Unitree Open-Sources UnifoLM-WLA-1.0 Humanoid Foundation Model

Unitree Robotics has fully open-sourced UnifoLM-WLA-1.0, a 6-billion-parameter vision-language-action foundation model designed for end-to-end humanoid manipulation.

The Hangzhou-based manufacturer said the release targets the software bottleneck that keeps most humanoid robots confined to laboratories, despite its own bipedal production having exceeded 18,000 cumulative units.

The architecture pairs an embodied reasoning backbone, UnifoLM-ER-1-4B, with an MMDiT flow-based action expert trained on approximately 2,500 hours of real-world robot data, including the BitRobot-HIW-500 dataset.

The model unifies desktop manipulation and mobile whole-body coordination into a single policy checkpoint, evaluated across 64 real-robot tasks on the Unitree G1 platform. These include 10 whole-body manipulation tasks such as taking out trash, loading a washing machine, making beds, and shelving inventory, alongside 54 tabletop tasks including folding towels, sorting components from conveyor belts, and plugging in power cords.

According to Unitree, the system supports both basic parallel grippers and two distinct types of multi-finger dexterous hands. Cross-end-effector generalization has historically been a significant barrier for robotics developers, as swapping a two-jaw clamp for a five-fingered hand typically breaks policy representations and requires comprehensive data recollection.

Unitree stated on social media that the release achieves new state-of-the-art results across multiple benchmarks among open-source models worldwide.

The release arrives amid an intensifying strategic debate across the embodied AI landscape, as frontier generalist models increasingly demonstrate zero-shot physical planning and tool use. Hardware advocates argue that algorithms commoditize rapidly, leaving competitive advantages with mass-production platforms and open data flywheels.

The open release follows Unitree's earlier UnifoLM-X2 dynamic world model, which powered autonomous bipedal sparring demonstrations. While X2 prioritized low-latency, contact-rich locomotion, UnifoLM-WLA-1.0 focuses on the industrial and domestic manipulation tasks where humanoid systems have traditionally struggled.