Skip to main content

Cognitive Planning with LLMs

Learning Objectives​

  • Learn how LLMs decompose complex natural language commands.
  • Understand prompt engineering for robotic task planning.
  • Create a sequence of actionable ROS 2 commands.

Task Decomposition​

LLMs are excellent at taking high-level goals (e.g., "Bring me the water") and decomposing them into sub-goals:

  1. Find_Water
  2. Navigate_To_Water
  3. Grasp_Water
  4. Return_To_User

Prompt Engineering​

The key is to instruct the LLM to output actions in a structured, machine-readable format (e.g., JSON) that your ROS 2 node can easily parse.

Takeaways​

  • LLMs provide the reasoning layer between human intent and robotic execution.
  • Structural prompting is essential for reliable task execution.

Review Questions​

  1. How does an LLM help in complex robotic task planning?