paper-with-me

Papers

Grounded Reinforcement Learning for Visual Reasoning

2025-05-29 · Gabriel Sarch, Snigdha Saha, Naitik Khandelwal, Ayush Jain, Michael J. Tarr, Aviral Kumar, Katerina Fragkiadaki

While reinforcement learning (RL) over chains of thought has significantly advanced language models in tasks such as mathematics and coding, visual reasoning introduces added complexity by requiring models to direct visual attention, interpret perceptual inputs, and ground abstract reasoning in spatial evidence. We introduce ViGoRL (Visually Grounded Reinforcement Learning), a vision-language model trained with RL to explicitly anchor each reasoning step to specific visual coordinates. Inspired by human visual decision-making, ViGoRL learns to produce spatially grounded reasoning traces, guiding visual attention to task-relevant regions at each step. When fine-grained exploration is required, our novel multi-turn RL framework enables the model to dynamically zoom into predicted coordinates as reasoning unfolds. Across a diverse set of visual reasoning benchmarks--including SAT-2 and BLINK for spatial reasoning, V*bench for visual search, and ScreenSpot and VisualWebArena for web-based grounding--ViGoRL consistently outperforms both supervised fine-tuning and conventional RL baselines that lack explicit grounding mechanisms. Incorporating multi-turn RL with zoomed-in visual feedback significantly improves ViGoRL's performance on localizing small GUI elements and visual search, achieving 86.4% on V*Bench. Additionally, we find that grounding amplifies other visual behaviors such as region exploration, grounded subgoal setting, and visual verification. Finally, human evaluations show that the model's visual references are not only spatially accurate but also helpful for understanding model reasoning steps. Our results show that visually grounded RL is a strong paradigm for imbuing models with general-purpose visual reasoning.

📄 PDF Abstract BibTeX arXiv:2505.23678

Code (1)

hiyouga/easyr1 공식 구현 pytorch

Tasks

reinforcement-learningReinforcement LearningReinforcement Learning (RL)Spatial ReasoningVisual Reasoning

Methods 이 논문이 사용한 방법론

Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…
Attention 설명 없음
SET Dynamic Sparse Training method where weight mask is updated randomly periodically

Similar Papers 제목 키워드 기반

Point-RFT: Improving Multimodal Reasoning with Visually Grounded Reinforcement Finetuning

2025-05-26 · Minheng Ni, Zhengyuan Yang, Linjie Li, Chung-Ching Lin 외

Recent advances in large language models have significantly improved textual reasoning through the effective use of Chain-of-Thought (CoT) and reinforcement learning. However, extending these successes to vision-language…

document understandingMultimodal ReasoningVisual Reasoning

GRIT: Teaching MLLMs to Think with Images

2025-05-21 · Yue Fan, Xuehai He, Diji Yang, Kaizhi Zheng 외

Recent studies have demonstrated the efficacy of using Reinforcement Learning (RL) in building reasoning models that articulate chains of thoughts prior to producing final answers. However, despite ongoing advances that …

Reinforcement Learning (RL)Visual Reasoning

MMRPT: MultiModal Reinforcement Pre-Training via Masked Vision-Dependent Reasoning

2025-12-08 · Xuhui Zheng, Kang An, Ziliang Wang, Yuhang Wang 외 arxiv

Multimodal pre-training remains constrained by the descriptive bias of image-caption pairs, leading models to favor surface linguistic cues over grounded visual understanding. We introduce MMRPT, a masked multimodal rein…

Reinforcement LearningVisual GroundingVisual Reasoning

Be Faithful When Response: Returning Fluent and Grounded Answers for Vision-Language Models Reinforcement Learning

2026-06-29 · Peng, Lee, Yin Zhang, Yanglin Zhang 외 arxiv

Reinforcement Learning (RL) is an important paradigm for improving the reasoning capabilities of Vision-Language Models (VLMs). However, directly applying RL to rollout multimodal reasoning can lead to instability, due t…

Reinforcement LearningMultimodal Reasoning

iVGR: Internalizing Visually Grounded Reasoning for MLLMs with Reinforcement Learning

2026-05-29 · Chang-Bin Zhang, Yujie Zhong, Qiang Zhang, Kai Han arxiv

While visually grounded Chain-of-Thought (CoT) has emerged as a promising paradigm to enhance fine-grained perception in multimodal large language models (MLLMs), its efficacy during the inference phase remains underexpl…

Reinforcement LearningVisual LocalizationVisual Grounding