Learning MPC for the Berkeley Autonomous Racing Car (BARC)
The Learning Model Predictive Control (LMPC) technique is applied to the autonomous racing problem. The goal of the controller is to minimize the time to complete a lap. The LMPC strategy uses the data from previous laps to learn from experience, improving its performance while satisfying safety requirements. Moreover, a system identification technique is proposed to estimate the vehicle dynamics.
In the LMPC framework the data from each lap are used to build a control invariant set and an approximation of the value function. These quantities are used to guarantee safety and performance improvement between to tasks repetition. Finally, the LMPC framework is extended to handle repetitive task, as the one represented by a vehicle driving continuously on a race track.
MS Thesis of Maximilian Brunner