Failure-Aware LLM-DWA Replanning for Mobile Robot Navigation in Dynamic Obstacle Environments
This study evaluates failure-aware LLM-DWA replanning in a simulated maze with three moving obstacles. The proposed method achieved the highest observed navigation success rate among the tested configurations, reaching the goal in 5/10 trials compared with 1/10 for one-shot LLM-DWA and 0/10 for NavFn-DWA and both periodic replanning baselines. Ablation experiments further examined how trigger timing, contact-based replanning, and LLM temperature affect navigation outcomes.
* Dabin Kim and Youngmin Lee are co-first authors.
Read research notes