A Survey on Safe Model Predictive Control and Reinforcement Learning

Recent progress in model predictive control (MPC) and reinforcement learning (RL) has led to a growing interest in many complex decision-making tasks. However, MPC heavily depends on the model and RL often disregards safety concerns, which prevents them from being applied to real-world applications. This survey presents the trends in MPC and RL from the perspective of safety concerns, based on different safety levels associated with MPC and different levels of interaction with the environment in RL. Also, integrated approaches that take advantage of the strengths of MPC and RL are explored. Finally, this paper outlines future directions for safe decision making based on MPC and RL