Context-Driven Task and Motion Planning with Language Models

언어 모델을 이용한 상황 추론 기반 작업 계획 생성

This paper presents a task planning framework for autonomous robots that combines Large Language Models (LLMs) and Vision-Language Models (VLMs) to enable context-driven reasoning and robust action generation. The framework infers preconditions by structuring environmental states and task goals into contextual representations, which are then refined through prompt engineering. These structured prompts are aligned with a predefined action library to generate task sequences that are both logically consistent and physically executable. By integrating symbolic task reasoning with motion planning, the method bridges the gap between high-level natural language understanding and low-level robotic control. Experiments conducted in simulated dynamic environments demonstrate that the approach produces reliable and adaptable task plans, even under incomplete or ambiguous natural language instructions. The results suggest that context-based inference and prompt optimization with LLM/VLMs provide a scalable pathway for developing autonomous robots capable of flexible task execution in real-world environments.