ConsiSpace: Learning Geometric Consistency Matters for Video Spatial Reasoning
For navigation-oriented perception and long-video QA, this work addresses the bottleneck of spatial consistency in video reasoning, achieving strong gains over existing methods.
Existing multimodal large language models fail at video spatial reasoning due to semantic-centric design and inability to aggregate consistent spatial evidence. ConsiSpace introduces a geometry-consistency-aware framework with a geometry-consistent memory and unified consistency self-supervised reinforcement learning, improving average score by 12.6 points over baselines on three benchmarks.
Video spatial reasoning is essential for navigation-oriented perception and long-video question answering, where models must infer spatial relations across long horizons under changing viewpoints. However, existing multimodal large language models (MLLMs) remain largely semantic-centric, and often fail to reliably aggregate consistent spatial evidence from redundant video observations, leading to inefficient or unstable reasoning. To address these issues, we propose ConsiSpace, a geometry-consistency-aware framework for geometry-sensitive video spatial reasoning that turns spatial consistency into both an evidence organization principle and an explicit post-SFT learning signal. We build a geometry-consistent memory (GCM) including implicit evidence tokens and explicit geometric cues, and leverage efficient organization strategies to compactly preserve task-related spatial evidence. Furthermore, we utilize unified consistency self-supervised reinforcement learning (UC-SSRL) after supervised fine-tuning to improve cross-view stability, with answer-, metric-, and topology-consistency rewards. Extensive experiments on three spatial-reasoning benchmarks, VSI-Bench, OSI-Bench, and MMSI-Video-Bench, show consistent gains, improving the average score by 12.6 points over the strongest baselines.