UnifoLM

Architecture

Unitree's open-source unified-large-model series for general-purpose robot learning across multiple embodiments. UnifoLM-WMA-0 (world-model–action): a world model of robot–environment physical interaction, used both as a simulation engine for synthetic data and as a policy-enhancement head that predicts future environmental states (video prediction translated into limb actions). UnifoLM-VLA-0 (vision–language–action): a VLA for general-purpose humanoid manipulation, built on a UnifoLM-VLM base. Open-sourced on Unitree's GitHub (WMA-0 Sept 2025; VLA-0 Jan 2026).

Key facts

Open-source

UnifoLM-WMA-0 (world-model-action) open-sourced Sept 2025; UnifoLM-VLA-0 (vision-language-action, code+weights) Jan 29, 2026 — on Unitree's official GitHub.

Powers (verified)

Unitree G1 integrates UnifoLM-WMA-0 (vision+language+action), per Yicai Global (Sept 16, 2025); G1 is the named real-robot platform for UnifoLM-VLA-0.

Series

UnifoLM-WMA-0 (world-model–action) + UnifoLM-VLA-0 (vision–language–action); designed for multiple robotic embodiments / general-purpose robot learning.

Wiring scope

Wired to G1 only. R1 and H1 left unwired — no primary source names them as UnifoLM platforms (honest absence over inferred presence).

DEPLOY maturity rationale

Open-source research release powering research-tier G1; classified research (brain), not commercial.

Developed by (1)

• Unitree Robotics

Powers (2)

• Unitree G1Unitree Robotics · humanoid
• Unitree R1Unitree Robotics · humanoid

Sources (3)

1. https://github.com/unitreerobotics/unifolm-world-model-action ↗
2. https://github.com/unitreerobotics/unifolm-vla ↗
3. https://www.yicaiglobal.com/news/chinas-unitree-open-sources-world-model-to-advance-robotics-ecosystem ↗ · 2025-09-16