CLEA introduces a novel architecture using specialized LLMs to enable robots to reliably execute complex tasks through adaptive planning and real-time adjustments.

• Integrates four specialized open-source LLMs with distinct functions to create a closed-loop system
• Enables hierarchical decomposition of complex tasks through semantic reasoning
• Improves one-shot success for long-term task execution in changing environments
• Offers real-time adaptation to unexpected obstacles or changes

This research advances robotic engineering by creating more reliable and flexible systems capable of operating in unpredictable real-world environments, potentially transforming factory automation and other industrial applications.

CLEA: Closed-Loop Embodied Agent for Enhancing Task Execution in Dynamic Environments