EvoNav: Evolutionary Reward Function Design for Robot Navigation with Large Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Zhao, Zhikai, Hua, Chuanbo, Berto, Federico, Ma, Zihan, Lee, Kanghoon, Li, Jiachen, Park, Jinkyoo
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909035560697856
author Zhao, Zhikai
Hua, Chuanbo
Berto, Federico
Ma, Zihan
Lee, Kanghoon
Li, Jiachen
Park, Jinkyoo
author_facet Zhao, Zhikai
Hua, Chuanbo
Berto, Federico
Ma, Zihan
Lee, Kanghoon
Li, Jiachen
Park, Jinkyoo
contents Robot navigation is a crucial task with applications to social robots in dynamic human environments. While Reinforcement Learning (RL) has shown great promise for this problem, the policy quality is highly sensitive to the specification of reward functions. Hand-crafted rewards require substantial domain expertise and embed inductive biases that are difficult to audit or adapt, limiting their effectiveness and leading to suboptimal performance. In this paper, we propose EvoNav, an evolutionary framework that automates the design of robot navigation reward functions via large language models (LLMs). To overcome prohibitively costly policy training, EvoNav evaluates each candidate proposal from the LLM via a progressive three-stage warm-up-boost procedure. EvoNav advances from analytical proxies with low-cost surrogates, such as small datasets and analytic rules, to lightweight rollouts and, finally, to full policy training, enabling computationally efficient exploration under effective feedback. Experiment results show that EvoNav produces more effective navigation policies than manually designed RL rewards and state-of-the-art reward design methods.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11859
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle EvoNav: Evolutionary Reward Function Design for Robot Navigation with Large Language Models
Zhao, Zhikai
Hua, Chuanbo
Berto, Federico
Ma, Zihan
Lee, Kanghoon
Li, Jiachen
Park, Jinkyoo
Robotics
Artificial Intelligence
Robot navigation is a crucial task with applications to social robots in dynamic human environments. While Reinforcement Learning (RL) has shown great promise for this problem, the policy quality is highly sensitive to the specification of reward functions. Hand-crafted rewards require substantial domain expertise and embed inductive biases that are difficult to audit or adapt, limiting their effectiveness and leading to suboptimal performance. In this paper, we propose EvoNav, an evolutionary framework that automates the design of robot navigation reward functions via large language models (LLMs). To overcome prohibitively costly policy training, EvoNav evaluates each candidate proposal from the LLM via a progressive three-stage warm-up-boost procedure. EvoNav advances from analytical proxies with low-cost surrogates, such as small datasets and analytic rules, to lightweight rollouts and, finally, to full policy training, enabling computationally efficient exploration under effective feedback. Experiment results show that EvoNav produces more effective navigation policies than manually designed RL rewards and state-of-the-art reward design methods.
title EvoNav: Evolutionary Reward Function Design for Robot Navigation with Large Language Models
topic Robotics
Artificial Intelligence
url https://arxiv.org/abs/2605.11859