Your Physical AI Partner

A Vega 1 Pro or Vega 1 U, assembled and tested, plus the software to program it. That means Python, C++, and Rust APIs, the SDK, and SimReady URDF and USD models of your exact robot. End effectors are ordered separately.

Yes. Vega 1 Pro and Vega 1 U run the same software, so code written for one runs on the other. Vega 1 U also shares Vega's upper-body kinematics, so data collected on a bench carries straight over to the mobile robot.

Nothing is locked in. Train with your own models and pipeline, and test in the simulator you already use. Mount your own end effectors, or DexGripper S and DexGripper D, on the published mechanical and electrical interfaces.

No autonomy is bundled. Position control, safety checks, and the sensors work out of the box. The task behavior is what your team builds on top.

An operator drives the robot in VR or with the open-source exoskeleton kit, in the space where the skill will run. Each session is recorded as training data for imitation learning. The same setup lets a person take over whenever a task needs a human.

It's built for them. We build for research labs, frontier AI labs, developers, and industrial application teams. The hardware is made to hold up to daily use.

Order from our store →, or talk to the team about your use case first. Vega 1 U is the faster start on a bench. Vega 1 Pro adds the mobile base, the adjustable torso, and the full reach.

Published papers from labs including UC Berkeley, Stanford, USC, Carnegie Mellon, UIUC, NVIDIA, Google DeepMind, and Amazon FAR ran on Vega. They are collected under Research Papers, and our own writing is on the Blog.