Robust-U1: Can MLLMs Self-Recover Corrupted Visual Content for Robust Understanding?

Multimodal Large Language Models (MLLMs) have demonstrated remarkable success in visual understanding, yet their performance degrades significantly under real-world visual corruptions. While existing robustness enhancement approaches exist, they are limited: black-box feature alignment lacks interpretability, and white-box text-based reasoning cannot restore lost pixel-level details. This work investigates a fundamental research question: Can MLLMs recover corrupted visual content by themselves? To address this, we propose Robust-U1, a novel framework that equips MLLMs with explicit visual self-recovery capability for robust understanding. The approach comprises three core stages: supervised fine-tuning for initial reconstruction, reinforcement learning with dual rewards (pixel-level SSIM and semantic-level CLIP similarity) for aligning high visual quality, and multimodal reasoning that jointly considers both the corrupted input and the recovered image. Extensive experiments demonstrate that Robust-U1 achieves state-of-the-art robustness on the real-world corruption benchmark and maintains superior performance under adversarial corruptions on general VQA benchmarks. Analysis confirms that high-quality visual recovery directly enhances reasoning performance, establishing self-recovery as a critical mechanism for robust visual understanding. The source code is available at https://github.com/jqtangust/Robust-U1.

Imagine trying to answer a question about a photo that’s blurry, grainy, or covered in digital noise. Even the smartest AI models struggle with this—they get confused and make mistakes. That’s a big problem because real-world photos are rarely perfect. Today’s fixes have two main limits. Some methods try to guess the answer without actually cleaning up the picture. Others clean it up using a separate, black-box tool that we don’t fully understand or control. We asked a different question: what if the AI could learn to fix the image inside its own mind before answering? We built a system that does exactly that. Think of it as teaching the AI to mentally sharpen a messy photo. Our training has three simple steps. First, we give the model basic repair practice. Next, we use a reward system to push the model to produce repairs that not only look right, pixel by pixel, but also make sense when you understand the scene. Finally, we show the model both the original messy photo and its own cleaned-up version, so it can reason about them together. The results are exciting. Our self-repairing AI is much more reliable when facing blurry, noisy, or even deliberately tricky images. It beats previous methods on standard tests. The takeaway is clear: giving AI the ability to quietly clean up what it sees before thinking is a powerful way to make it tougher and more trustworthy in the real world. We’ve shared the code publicly so others can build on this idea.