מושגי ליבה
LLMs are used to translate natural language task descriptions into formal task specifications, improving task success rates in complex environments.
תקציר
I. Introduction
Effective human-robot interaction requires robots to understand, plan, and execute complex tasks described in natural language.
Recent advances in large language models (LLMs) show promise for translating natural language into robot action sequences.
Existing approaches face challenges with complex environmental and temporal constraints when using LLMs directly for planning tasks.
II. Problem Description
Aim to convert natural language instructions into motion plans for robots.
Generate constraint-satisfying trajectories based on instructions and environment state.
III. Methods
Three approaches compared: LLM End-to-end Motion Planning, LLM Task Planning, Autoregressive LLM Specification Translation & Checking + Formal Planner.
Autoregressive re-prompting technique introduced for semantic error correction.
IV. Experimental Design
Six different task scenarios evaluated in 2D environments.
Performance compared across methods using GPT-3 and GPT-4 as the LLM.
V. Results
AutoTAMP outperforms other methods in handling tasks with complex geometric and temporal constraints.
Syntactic and semantic error correction significantly improves translation performance.
VI. Related Work
Previous research on task and motion planning, using LLMs for TAMP, translating language to task representations, and re-prompting of LLMs discussed.
VII. Conclusion
AutoTAMP framework leverages pre-trained LLMs for translating NL to formal task specifications, improving success rates in challenging environments.
סטטיסטיקה
LLMsを使用して自然言語のタスク記述を形式的なタスク仕様に変換し、複雑な環境でのタスク成功率を向上させる。