DEPLOYDatabase

Brain

SpatialVLA

VLA with Ego3D position encoding and adaptive action grids, pretrained on 1.1M real robot episodes. Multi-institutional research effort demonstrating spatial reasoning for cross-embodiment manipulation.

Foundation model · Maturity: Research · Open source

Machine-readable surfaces

Architecture

VLM + 3D spatial position encoding, adaptive action grids

Key facts

SpatialVLA design (primary project page)
The SpatialVLA project page describes a spatial-enhanced vision-language-action model trained on 1.1 million real robot episodes, with Ego3D Position Encoding and Adaptive Spatial Grids (also called Adaptive Action Grids). Authors are affiliated with Shanghai AI Laboratory, Fudan, Shanghai Jiao Tong, Zhejiang University, ShanghaiTech, and Northwestern Polytechnical University. Treat dataset size as author-stated (SpatialVLA)
Evaluation scope (primary project page)
The page says SpatialVLA was evaluated across 7 robot learning scenarios, 16 real-robot tasks, and 48 simulation setups, focusing on zero-shot control, spatial understanding, and adaptation to new setups, and claims faster inference with fewer tokens per action. Treat SOTA and speed claims as author-reported (SpatialVLA)
Official code and model release
The SpatialVLA repository provides the project code and links to model resources for the spatial vision language action system (SpatialVLA repository)
SpatialVLA research paper
The SpatialVLA paper describes Ego3D position encoding and adaptive action grids as design components for spatially informed vision language action control (SpatialVLA paper)
Spatial action representation
SpatialVLA combines Ego3D Position Encoding with Adaptive Spatial or Action Grids to represent spatial information for vision-language-action control (SpatialVLA).
Cross-embodiment resources
The official SpatialVLA repository provides code and model resources for evaluating the spatial VLA across robot setups (SpatialVLA repository).

Common questions

What is SpatialVLA?
VLA with Ego3D position encoding and adaptive action grids, pretrained on 1.1M real robot episodes. Multi-institutional research effort demonstrating spatial reasoning for cross-embodiment manipulation.
Is SpatialVLA open source?
Yes. SpatialVLA is recorded as open-source / open-weights on the DEPLOY registry, meaning model weights or source code are publicly available.
What type of AI is SpatialVLA?
SpatialVLA is a foundation model, built on a VLM + 3D spatial position encoding, adaptive action grids architecture on the DEPLOY registry.
What is SpatialVLA's maturity stage?
SpatialVLA is at the research stage on the DEPLOY maturity ladder. Research stage means active development without commercial deployments on file.
Which robots run on SpatialVLA?
No robot models on the DEPLOY registry are recorded as running SpatialVLA. DEPLOY wires brain-to-model connections only when the wiring is verifiable from primary sources; absence may reflect pre-deployment or unverified manufacturer claims.

Sources (3)

  1. https://spatialvla.github.io/
  2. https://github.com/SpatialVLA/SpatialVLA
  3. https://arxiv.org/html/2501.15830v5
Methodology: Verified · 3 sources (no primary) · last reviewed 2026-10-09

Verification posture

Verified

Low confidence

Review state

Stable

Last reviewed 2026-10-09

Maturity + lifecycle

Maturity stage: research

Sources by quality tier

1
unclassified
Unclassified source
1
code-repository
Code repository
1
preprint
Preprint

The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.

Methodology surface for SpatialVLA.

Canonical ID 5f5bb142-a08e-4f41-a576-8243ec4659bd