UnifoLM-WLA-1.0
$ 0
UnifoLM-WLA-1.0 is Unitree’s open foundation model for humanoid robots. One set of weights runs 64 tabletop and whole-body tasks on the Unitree G1.
Available on backorder
Brain Score
3
Specifications and details:
| Nationality | China |
|---|---|
| Website | https://www.unitree.com/ |
| Model type | Foundation Model |
| Manufacturer | Unitree Robotics |
| Release date | September 10, 2026 |
Description
UnifoLM-WLA-1.0 is a general-purpose foundation model for humanoid robots from Unitree Robotics. Unitree unveiled it in September 2026 as the newest member of its open UnifoLM family. The model has about 6 billion parameters, and one set of weights handles 64 different tasks. It also blends three jobs in one system: it reasons about the scene, predicts which parts will move and then outputs actions. That mix gives it a light form of world modeling without full video generation. As a result, it marks Unitree’s biggest step yet from hardware maker toward robot brain builder.
The team tested the model on the Unitree G1, with 54 tabletop tasks and 10 whole-body tasks. For example, the robot can use its legs and torso while it handles objects. In addition, the model works with both simple grippers and several five-fingered hands. Unitree trained it on about 2,500 hours of real robot data, including its own open datasets. Developers can now download the base weights from Hugging Face under the Apache 2.0 license. Meanwhile, Unitree still lists the post-training code and two more datasets as coming soon.
New! 2026 Humanoid
Robot Market Report
198 pages of exclusive insight from global robotics experts — uncover funding trends, technology challenges, leading manufacturers, supply chain shifts, and surveys and forecasts on future humanoid applications.
now Google DeepMind
Contact Humanoid.guide
Website: https://unigen-x.github.io/unifolm-wla.github.io/
