LTL

We introduce Self-Constrained Reasoning, a novel approach that embeds an LTL-specific synchronous grammar into hierarchical reinforcement learning, enabling LLM-based task specification while preserving formal validity and generalization.