LLM-Enhanced A* Path Planning

Overview

Redesigned and improved the LLM-A* hybrid path planning system from the ground up, integrating refined LLM waypoint guidance with classical A* search.

The Challenge

Traditional path planning algorithms can be computationally expensive. The goal was to leverage LLM reasoning capabilities to guide classical planners.

The Solution

Developed a novel integration of LLM-generated waypoint suggestions with A* search. Compared chain-of-thought, minimalistic, and RePE prompting methods.

Key Results

• 23.4% reduction in node expansions on 10×10 grids
• 21.6% improvement on 20×20 grids
• 17.8% boost in waypoint accuracy
• Faster, more resource-efficient navigation