Implementation and Analysis of a Clothoid-based Model Predictive Controller

Sammanfattning: For the last couple of years autonomous driving has increased in popularity as a research area, and it continues to grow. A topic within autonomous driving is path following, which is the subject studied in this project. One of the popular controllers to use for controlling a vehicle is the model predictive controller, because it finds an optimal control input for the vehicle based on the model of the vehicle, and its estimated future behaviour within the prediction horizon - which covers a distance ahead of the vehicle. To increase the length of this distance, one can use a new controller - the clothoid-based model predictive controller. The clothoid-based model predictive controller is a linear time-varying model predictive controller that uses a clothoid-based vehicle model to find an optimal input based on the vehicle’s behaviour at the kink-points. The kink-points are way-points that are used to create the clothoids, and the distance between them can be very far. Therefore, it is possible to cover a large distance ahead of the vehicle with a small prediction horizon. In this thesis, the controller is implemented at the Smart Mobility Laboratory at KTH Royal Institute of Technology so that it can be tested and evaluated for future use. The controller is implemented on a 1 : 32 scaled radio truck that is monitored by a motion capture system, and remotely controlled by a desktop computer. The outcome of the implementation is a new controller for the remote controlled radio trucks with a fast control algorithm, where the greatest mean deviation from the path was 0.117m.