance each time step based on a prediction model, which is propagated to get the future states of the vehicle. The proposed method showed an advantage over both the original pure pursuit controller and the Basic Stanley (BS) controller in terms of the RMS of the lateral error. The performance is being measured using several simulations on the Car-sim simulator. A wide range of velocities was used to test the vehicle's behavior while carrying out the double lane change maneuver.