Work

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

Year: 2026

Type: article

Abstract: This research addresses the local minima problem in the Dynamic Window Approach (DWA) algorithm. The conventional DWA, which does not incorporate prior environmental knowledge, often exhibits degraded goal-reaching performance in complex scenarios, such as environments with U-shaped obstacles, and even when it reaches the goal, the planning time can be relatively long. To overcome this limitation, we propose an improved DWA by integrating Large Language Models (LLMs). Leveraging the reasoning capabilities of LLMs, prior environmental information is interpreted, and appropriate intermediate waypoints are generated. Experimental results in both 2D grid environments and 3D simulation platforms demonstrate that the proposed LLM-based hybrid method achieves higher efficiency and shorter goal-reaching times in U-shaped obstacle scenarios compared to the conventional DWA. These findings highlight the effectiveness of combining the reasoning capabilities of LLMs with DWA to improve navigation performance in complex environments. A video demonstration is available at https://youtu.be/Otn53HS4KC4 .

Source: Scientific Reports

DOI: 10.1038/s41598-026-39524-1

Authors: Jeonghee Seo, Eunsung Kim, Andrew Jaeyong Choi

Institution: Gachon University

Language: English

Topic: Robotic Path Planning Algorithms

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $2,690

Funders: Gachon University, Institute for Information and Communications Technology Promotion