RoboParty open-sources a whole humanoid, from mechanical drawings to motion algorithms

Humanoid robotics company RoboParty has released RP1 as a full-stack open-source robot. QbitAI's October 27 report stresses that the release goes beyond an SDK or partial algorithm code: mechanical drawings, the electrical system, in-house joints, low-level control, simulation, training and deployment are all handed to developers.

The report draws a clear line between degrees of openness. With only a high-level API, a researcher can issue commands and tune parameters; with control algorithms open, they can study how motion is generated and executed; only when drawings, electrical schematics, joint documentation, embedded software and simulation tooling are all available does the design space really open. RP1 aims at that last layer.

The robot stands 135cm tall and comes in configurations with 24 to 30 degrees of freedom, aimed at education and research, data collection, guided tours and entertainment. Motion is driven by RoboParty's in-house ROMOMO joint modules plus a companion control and training stack; the demonstrations show continuous walking, human motion tracking, recovery from perturbations and getting up after a fall.

For research groups, how deep the openness goes determines how much autonomy they have. The report describes a familiar problem: a robot walks perfectly in a paper and falls over on a different unit. Joint layout, device protocols, control frequency and even the zero-position definition — which physical position counts as zero degrees — can all change experimental results. Without the underlying material, teams burn time adapting to a platform, and some questions get bounded by the hardware.

The deeper issue is diagnosis. With only a high-level interface, a disappointing experiment is hard to attribute: was the policy undertrained, did the joints lag, or does the simulator diverge from the real machine? Layer after layer of black box sits in between. Full-stack openness flattens that chain.

“Open source” has been used loosely in humanoids, with most projects stopping at model weights, an SDK or part of the algorithm stack. Publishing the mechanical structure and electrical system together accepts being copied and modified; for projects like this, the durable advantage usually sits in joint modules, tooling and engineering support rather than the drawings.

What to watch: the exact license terms, how complete the drawings and code actually are, and whether the community can reproduce the demonstrated motion on real hardware. For a humanoid, downloading is only step one; running stably outside the lab is the real test of full-stack open source.