Georg Schäfer
Georg Schäfer is an assistant professor and researcher in the Department of Information Technologies and Digitalisation at Salzburg University of Applied Sciences, where he is affiliated with the Josef Ressel Centre for Intelligent and Secure Industrial Automation. He is known for his research on safe reinforcement learning for industrial automation, particularly frameworks that combine deep reinforcement learning with model predictive control to guarantee constraint satisfaction during training and deployment on physical hardware such as the Quanser Aero 2 testbed. His work bridges Python-based reinforcement learning training with Simulink models and physical systems through systematic software architectures enabling zero-shot deployment.
“Rather than evaluating safety purely online, which is often computationally prohibitive for the high-frequency control loops required in modern CPSs, our approach uses offline MPC as an oracle to pre-define a 'feasible state-action space' (F)”
Source→“Empirical observations indicate that the agent successfully explores the extreme boundaries of the state space, achieving high control performance while remaining within the defined feasible bounds for the vast majority of the training process”
Source→“The authors wrapped their compiled Simulink model in a Farama Gymnasium interface, enabling high-frequency, zero-shot deployment to the physical hardware”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.