Learning MPC:                                                                                   A control framework to teach safety critical tasks

Learning MPC: A control framework to teach safety critical tasks

Learning Model Predictive Control (LMPC) is a technique we develop at the MPC lab to control system performing iterative tasks. The controller is reference-free and is able to improve its performance by learning from previous iterations. In LMPC, the data collected from each task execution are exploited to guaranty safety and to improve the system performance. A safe set and a terminal cost function build from data are used to guarantee recursive feasibility and non-decreasing performance for the closed-loop system at each task execution.

More details on the LMPC design and its properties can be found here.

The LMPC is suitable to teach autonomous vehicles to race as it guaranties safety and performance improvement i.e. decreasing lap time. Here, the experiments we did on our Berkeley Autonous Race Car (BARC) open source platform.

More details on the LMPC design for autonous racing can be found here.

Learning MPC for the Berkeley Autonomous Racing Car (BARC)

Learning MPC for the Berkeley Autonomous Racing Car (BARC)

Research on the Berkeley Autonomous Car Recognized by the US Secretary of Transportation

Research on the Berkeley Autonomous Car Recognized by the US Secretary of Transportation