taxonomy · benchmark · Ability Study

SpatialTree : How Spatial Abilities
Branch Out in MLLMs

Yuxi Xiao▲,★,*, Longfei Li♦,★,*, Shen Yan★, Xinhang Liu★, Sida Peng▲,
Yunchao Wei♦, Xiaowei Zhou▲, Bingyi Kang★,†

▲Zhejiang University · ★ByteDance Seed · ♦Beijing Jiaotong University

*Equal Contribution †Project Lead

arXiv 📄 PDF 🏆 Leaderboard
SpatialTree organizes spatial abilities into L1–L4, linking perception to agentic competence.

How does spatial intelligence emerge?

L1: low-level perception

L1 Perception

L2: mental mapping

L2 Mental Mapping

L3: mental simulation

L3 Mental Simulation

L4: agentic competence

L4 Spatial Agent

“Each stage builds on the stage before it.”

— Jean Piaget’s cognitive development theory

Overview

TL;DR

We propose SpatialTree, a cognitive-science-inspired taxonomy of spatial intelligence in multimodal large language models(MLLMs), along with SpatialTree-Bench and a proof-of-concept study on improving spatial abilities.

Takeaways
  • Dependencies Structure: Low-level abilities are relatively orthogonal, whereas higher-level abilities are increasingly compositional and coupled.
  • Ability Acquisition: Exhaustive supervised fine-tuning (SFT) over a broad range of low-level perceptual data can maximize cross-level transfer, benefiting higher-level spatial abilities.
  • Auto-thinking: Naive post-training with RLVR does not work; explicitly encouraging reasoning at higher levels while suppressing it for low-level perceptual skills makes broad improvements.

Cognitive Hierarchy

We design the SpatialTree taxonomy, inspired by cognitive science, to scaffold spatial abilities across four hierarchical levels.

Click a layer to see its focus, representative abilities.

L4 · Agentic Competence

Converts perception + memory into executable behaviors through the Spatial Action Mapping, covering goal execution, open-world exploration, and manipulation.

Data Samples

SpatialTree-Bench

SpatialTree-Bench measures 16 closed/open MLLMs across L1–L4 with balanced weighting, revealing orthogonal perception atoms and tightly coupled higher layers.

Distribution of Benchmark Categories

Closed-source Open-source
Model Avg. Perception Mental Mapping Mental Simulation Agentic Competence
Gemini 2.5 Pro 50.1 49.3 53.8 51.3 45.8
Gemini 2.5 Pro NT 41.4 38.8 46.9 42.4 37.6
Qwen3VL-235B 40.0 34.6 46.6 39.9 38.9
Gemini 2.5 Flash 39.0 38.3 46.3 37.8 33.5
GLM4.5V 36.0 35.0 34.4 36.3 38.3
Gemini 2.5 Flash NT 35.8 33.5 41.6 31.4 36.7
Seed1.6-Vision 35.7 36.3 36.6 35.4 34.6
Qwen3VL-30B 35.3 30.2 40.9 36.7 33.4
Qwen 2.5VL-72B 33.0 28.8 37.8 34.6 31.3
GPT-4o 31.9 26.8 34.9 33.2 32.5
Qwen 2.5VL-32B 27.9 24.0 26.8 34.3 26.4
Qwen 2.5VL-7B 27.5 20.9 28.9 32.4 27.8
Kimi-VL-A3B 24.4 20.3 25.9 27.2 24.2

Hierarchy insight: L1 abilities are largely orthogonal, while L3 and L4 scores are strongly correlated.

Ability Study

Dependency visualization

Ability Dependency Analysis

(Region A + Region B): Low-level abilities are largely orthogonal, while high-level abilities are tightly coupled through integration.

(Region C): Low-level abilities form clear prerequisites that higher-level abilities depend on.

Finding 1

Cross-Ability Transfer: Single-ability L1 SFT induces cross-level transfer, while yielding limited or slightly negative effects on same-level abilities.

Ability transfer visualization

Finding 2

Multi-ability Synergy: The holistic integration across multiple fundamental abilities achieves synergistic gains far exceeding their individual effects.

Methods Avg. L1 Perception L2 Mental Mapping L3 Mental Simulation L4 Agentic Competence
Geom. Motion Rel. Local. Orient. Underst. Memory Caus. Reas. Seq. Plan. Goal Exec. Open Expl.
Baseline 25.0 20.9 28.6 28.9 24.2 34.2 22.6 21.7 27.2 31.7 22.1 26.5
B+Dist. 24.5 24.1 +3.2 26.6 -2.0 23.2 -5.8 19.6 -4.6 34.3 +0.1 24.6 +2.0 21.8 +0.1 26.1 -1.1 30.8 -0.9 25.5 +3.4 26.1 -0.4
B+Corr. 25.2 17.6 -3.2 23.9 -4.7 30.2 +1.3 18.9 -5.3 35.6 +1.4 21.9 -0.7 24.6 +2.9 21.8 -5.4 33.9 +2.2 24.7 +2.6 35.9 +9.4
B+Size 23.5 24.3 +3.4 22.6 -6.0 21.4 -7.5 21.7 -2.5 34.5 +0.3 21.9 -0.8 19.2 -2.5 23.4 -3.8 30.3 -1.5 21.5 -0.6 24.3 +2.2
B+Dist.+Size+Corr. 26.1 25.5 +4.6 29.3 +0.7 29.4 +0.5 16.4 -7.8 33.7 +0.5 23.0 +0.4 24.2 +2.5 25.2 -2.0 34.2 +2.5 26.0 +3.9 28.5 +2.0
B+Dist.+Size+Corr.+Mot. 27.3 28.6 +7.7 24.6 -4.0 20.6 -8.3 26.3 +2.1 36.0 +1.8 22.2 -0.4 22.6 +0.9 28.2 +1.0 32.8 +1.1 23.3 +1.1 35.9 +9.4
Baseline+75@(all spat.) 23.6 24.9 +4.0 22.6 -6.0 25.9 -3.0 17.4 -6.8 31.2 -3.0 22.2 -0.4 20.6 -1.1 25.7 -1.5 30.2 -1.5 19.7 -2.4 22.8 -3.7

SFT Comparisons. "B+Dist.", "B+Corr.", and "B+Size" denote the baseline augmented with distance, correspondence, and size tuning data, respectively. Changes are color-coded as notable gains, neutral influence, and drops.

Finding 3

Auto-Think Strategy: We uncover a trade-off in RL where excessive reasoning hurts intuitive perception. We propose a hierarchy-aware "Auto-think" strategy that suppresses reasoning for low-level sensing while encouraging it for complex planning.

Methods Avg. L1 Perception L2 Mental Mapping L3 Mental Simulation L4 Agentic Competence
Geom. Motion Rel. Local. Orient. Underst. Memory Caus. Reas. Seq. Plan. Goal Exec. Open Expl.
SFT Baseline 27.3 28.624.620.626.336.0 22.222.6 28.232.8 23.335.9
L1 RL@think 26.6 -0.7 28.4 +0.9 30.0 +5.4 30.5 +9.9 19.5 -6.8 34.5 -1.4 24.9 +2.8 18.5 -4.1 25.7 -2.4 34.0 +1.1 24.1 +0.8 29.6 -6.3
L2 RL@think 26.7 -0.5 24.4 -4.1 22.7 -1.9 22.3 +1.6 17.0 -9.4 35.0 -0.9 26.6 +4.5 16.1 -6.5 29.0 +0.8 31.4 -1.4 25.6 +2.3 34.5 -1.4
L3 RL@think 27.7 +0.4 16.2 -12.3 24.0 -0.6 24.1 +3.5 21.4 -4.9 38.5 +2.5 26.0 +3.9 21.8 -0.8 31.3 +3.1 34.7 +1.9 26.5 +3.2 38.4 +2.5
L4 RL@think 28.5 +1.2 23.8 -4.8 25.3 +0.7 22.1 +1.4 23.5 -2.8 33.9 -2.1 25.1 +3.0 20.5 -2.2 32.0 +3.8 34.2 +1.4 27.1 +3.9 38.8 +2.9
Full RL@think 30.1 +2.9 29.7 +1.1 24.7 +0.1 27.2 +6.5 21.0 -5.3 34.8 -1.2 27.4 +5.2 16.7 -5.9 33.6 +5.5 37.6 +4.8 25.4 +2.1 41.7 +5.8
Full RL@auto-think 30.8 +3.6 31.9 +3.3 28.6 +4.0 22.0 +1.3 23.1 -3.2 36.8 +0.8 28.0 +5.8 22.6 -0.1 33.5 +5.4 35.6 +2.8 23.4 +0.1 44.1 +8.3

RLVR Comparisons. The table compares the SFT baseline with versions enhanced by RL on different tasks. Changes are color-coded as notable gains, neutral influence, and drops.

Citation

If you find our project helpful to your research, you can cite us with:

@article{spatialtree2025,
  title={SpatialTree: How Spatial Abilities Branch Out in MLLMs},
  author={Xiao, Yuxi and Li, Longfei and Yan, Shen and Liu, Xinhang and Peng, Sida and Wei, Yunchao and Zhou, Xiaowei and Kang, Bingyi},
  journal={arXiv preprint},
  year={2025}
}