Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
UnifoLM-WLA-1.0 is a general-purpose humanoid robot foundation model from Unitree Robotics with 6B parameters. It builds on the UnifoLM-ER-Flow multimodal backbone and adds an MMDiT action expert. It combines embodied reasoning, interaction-centric world modeling and action generation in a single vision-language-action representation.
The world-modeling component uses optical flow to extract dynamic regions of a scene and a VQ-VAE to encode them as fixed-length discrete tokens. Conditioned on the current image and a task description or action, the model predicts tokens for future dynamic regions. The embodied-reasoning part, UnifoLM-ER-1, is built on Qwen3-VL-4B and trained on more than 5 million spatial-understanding samples. Unitree reports that it leads open-source models on seven of 16 perception and understanding benchmarks.
The action model is trained on about 2,500 hours of real-robot data, including the Unitree Open Datasets and BitRobot-HIW-500. A single model is reported to coordinate 64 tasks across tabletop and whole-body manipulation. It supports two-finger grippers and several five-finger dexterous hands through a unified action space. According to a YouTube creator, the models and dataset have been released, but no repository or weights links were provided here, so the release details are unverified.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Open-source 6B-parameter robot model covering 64 tasks including whole-body coordination, supporting different grippers and hands. Predicts scene changes and generates actions; models and dataset released, under 9 GB.
Closest neighbours by meaning, not just by category.
Research · BitRobot Foundation
Open dataset of 500+ hours of whole-body teleoperated Unitree G1 humanoid data recorded in real homes.
Model · Xiaomi
Open-weights Vision-Language-Action robot foundation model from Xiaomi, pre-trained on 100K+ hours of UMI trajectories.
Library
Embodied harness from NUS Show Lab that lets vision-language models control robots via discrete semantic action units.
Model · Perceptron
Perceptron's vision-language model for physical AI, covering perception, physical reasoning and robot action prediction.
Model
Action-conditioned robotics world model fine-tuned from Cosmos-Predict2.5-2B using 2D skeleton conditioning across embodiments.
Model · Black Forest Labs
7B open-weights world action model from Black Forest Labs that jointly predicts video and actions for robot control.