OVO-S-Bench: A Hierarchical Benchmark for Streaming Spatial Intelligence in Multimodal LLMs
2026-06-02 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors created OVO-S-Bench, a new test for AI models that need to understand places and layouts from ongoing first-person video streams. The benchmark includes 1,680 carefully checked questions that test different skills like seeing what's immediately around, keeping track of objects over time, imagining spaces, and making maps from the video. They tested many large language models and found even the best ones struggle most with making maps from this kind of data. The authors also show that some reasoning methods can make mistakes worse if the model doesn’t directly use the video stream. This work highlights current challenges in teaching AI to understand spaces in real time.
multimodal agentsegocentric streamsspatial intelligenceallocentric mappingchain-of-thought reasoninglarge language models (LLMs)quality assurancespatiotemporal contextstreaming evaluationspatial simulation
Authors
Yifei Li, Pengyiang Liu, Yuhang Zang, Zhongyue Shi, Qi Fu, Hongye Hao, Jiwen Lu
Abstract
Multimodal agents in robotics, AR, and autonomous driving must reason about places and layouts from continuous egocentric streams, often using evidence outside the current view. Existing benchmarks either evaluate offline over full videos or target events rather than spatial structure. We introduce OVO-S-Bench, a fully human-annotated benchmark for streaming spatial intelligence, comprising 1,680 questions over 348 source videos. Annotation involves 12 trained annotators, each also serving as a blind cross-reviewer, across roughly 804 person-hours of multi-round quality assurance. Each question carries a query timestamp and an evidence interval, and at evaluation, the model sees only the prefix preceding the query. Questions span four levels of increasing abstraction: instantaneous egocentric perception, spatiotemporal context tracking, spatial simulation and reasoning, and allocentric mapping. Across 38 proprietary and open-source MLLMs, Gemini-3.1-Pro trails human experts by 27 points, 59.2 vs. 86.6, with allocentric mapping as the dominant bottleneck. Notably, streaming and spatially fine-tuned MLLMs underperform their own backbones. We further find that chain-of-thought reasoning amplifies spatial errors when ungrounded in the stream. By exposing these limitations, OVO-S-Bench establishes a demanding testbed for next-generation streaming spatial MLLMs.