A RoboParty Lab team member at RoboParty discusses building humanoid parkour motions with Party OS and the lab’s approach to physical AI research.
The presenter describes humanoid parkour as the goal and identifies data as the first major challenge. RoboParty Lab’s Party OS includes H H Tools, an open-source tool for processing data for motion retargeting.
For a controller that handles fast movements and teleoperation, the presenter uses Mimic Light, a supervised learning framework in Party OS’s open-source BFM base. The framework is presented as a way to train tracking behavior.
The presenter says supervised methods can struggle when a robot encounters situations outside its training data. For stronger recovery from disturbances, they point to UFO, an unsupervised framework that supports teleoperation and dynamic movement.
RoboParty Lab frames real-world readiness as a cycle of building, breaking, and fixing robots. The lab aims to give researchers access to robots, datasets, data processing, sim-to-real work, and ways to present and share results, so they can focus on research or open-source contributions.
The presenter describes the lab as a group focused on doing meaningful work. Its stated aim is to let people bring their ideas and abilities to the platform.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
TAKASU Masakazu, host of Nico-Tech Shenzhen, reports from ROSCon China 2026 in Shenzhen on the robotics demos, developer tools, and embodied AI themes across the event.
Professor A. Fernando Ribeiro, associate professor at Universidade do Minho and RoboCup trustee, recalls Gerhard K. Kraetzschmar’s leadership and friendship in RoboCup and how he inspired young people through engineering outreach in Portugal and Leipzig.
