Weights-varying MPC for Autonomous Vehicle Guidance: a Deep Reinforcement Learning Approach

Model Predictive Control (MPC) can achieve excellent results for complex control tasks like path-following of autonomous vehicles. However, its performance depends on the right choice of a cost function for its internal optimization problem. Optimizing the cost function to different objectives is challenging and time-consuming. In this paper, we propose to automatically learn context-dependent optimal weights for the cost function with Deep Reinforcement Learning and to adapt the weights online. We show that our approach outperforms the results of a human expert.     «

Model Predictive Control (MPC) can achieve excellent results for complex control tasks like path-following of autonomous vehicles. However, its performance depends on the right choice of a cost function for its internal optimization problem. Optimizing the cost function to different objectives is challenging and time-consuming. In this paper, we propose to automatically learn context-dependent optimal weights for the cost function with Deep Reinforcement Learning and to adapt the weights online....     »