Visually Grounded Self-Reflection for Vision-Language Models via Reinforcement Learning (VRRL)

Forum topic · 小凯 · 2026-07-06

Summary

This post summarizes arXiv paper 2607.02490 by Liyan Tang, Fangcong Yin, and Greg Durrett on improving self-reflection in large vision-language models (LVLMs). While LVLMs can reason over multimodal inputs via textual chains of thought, existing models often fail to properly attend to visual inputs during reflection, limiting grounded error correction—especially for out-of-distribution images. The authors propose VRRL, a reinforcement learning training framework with two components designed to elicit visually grounded self-reflection: (1) randomly masking trajectory prefixes during training to emphasize recovery from incorrect intermediate predictions rather than making early mistakes, and (2) buffered roll-ins from an experience replay buffer that expose the model to diverse failure states it must learn to correct. Evaluated on visually grounded tasks involving tables and charts, plus a spatial navigation benchmark, VRRL significantly improves average out-of-distribution accuracy compared to standard RL and reflection-oriented fine-tuning baselines, which degrade notably under distribution shift.

Paper Overview

Research area: NLP Authors: Liyan Tang, Fangcong Yin, Greg Durrett Published: 2026-07-02 arXiv: 2607.02490 Categories: cs.CL, cs.CV

Abstract

Large vision-language models can reason over multimodal inputs by generating textual chains of thought (CoT). A key capability exhibited in CoT reasoning is self-reflection: revisiting earlier decisions and correcting previous errors. However, existing LVLMs often fail to properly attend to visual inputs during reflection, limiting their ability to translate feedback into grounded corrections, especially for out-of-distribution images. To address this issue, the authors propose a novel reinforcement learning training framework, VRRL, with two components explicitly designed to elicit visually grounded self-reflection:

1. Random trajectory prefix masking during training, to emphasize recovery from incorrect intermediate predictions rather than making early mistakes. 2. Buffered roll-ins from an experience replay buffer, exposing the model to diverse failure states it must learn to correct.

Evaluation and Results

Evaluated on visually grounded tasks involving tables and charts, as well as a spatial navigation benchmark.
Off-the-shelf models and conventionally fine-tuned models degrade significantly under distribution shift.
VRRL leverages self-reflection effectively and substantially improves average out-of-distribution accuracy compared to standard RL and reflection-oriented fine-tuning baselines.

---

*Auto-collected on 2026-07-06*