{"slug": "thinking-from-above-efficient-spatial-reasoning-with-bird-s-eye-view-visual", "title": "Thinking from Above: Efficient Spatial Reasoning with Bird's-Eye-View Visual Tokens", "summary": "Cambrian-BEV, a spatial reasoning method that back-projects visual patch features from video frames into a shared bird's-eye-view coordinate grid, reaches 63.48 on the VSI-Bench spatial reasoning benchmark, 8.5 points above an identically fine-tuned frame-token control, according to the arXiv paper 2610.22700v1. The parameter-free geometric tokenization compresses 128 frames containing roughly 100K DINOv3 patch features into a fixed 27×27 grid of 729 BEV tokens, cutting LLM prefill time by 2.9× and peak memory by 3.6× while using 12.6× fewer visual tokens. The method, which needs no architectural change beyond BEV tokenization and keeps a frozen DINOv3 vision encoder, gains 24.5 points on room size and 34.1 on relative direction, and scores 44.68 on ReVSI, 4.5 above its control.", "body_md": "arXiv:2610.22700v1 Announce Type: new \nAbstract: Existing video multimodal large language models (MLLMs) organize observations primarily in time, but embodied spatial reasoning requires organizing them in space. Consequently, their visual representations can remain locally consistent across nearby frames while lacking a persistent spatial context that is stable across viewpoints and time. We exploit depth and camera pose, which are naturally available to embodied agents, to explicitly establish this spatial correspondence. We introduce Cambrian-BEV, which extends Cambrian-S with a parameter-free geometric tokenization scheme that back-projects visual patch features from each frame into a shared bird's-eye-view (BEV) coordinate frame and averages features that fall within the same spatial cell. Thus, observations of the same physical location, regardless of viewpoint or time step, are aggregated into a common spatial token. In addition to providing a persistent scene representation, this formulation acts as an efficient compression mechanism: 128 frames containing approximately 100K DINOv3 patch features are consolidated into a fixed 27$\\times$27 grid of only 729 BEV tokens, cutting LLM prefill time by $2.9\\times$ and peak memory by $3.6\\times$. Cambrian-BEV requires no architectural modification beyond BEV tokenization and uses a frozen DINOv3 vision encoder with a trainable multimodal projector and LLM. On VSI-Bench, our model reaches 63.48, $+8.5$ points over an identically fine-tuned frame-token control with $12.6\\times$ fewer visual tokens, with particularly large gains on room size ($+24.5$) and relative direction ($+34.1$). On ReVSI, it reaches 44.68, $+4.5$ over the corresponding control.", "url": "https://wpnews.pro/news/thinking-from-above-efficient-spatial-reasoning-with-bird-s-eye-view-visual", "canonical_source": "https://arxiv.org/abs/2610.22700", "published_at": "2026-10-19 04:00:00+00:00", "updated_at": "2026-10-19 04:22:05.488185+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "computer-vision", "ai-research", "large-language-models"], "entities": ["Cambrian-BEV", "Cambrian-S", "DINOv3", "VSI-Bench", "ReVSI", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/thinking-from-above-efficient-spatial-reasoning-with-bird-s-eye-view-visual", "markdown": "https://wpnews.pro/news/thinking-from-above-efficient-spatial-reasoning-with-bird-s-eye-view-visual.md", "text": "https://wpnews.pro/news/thinking-from-above-efficient-spatial-reasoning-with-bird-s-eye-view-visual.txt", "jsonld": "https://wpnews.pro/news/thinking-from-above-efficient-spatial-reasoning-with-bird-s-eye-view-visual.jsonld"}}