Any House Any Task: Scalable Long-Horizon Planning for Abstract Human Tasks

This research presents AHAT, a household task planner designed for long-horizon planning in large environments with ambiguous human instructions. AHAT uses an LLM to map task instructions and textual scene graphs into grounded subgoals, which are then solved through symbolic reasoning. The paper introduces TGPO, a reinforcement learning algorithm that integrates external correction of reasoning traces, leading to significant performance improvements over existing methods in human-style household tasks.