Unitree Robotics open-sources UnifoLM-WLA-1.0, a 6B humanoid robot foundation model
Unitree Robotics released UnifoLM-WLA-1.0, a 6B-parameter humanoid robot foundation model, on GitHub under Apache 2.0. It was trained on about 2,500 hours of real-robot data and coordinates 64 tasks across desktop and whole-body manipulation. The release includes model weights, training code, and datasets.
China context
- Original name
- 宇树科技
- Outside China
- Open weights · github.com
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers can download the model weights and training code from GitHub to fine-tune or adapt UnifoLM-WLA-1.0 for their own robot platforms and tasks.
- For investors
- Unitree's open-source strategy may accelerate adoption of its humanoid robots and create an ecosystem around its foundation model, increasing demand for its hardware.
Translated from Chinese. Quotes and facts link to the original sources.
Unitree Robotics open-sourced UnifoLM-WLA-1.0, a 6B-parameter general-purpose humanoid robot foundation model. The model is built on large-scale multimodal perception data and interaction-centric world modeling, achieving leading results on multiple embodied reasoning benchmarks. It supports two-finger grippers and multiple five-finger dexterous hands, with strong generalization across tasks and end effectors. The repository includes model weights, training and fine-tuning code, and three datasets: UniBot-V1 Challenge Dataset, UnifoLM-WBT-Dataset, and UnifoLM-Dex1-Dataset. The project is released under Apache License 2.0.
UnifoLM-WLA-1.0 is a 6B-parameter model trained on approximately 2,500 hours of real-robot data. It coordinates 64 tasks spanning desktop manipulation and whole-body manipulation. The model supports two-finger grippers and multiple five-finger dexterous hands. The repository provides code for training action experts, fine-tuning, and LoRA fine-tuning. It builds upon starVLA and Qwen-Image.
Robotics developers outside China can now access a production-grade humanoid foundation model and its training code, reducing the cost and time to build manipulation capabilities for their own robots. Unitree's open-source release puts pressure on other humanoid robot makers to match its data scale and task coverage or risk falling behind in developer adoption.
The open-source release lowers the barrier for robotics startups and researchers to build on a state-of-the-art humanoid foundation model, accelerating product development. For Unitree, it positions the company as a leader in embodied AI and may drive adoption of its hardware platforms.
Observable next signals include whether Unitree releases additional model variants or datasets, whether third-party developers fine-tune UnifoLM-WLA-1.0 for new tasks, and whether other robotics companies respond with their own open-source foundation models. The model's performance on independent benchmarks outside Unitree's reported results remains to be verified.