CoPAL introduces a novel architecture where language models guide robots through complex tasks with real-time error correction.

• Orchestrates multiple cognitive levels across reasoning, planning, and motion
• Implements a corrective planning approach that detects and resolves execution failures
• Demonstrates effectiveness in real-world scenarios including pizza preparation
• Bridges the gap between high-level task understanding and physical execution

This research advances engineering by creating more robust autonomous systems that can handle unpredictable environments—a critical advancement for factory automation and human-assistant robots.

CoPAL: Corrective Planning of Robot Actions with Large Language Models