UnifoLM-ER-Flow

UnifoLM-ER-Flow is a model released by Unitree Robotics on September 11, 2026, alongside UnifoLM-ER-1 as part of the UnifoLM-WLA-1.0 open-source plan. The weights are hosted on the official unitreerobotics Hugging Face organization under the Apache 2.0 license. Rather than being a simple checkpoint of UnifoLM-ER-1, UnifoLM-ER-Flow extends it with interaction-centric world modeling and discrete action learning: it models how the environment responds to robot actions, enabling flow-based action generation for contact-rich whole-body manipulation. It shares the Qwen3-VL architecture family (Qwen3VLForConditionalGeneration) with UnifoLM-ER-1. Within the UnifoLM-WLA ecosystem, UnifoLM-ER-Flow complements the ER-1 VLA model and the UniBot-V1 Challenge Dataset, forming Unitree's open stack for general-purpose humanoid robot learning. Post-training code and the UnifoLM-WLA-Base model remain planned future releases. Performance figures reported by Unitree are company-reported and have not been independently verified.