This research introduces a novel approach that extracts knowledge from Large Language Models to produce object-level plans that guide robots in completing complex tasks.

• Bridges the gap between LLMs and physical task planning by focusing on object states rather than direct action sequences
• Uses Functional Object-Oriented Networks (FOON) to represent how objects change during manipulation tasks
• Significantly outperforms baseline approaches in pick-and-place scenarios
• Provides a more robust foundation for factory automation and industrial robotics

This advancement matters for engineering because it creates a more reliable framework for robotic systems to understand and execute complex manipulation tasks in real-world environments, potentially transforming factory automation.

Bootstrapping Object-level Planning with Large Language Models