🤖 AI Summary
To address the inexecutability of high-level plans for autonomous robots operating in open-world environments—caused by physical, logical, and semantic errors—this paper proposes a multi-cognitive-level collaborative architecture. The architecture integrates (i) high-level semantic reasoning via an LLM (LLaMA-2), (ii) mid-level logical decomposition using a symbolic planner (PDDL), and (iii) low-level motion generation through ROS. Crucially, we introduce a novel closed-loop dynamic re-planning mechanism that classifies and responds to physical, logical, and semantic failures in real time, enabling executable validation and adaptive refinement of LLM-generated plans. Evaluation across Blocks World, Barman, and a real-world pizza-making task demonstrates significant improvements over baseline methods: a 41% increase in plan executability, a 37% gain in semantic correctness, and an average re-planning latency under 1.2 seconds.
📝 Abstract
In the pursuit of fully autonomous robotic systems capable of taking over tasks traditionally performed by humans, the complexity of open-world environments poses a considerable challenge. Addressing this imperative, this study contributes to the field of Large Language Models (LLMs) applied to task and motion planning for robots. We propose a system architecture that orchestrates a seamless interplay between multiple cognitive levels, encompassing reasoning, planning, and motion generation. At its core lies a novel replanning strategy that handles physically grounded, logical, and semantic errors in the generated plans. We demonstrate the efficacy of the proposed feedback architecture, particularly its impact on executability, correctness, and time complexity via empirical evaluation in the context of a simulation and two intricate real-world scenarios: blocks world, barman and pizza preparation.
