Chinese robotics startup RoboParty has unveiled RP1, a bipedal humanoid it describes as the world’s first high-performance, full-stack open-source robot of its kind.
The machine debuted at the IROS 2026 conference in Pittsburgh on September 28, ahead of a wider October 3 announcement. RoboParty is aimed at researchers, educators, robotics developers and embodied AI teams, and promises a system they can use, modify, train and extend in real workflows.
At its booth, RoboParty ran a “Kick Me” demonstration in which visitors pushed and kicked the robot, which then adjusted its posture and recovered balance. The company reports peak joint torque of up to 160 Nm. The motion control runs on PartyOS, RoboParty’s open R&D foundation, which integrates a training framework called UFO for discovering humanoid motor skills through unsupervised reinforcement learning.
It builds on RPO, or Roboto Origin, RoboParty’s first open-source humanoid project, released in January and now carrying more than 2,500 GitHub stars. The company plans to progressively release motion control, simulation environments, SDKs and training tools, with broader hardware rollouts and a mass-production program later in the fourth quarter. Pricing has not been disclosed.
The bet is that open hardware can accelerate embodied AI the way open software sped up machine learning – though whether outside teams can reproduce the IROS demos will be the real measure.
