Tag
This paper proposes a multi-role reinforcement learning framework using solver feedback to convert natural language into PDDL specifications for symbolic planning, improving success rates and faithfulness compared to existing methods.