Graph Contextual Reinforcement Learning for Efficient Directed Controller Synthesis
This work addresses efficiency in controller synthesis for formal methods, though it is incremental as it builds on existing RL-based methods with a novel graph-based enhancement.
The paper tackled the problem of inefficient controller synthesis by introducing GCRL, which integrates Graph Neural Networks with reinforcement learning to capture broader context, resulting in superior learning efficiency and generalization in four out of five benchmark domains.
Controller synthesis is a formal method approach for automatically generating Labeled Transition System (LTS) controllers that satisfy specified properties. The efficiency of the synthesis process, however, is critically dependent on exploration policies. These policies often rely on fixed rules or strategies learned through reinforcement learning (RL) that consider only a limited set of current features. To address this limitation, this paper introduces GCRL, an approach that enhances RL-based methods by integrating Graph Neural Networks (GNNs). GCRL encodes the history of LTS exploration into a graph structure, allowing it to capture a broader, non-current-based context. In a comparative experiment against state-of-the-art methods, GCRL exhibited superior learning efficiency and generalization across four out of five benchmark domains, except one particular domain characterized by high symmetry and strictly local interactions.