Multi-scenario Learning MPC for Automated Driving in Unknown and Changing Environments

System dynamics identification significantly impacts trajectory tracking performance for autonomous driving in a dynamic environment. In this paper, a multi-scenario learning model predictive control (MPC) optimization strategy is proposed to reduce model complexity and improve system generalization and robustness. First, the Gaussian process is simplified to reduce the complexity of the system's residual model while ensuring the optimization problem's convexity. Then, a meta-learning based multi-scenario model is proposed through online adjusting weight factors to identify the dynamic characteristics when the vehicle drives in a new scenario. Finally, the developed learning model is integrated into a stochastic MPC framework for robust optimization by considering environmental changes and parameter uncertainties. Simulation results show the efficient performance of our proposed method in terms of model prediction accuracy and trajectory tracking.