Spatial Harness grounds VLM spatial reasoning in physics-engine states
On MindCube it raises Qwen3.5-27B from 52.70% to 71.14%, and to 98.25% when poses are annotated.
Stories per day · 30 days
7 RSS
Relatedbenchmark4multimodal4vision-language4benchmarks3world-models3qwen2reinforcement-learning2vision-language-models2vlm23d-mapping13d-reasoning1active-exploration1active-perception1agentic-reasoning1coding-agents1depth-generation1egocentric-vision1evaluation1
41 stories, newest first · one card per story
Stories tagged #spatial-reasoning
EvidenceConnect data training produces ZwZ2 multimodal perception gains
Zoomed-region questions plus RLVR improved spatial, long-context, agent, and general visual reasoning without benchmark-specific data.
GeoUG uses generated depth to improve spatial understanding
Reported circular accuracies are 97.9% on CV-Bench-3D, 90.8% on 3DCueBench, and 94.2% on DA-2K.
SATE-Bench tests MLLMs on active exploration of occluded scenes
The benchmark has 316 scenes and 2,525 multi-turn pairs; GRPO increases view-seeking but lowers accuracy.
Paper introduces Ψ-0.6 vision-language world model with streams of thought
The authors say Ψ-0.6 beats leading open-weight models zero-shot on multi-view spatial reasoning and can simulate counterfactual scenes.
SpatialBrain stores spatial-reasoning experience with a fixed backbone
On five benchmarks it beats direct answers from the same backbone on four and ties on the fifth.
S-Agent 2.0 adds spatial tools to general coding agents
With Qwen3.8-27B it scores 66.5 on MMSI-Bench and 82.3 on ReVSI, above SpatialClaw.
Holi-World builds world records and a 4B model from ego video
About 100 scenes yield roughly 80K QA pairs used to train HOLIWORLD-VL-4B, evaluated on HOLIWORLD-Bench.
SpatialScout tests VLM agents on active 3D scene mapping
Claude Opus 5.5 scores 41.71% mean overlap; only 3.25% of targets exceed IoU 0.9.
Look, Reason, Draw a Region: Recursive Refinement for Spatial Context Prediction
Spatial-TRM raises RoboSpatial-Home context accuracy to 29.51% with 11.2M trainable parameters.
Grid-Bench: Can LLMs Predict How Discrete Worlds Change?
Grid-Bench tests whether LLMs can predict full grid states after sequences of spatial operations.
CoWorld Reasoner: Agentic Multimodal Reasoning with the Evolving Video World Model-Driven Harness
CoWorld Reasoner lets a VLM call a video world model as a tool, lifting accuracy on all seven reasoning benchmarks.
RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment
RoTA-Bench shows vision-language models struggle to recover target 3D projections through successive rotations.
When Left Is Right? Camera-Motion Understanding beyond Screen
CameraVLM, a 9B model, judges camera motion better than prior VLMs and matches humans.
When VLMs Destroy What They Know: Spatial Representations Fall Below Lexical Priors
A new benchmark finds vision-language models drop spatial relations from the representations used downstream.
SpatialASM: Internalizing Active Mentalization for Spatial Reasoning
SpatialASM trains vision-language models to plan viewpoints and imagine scenes, improving accuracy on seven spatial benchmarks.
S-Space: Exploring Spatial Workspace in Multimodal Language Models
Researchers identify S-Space, a linear subspace where multimodal models encode object coordinates.
SpatialASM: Internalizing Active Mentalization for Spatial Reasoning
SpatialASM trains vision-language models to plan viewpoints and simulate scenes for spatial reasoning.
SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models
SpaceCast-Bench tests predictive spatial reasoning; the best of 21 models scores 58.0% versus 87.2% for humans.
SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces
SpatialOPSD distills verified spatial coding-agent traces into a tool-free MLLM that beats SFT and GRPO.
SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models
SpaceCast-Bench finds VLMs weak at predictive spatial reasoning; the best model scores 58 percent versus 87.2 percent for humans.
Distilling Routed 3D Privilege for Spatial Reasoning in Vision-Language Models
GPD distills routed 3D geometry into an RGB-only VLM, lifting a 4B model to 57.1 on VSI-Bench.
SpatialCORE: Confidence-Aware Grounded Spatial Reasoning in Large Vision--Language Models
SpatialCORE uses grounding confidence as a reward so vision-language models reason from accurately localized objects.
Soft Spatial Reasoning
Soft Spatial Reasoning lets vision-language models mix token embeddings instead of committing to one spatial token.