How to Prevent the Robot Uprising - An Introduction to MPC Based Safe Reinforcement Learning

Explore a conference talk that introduces Model Predictive Control (MPC) based safe reinforcement learning frameworks for safety-critical systems. Learn how to implement robust safety systems that provide mathematical guarantees during the training process of autonomous systems, preventing catastrophic failures while maintaining learning capabilities. Discover the computational aspects of MPC as a control strategy and its practical applications to reinforcement learning environments. Examine real-world experiments from academic research at University of Lübeck, including bachelor thesis findings that demonstrate the effectiveness of using mathematical models as supervisors during training phases. Understand how environmental and system models can act as safety boundaries, ensuring that reinforcement learning agents operate within acceptable behavioral parameters. The presentation covers both theoretical foundations and practical implementations, with potential live demonstrations of simplified MPC-based safe learning architectures that showcase how to train intelligent systems without risking infrastructure damage or safety violations.