Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wu, Di, Fan, Jiaxin, Zang, Junzhe, Wang, Guanbo, Yin, Wei, Li, Wenhao, Jin, Bo
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2505.22050
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918089874997248
author Wu, Di
Fan, Jiaxin
Zang, Junzhe
Wang, Guanbo
Yin, Wei
Li, Wenhao
Jin, Bo
author_facet Wu, Di
Fan, Jiaxin
Zang, Junzhe
Wang, Guanbo
Yin, Wei
Li, Wenhao
Jin, Bo
contents Embodied planning requires agents to make coherent multi-step decisions based on dynamic visual observations and natural language goals. While recent vision-language models (VLMs) excel at static perception tasks, they struggle with the temporal reasoning, spatial understanding, and commonsense grounding needed for planning in interactive environments. In this work, we introduce a reinforcement fine-tuning framework that brings R1-style reasoning enhancement into embodied planning. We first distill a high-quality dataset from a powerful closed-source model and perform supervised fine-tuning (SFT) to equip the model with structured decision-making priors. We then design a rule-based reward function tailored to multi-step action quality and optimize the policy via Generalized Reinforced Preference Optimization (GRPO). Our approach is evaluated on Embench, a recent benchmark for interactive embodied tasks, covering both in-domain and out-of-domain scenarios. Experimental results show that our method significantly outperforms models of similar or larger scale, including GPT-4o-mini and 70B+ open-source baselines, and exhibits strong generalization to unseen environments. This work highlights the potential of reinforcement-driven reasoning to advance long-horizon planning in embodied AI.
format Preprint
id arxiv_https___arxiv_org_abs_2505_22050
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Reinforced Reasoning for Embodied Planning
Wu, Di
Fan, Jiaxin
Zang, Junzhe
Wang, Guanbo
Yin, Wei
Li, Wenhao
Jin, Bo
Artificial Intelligence
Machine Learning
Embodied planning requires agents to make coherent multi-step decisions based on dynamic visual observations and natural language goals. While recent vision-language models (VLMs) excel at static perception tasks, they struggle with the temporal reasoning, spatial understanding, and commonsense grounding needed for planning in interactive environments. In this work, we introduce a reinforcement fine-tuning framework that brings R1-style reasoning enhancement into embodied planning. We first distill a high-quality dataset from a powerful closed-source model and perform supervised fine-tuning (SFT) to equip the model with structured decision-making priors. We then design a rule-based reward function tailored to multi-step action quality and optimize the policy via Generalized Reinforced Preference Optimization (GRPO). Our approach is evaluated on Embench, a recent benchmark for interactive embodied tasks, covering both in-domain and out-of-domain scenarios. Experimental results show that our method significantly outperforms models of similar or larger scale, including GPT-4o-mini and 70B+ open-source baselines, and exhibits strong generalization to unseen environments. This work highlights the potential of reinforcement-driven reasoning to advance long-horizon planning in embodied AI.
title Reinforced Reasoning for Embodied Planning
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2505.22050