UnifoLM-VLA-0 large model, developed by Unitree Robotics, assists in the operation of general-purpose humanoid robots.

Unitree Robotics announced the open-sourcing of its large-scale vision-language-action model, UnifoLM-VLA-0. Based on the Qwen2.5-VL-7B architecture, the model was trained on 340 hours of real-device data and integrates 2D/3D spatial perception and dynamic prediction capabilities, overcoming the limitations of traditional VLMs in physical interaction.