LLM-DWA: a hybrid path planning framework combining large language models with the dynamic window approach.

Area of Science:

• Robotics
• Artificial Intelligence

Background:

• The Dynamic Window Approach (DWA) algorithm faces challenges with local minima and inefficient path planning in complex environments.
• Conventional DWA lacks the ability to incorporate prior environmental knowledge, leading to degraded performance, particularly with U-shaped obstacles.

Purpose of the Study:

• To improve the goal-reaching performance and efficiency of the DWA algorithm.
• To address the local minima problem and reduce planning time in complex navigation scenarios.

Main Methods:

• Integration of Large Language Models (LLMs) with the Dynamic Window Approach (DWA).
• Utilizing LLMs' reasoning capabilities to interpret environmental information and generate intermediate waypoints.
• Experimental validation in 2D grid environments and 3D simulation platforms.

Main Results:

• The proposed LLM-based hybrid method significantly improves efficiency in U-shaped obstacle scenarios.
• Shorter goal-reaching times were observed compared to the conventional DWA.
• Demonstrated enhanced navigation performance in complex environments.

Conclusions:

• Combining LLMs with DWA effectively overcomes limitations of the conventional approach.
• LLM-enhanced DWA offers a promising solution for efficient and robust robot navigation in complex environments.
• The hybrid method shows superior performance in scenarios with challenging obstacle configurations.