---
pipeline_tag: robotics
license: apache-2.0
---

# WSA: A 3D-Centric World-Spatial-Action Model for Generalizable Robot Control

This repository contains the weights for **WSA-Base**, a 3D-centric World-Spatial-Action embodied foundation model for generalizable robot control. 

- **Paper:** [WSA$_1$: a 3D-Centric World-Spatial-Action Model for Generalizable Robot Control](https://huggingface.co/papers/2607.03941)
- **Project Page:** [https://zaleni.github.io/WSA1/](https://zaleni.github.io/WSA1/)
- **Repository:** [https://github.com/zaleni/WSA](https://github.com/zaleni/WSA)

## Overview

WSA (World-Spatial-Action) is an embodied foundation model that unifies instruction-aligned 2D visual planning, action-conditioned 3D world modeling, and 3D-aware action generation into a shared latent space. WSA-Base utilizes a Qwen3-VL-2B backbone and is pre-trained on diverse datasets to enable data-efficient, generalizable robot manipulation.

For details on how to set up the environment, run inference, or perform downstream fine-tuning, please refer to the [official GitHub repository](https://github.com/zaleni/WSA).