✦
Summary
UnifoLM-WLA-1.0 is Unitree Robotics general-purpose humanoid foundation model for coordinated manipulation. A single six-billion-parameter model handles desktop and whole-body tasks, with support for two-finger grippers and several five-finger hands, aiming to transfer capabilities across tasks and end effectors.
Key Features
The general-purpose humanoid model has approximately 6B parameters.
One model coordinates 64 desktop and whole-body manipulation tasks.
Demonstrations include two-finger grippers and multiple five-finger dexterous hands.
UnifoLM-ER-1 builds on the Qwen3-VL-4B vision-language backbone.
The reported training set contains approximately 2,500 hours of real-robot data.
Embodied reasoning training includes more than five million samples.
Dynamic-region prediction helps represent interaction-relevant future changes.
The project links code, model releases, and datasets.
About UnifoLM-WLA-1.0
The system combines embodied reasoning, predictions of dynamic visual regions, and action learning. Its embodied reasoner builds on Qwen3-VL-4B and more than five million reasoning samples. World-language-action training then draws on approximately 2,500 hours of real robot interaction data.
The project is useful for researchers developing unified humanoid policies that connect spatial understanding with execution. It demonstrates 64 tasks and supplies linked code, models, and datasets. Actual deployment requires compatible robot interfaces and the task-specific setup described by the release.