This project develops new foundations for scalable and adaptable logical controller synthesis. Classical synthesis approaches compute a controller for one fixed and fully specified plant model, which often leads to severe scalability limitations and brittle solutions when the system changes. Our work on Universal Safety Controllers (USCs) replaces this paradigm with generic controllers that can adapt to entire families of plants through symbolic conditions, called prophecies, describing relevant aspects of plant behavior. This shifts controller synthesis from designing a single controller for a single model to synthesizing reusable and explainable control principles that generalize across systems. More recently, we combine this idea with learning-based methods that infer compact and human-readable logical representations of prophecies from example plants, significantly improving efficiency and interpretability while preserving formal guarantees.

(Past) Group Members Involved:

• Satya Prakash Nayak
• Anne-Kathrin Schmuck

Main Collaborators:

• Bernd Finkbeiner (CISPA / TU Munich)
• Niklas Metzger (TU Munich)

Publications:

• * B. Finkbeiner, N. Metzger, S.P. Nayak, A.-K. Schmuck. Universal Safety Controllers with Learned Prophecies. 2026. AAAI’26. (extended version)
• * B. Finkbeiner, N. Metzger, S.P. Nayak, A.-K. Schmuck. Synthesis of Universal Safety Controllers. 2025. TACAS’25 (PDF, extended version)