Large-Language-Model-Guided State Estimation for Partially Observable Task and Motion Planning

Researchers developed CoCo-TAMP, a robot planning framework that uses large language models to improve state estimation in partially observable environments. The system leverages LLMs' common-sense reasoning to predict object locations and co-locations, achieving 62-73% reduction in planning time compared to baseline methods.

• →CoCo-TAMP framework integrates LLMs with robotic planning to handle partially observable environments more efficiently.
• →The system uses two types of common-sense knowledge: location-based object likelihood and object co-location patterns.
• →LLMs eliminate the need for manual engineering of common-sense knowledge in robotic planning systems.
• →Real-world demonstrations showed 72.6% improvement in planning and execution time over baseline methods.
• →The approach enables more efficient solutions to long-horizon task and motion planning problems in robotics.