# On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMs

arXiv: https://arxiv.org/abs/2602.12506 · HTML: https://mlsanity.com/paper/2602.12506/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 70 of its 97 references that are in the map.
- First posted on arXiv 13 Feb 2026.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)
- Multimodal LLMs / vision-language models / [Embodied and 3D vision-language models](https://mlsanity.com/mllm/embodied-and-3d-vision-language-models/)

## Builds on (most cited first, 30 of 70)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [LLaVA-OneVision: Easy Visual Task Transfer](https://mlsanity.com/paper/2408.03326/) (2024)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models](https://mlsanity.com/paper/2504.10479/) (2025)
- [Understanding R1-Zero-Like Training: A Critical Perspective](https://mlsanity.com/paper/2503.20783/) (2025)
- [OpenAI o1 System Card](https://mlsanity.com/paper/2412.16720/) (2024)
- [Kimi k1.5: Scaling Reinforcement Learning with LLMs](https://mlsanity.com/paper/2501.12599/) (2025)
- [Kosmos-2: Grounding Multimodal Large Language Models to the World](https://mlsanity.com/paper/2306.14824/) (2023)
- [Tulu 3: Pushing Frontiers in Open Language Model Post-Training](https://mlsanity.com/paper/2411.15124/) (2024)
- [Shikra: Unleashing Multimodal LLM's Referential Dialogue Magic](https://mlsanity.com/paper/2306.15195/) (2023)
- [Emerging Properties in Unified Multimodal Pretraining](https://mlsanity.com/paper/2505.14683/) (2025)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models](https://mlsanity.com/paper/2503.06749/) (2025)
- [VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model](https://mlsanity.com/paper/2504.07615/) (2025)
- [Ferret: Refer and Ground Anything Anywhere at Any Granularity](https://mlsanity.com/paper/2310.07704/) (2023)
- [The Entropy Mechanism of Reinforcement Learning for Reasoning Language Models](https://mlsanity.com/paper/2505.22617/) (2025)
- [Video-R1: Reinforcing Video Reasoning in MLLMs](https://mlsanity.com/paper/2503.21776/) (2025)
- [Reasoning Models Don't Always Say What They Think](https://mlsanity.com/paper/2505.05410/) (2025)
- [Hallucination of Multimodal Large Language Models: A Survey](https://mlsanity.com/paper/2404.18930/) (2024)
- [Kimi-VL Technical Report](https://mlsanity.com/paper/2504.07491/) (2025)
- [Demystifying Long Chain-of-Thought Reasoning in LLMs](https://mlsanity.com/paper/2502.03373/) (2025)
- [VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning](https://mlsanity.com/paper/2504.08837/) (2025)

## Built on by (most cited first, 11 of 11)

- [HopChain: Multi-Hop Data Synthesis for Generalizable Vision-Language Reasoning](https://mlsanity.com/paper/2603.17024/) (2026)
- [Evidence-RL: Towards Evidence-intensive Visual Reasoning](https://mlsanity.com/paper/2608.08021/) (2026)
- [Thinking With Tools, Not With Pixels: Tool Calls as Text Scaffolds for Visual Reasoning](https://mlsanity.com/paper/2608.09682/) (2026)
- [When to Call an Apple Red: Humans Follow Introspective Rules, VLMs Don't](https://mlsanity.com/paper/2604.06422/) (2026)
- [Beyond Visual CoT: Internalized Visual Thinking for Proactive Video Reasoning](https://mlsanity.com/paper/2608.15869/) (2026)
- [The Cognitive Penalty: Ablating System 1 and System 2 Reasoning in Edge-Native SLMs for Decentralized Consensus](https://mlsanity.com/paper/2604.16913/) (2026)
- [Bridging Physical Reasoning and Task Generalization via Visual Action Outcome Reasoning Alignment](https://mlsanity.com/paper/2607.06522/) (2026)
- [When Does Consensus Mean Correctness? Measuring the Agreement-Accuracy Coupling with Semantics-Preserving Re-Rendering](https://mlsanity.com/paper/2608.05670/) (2026)
- [Consistency Has a Computable Blind Spot: A Commutation Theory of Label-Free Reliability for Vision-Language Figure Reading](https://mlsanity.com/paper/2608.05675/) (2026)
- [One Perturbation Is Not Enough: Identifiability and Blind Baselines for Behavioral AI Evaluation](https://mlsanity.com/paper/2609.06190/) (2026)
- [Taming CoT Obfuscation in VLMs: From Mechanistic Evidence to Activation Enforcement](https://mlsanity.com/paper/2609.24243/) (2026)

## Abstract (arXiv)

Reinforcement learning (RL) finetuning has become a key technique for enhancing large language models (LLMs) on reasoning-intensive tasks, motivating its extension to vision-language models (VLMs). While RL-tuned VLMs improve on visual reasoning benchmarks, they remain vulnerable to weak visual grounding, hallucinations, and over-reliance on textual cues. We show that simple, controlled textual perturbations, including misleading captions or incorrect chain-of-thought (CoT) traces, cause substantial drops in robustness and confidence, and that these effects are more pronounced when CoT consistency is taken into account across open-source multimodal reasoning models. In contrast, closed models exhibit similar failure modes but maintain markedly greater robustness and reasoning consistency, suggesting that the gap reflects a shortcoming in current open-source RL finetuning rather than an inherent limitation of the task. To better understand these vulnerabilities, we further analyze RL finetuning dynamics and uncover an accuracy-faithfulness trade-off: finetuning raises benchmark accuracy, but can simultaneously erode the reliability of the accompanying CoT and its robustness to contextual shifts. Although adversarial augmentation improves robustness, it does not by itself prevent faithfulness drift. Incorporating a faithfulness-aware reward can restore alignment between answers and reasoning, but when paired with augmentation, training risks collapsing onto shortcut strategies and robustness remains elusive. Together, these findings highlight the limitations of accuracy-only evaluations and motivate training and assessment protocols that jointly emphasize correctness, robustness, and the faithfulness of visually grounded reasoning.