Reactive Synthesis

Reactive synthesis is a computational approach to automatically designing controllers that guarantee desired system behavior, often modeled as a game against an uncertain environment. Current research emphasizes extending beyond purely "winning" strategies to encompass notions of admissibility and regret minimization, particularly relevant in robotics and human-robot interaction where winning strategies may not exist. This involves developing algorithms and models, including those based on symbolic methods, reinforcement learning (both model-free and model-based), and neuro-symbolic approaches, to synthesize controllers that are both efficient and robust to uncertainty. The field's impact spans theoretical computer science and practical applications in areas like robotics, control systems, and the safe deployment of machine learning models.