UniVR: Thinking in Visual Space for Unified Visual Reasoning

2026-07-14

AI Summary

This work addresses the challenge of unifying complex reasoning, fine-grained physical dynamics modeling, and long-horizon planning solely from visual inputs. To this end, it proposes UniVR, a novel framework that, for the first time, enables simultaneous learning across diverse visual reasoning tasks under purely visual conditions—without relying on task-specific heuristics or image–text pairs. The approach introduces VR-GRPO, a reinforcement learning paradigm that integrates both global and step-level rewards to ensure logical coherence and physical consistency in generated reasoning trajectories. The study also establishes VR-X, the first purely visual benchmark for multi-capability evaluation, on which UniVR achieves performance gains of up to 25% and demonstrates significant superiority over existing methods across multiple multimodal understanding tasks.

Abstract

Learning broad world knowledge directly from raw visual data is a fundamental capability of intelligence. We introduce UniVR, the first investigation into simultaneously learning complex reasoning, fine-grained physical dynamics, and long-term planning from pure visual demonstrations. At its core, UniVR features VR-GRPO, a reinforcement learning paradigm with complementary global and step-level rewards. This approach enforces logical coherence and physical consistency throughout the reasoning process without requiring task-specific heuristics or image-text pairs. To train and evaluate UniVR, we construct VR-X, a large-scale benchmark curated from 16 diverse sources spanning long-horizon manipulation, spatial puzzles, and physical reasoning. It is the first comprehensive suite to assess these heterogeneous capabilities under a purely visual protocol. Remarkably, UniVR achieves up to a 25% improvement on VR-X, and its superior visual reasoning also boosts performance on various multimodal understanding benchmarks. These findings underscore the vast potential of reasoning within visual spaces, with all code, data, and models are open-sourced for further research.

Problem

• visual reasoning
• physical dynamics
• long-term planning
• pure visual demonstrations
• world knowledge

Innovation

• visual reasoning
• reinforcement learning
• physical dynamics
• long-term planning
• pure visual learning