Research map / Vision-language and multimodal / Multimodal LLMs
SpatiO: Adaptive Test-Time Orchestration of Vision-Language Agents for Spatial Reasoning
ChanYeong Hwang, Miso Choi, Sunghyun On, Jinkyu Kim, Jungbeom Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the underperformance penalty delta_i in the reward (ablation) | SpatiO · CV-Bench | accuracy (%): 72.0 -> 87.4 | Improved |
| use Final Reasoning Agent instead of majority voting (ablation) | SpatiO · CV-Bench | accuracy (%): 77.2 -> 87.4 | Improved |
| add TTO reliability weights instead of uniform weights (ablation) | SpatiO · CV-Bench | accuracy (%): 75.9 -> 87.4 | Improved |
| add short-term EMA (ablation) | SpatiO · CV-Bench | accuracy (%): 83.2 -> 87.4 | Improved |
| add long-term EMA (ablation) | SpatiO · CV-Bench | accuracy (%): 81.5 -> 87.4 | Improved |
| add Bayesian trust update (step 3) (ablation) | SpatiO · CV-Bench 500 samples | accuracy (%): 79.66 -> 88.12 | Improved |
| add reward computation only (step 1) (ablation) | SpatiO · CV-Bench 500 samples | accuracy (%): 86.10 -> 73.02 | Mixed |
| use trust-based role assignment instead of fixed (ablation) | SpatiO · 3DSRBench | accuracy (%): 64.8 -> 73.3 | Improved |
| use Top-k=4 instead of 3 agents (ablation) | SpatiO · 3DSRBench | accuracy (%): 73.3 -> 75.1 | Improved |
| use Top-k=5 instead of 3 agents (ablation) | SpatiO · 3DSRBench | accuracy (%): 73.3 -> 69.7 | Hurt |
| replace human role prompts with LLM-generated ones (ablation) | SpatiO · 3DSRBench | accuracy (%): 73.3 -> 71.4 | No clear effect |
| swap 2 of 5 pool agents for alternatives (ablation) | SpatiO · 3DSRBench | accuracy (%): 73.3 -> 70.9 | No clear effect |
Closest papers by meaning
- SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning (NeurIPS 2026 · similarity 0.93)
- Perceive, Interact, Reason: Building Tool-Augmented Visual Agents for Spatial Reasoning (NeurIPS 2026 · similarity 0.92)
- Geometrically-Constrained Agent for Spatial Reasoning (CVPR 2026 · similarity 0.91)
- SpatialScore: Towards Comprehensive Evaluation for Spatial Intelligence (CVPR 2026 · similarity 0.91)
- SpatialLadder: Progressive Training for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.91)
- Visual Agentic AI for Spatial Reasoning with a Dynamic API (CVPR 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
