open source

Research code and resources.

Research implementations

  • VEGA-3D — Implicit 3D priors from video generation models for spatial understanding and embodied tasks. Project page.
  • TurboVLA — A compact vision-language-action model for efficient robotic manipulation. Project page.
  • GaugeAlign — The winning solution for AI City Challenge 2026, Track 1. Inference-time coordinate canonicalization for multi-camera 3D perception.

Community resources

Awesome World Action Model
A curated collection of papers and resources on vision-language-action models and world action models.