roboparty/party_os

Party OS is a project from RoboParty Lab that lays out a roadmap for building software and capabilities for humanoid robots, the kind that walk and interact like a person. Rather than being one single program, it acts as an index page linking together a family of separate repositories that each handle one stage of that roadmap. The plan starts with basic building blocks: collecting and generating movement data, then retargeting motion, meaning translating a human's or another robot's movements onto a different humanoid body shape. From there it moves into imitation learning, where a robot learns by copying recorded movements, and unsupervised learning, where the system learns patterns without labeled examples. The stated long term goal is to build up to more advanced abilities such as human object interaction, human system interaction, vision language action models, and agent like humanoid behavior, so the software climbs from simple tools at the bottom to more complex interactive skills at the top. The README lists four specific sub projects tied to this roadmap. A data generation tool is planned but not yet released. A motion retargeting tool called human-humanoid-tools is already public and converts human or robot movement data so it can be reused on a different humanoid's body. A supervised learning project called MimicLite covers the pipeline for loading movement data, tracking it, training a model, evaluating results, and connecting that to deployment on a real or simulated robot. An unsupervised learning framework called UFO is also linked as a separate repository. There is also a link to an external written knowledge base covering humanoid robot motion control, hosted on Feishu, a Chinese workspace and documentation tool, rather than in this repository itself. This particular repository does not contain runnable code itself. It functions mainly as a map pointing to the other repositories and documentation that make up the full Party OS system, most of which live in separate repos under the same RoboParty account. The project is released under the GNU General Public License version 3.0, a copyleft license, though the README notes that each linked sub project may carry its own separate license, so anyone using a specific piece should check that repo's own terms.