RoboParty Open-Sources Its RP1 Humanoid, Hardware Files Included

Open-source hardware has mostly stayed at the component level in humanoid robotics: a motor design here, a controller board there, rarely a complete robot. RoboParty, a China-linked open-source humanoid robotics developer community, is pushing further, releasing a full development stack for its RP1 humanoid, known in its research papers as ROBOTO 01, that spans mechanical design files down to the motion-learning policies that make the robot walk. The release includes STEP models, drawings, PCB production files, wiring-harness information and bills of materials for the hardware, alongside control software, robot models and motion-learning resources for the software side, following RP1's debut at the IEEE/RSJ International Conference on Intelligent Robots and Systems, known as IROS.

RoboParty's community presence, documentation largely in Chinese and links to QQ, WeChat and Bilibili, points to a China-based developer base, though no formal headquarters or corporate registration is disclosed in the project's public materials. The project's niche is specific and narrow: rather than selling a finished consumer or industrial humanoid, RoboParty is positioning RP1 as a reference platform other developers, researchers and small robotics teams can build on, modify and extend, competing less with commercial humanoid makers like Unitree or Agility Robotics and more with academic and hobbyist robotics platforms that have historically lacked a credible, full-stack humanoid option.

Why Releasing the Whole Stack Matters More Than Releasing Code

Most open-source robotics releases so far have been partial by necessity. A company might open-source its control software while keeping hardware designs proprietary, protecting the manufacturing side of the business while still building developer goodwill. Or a hardware project might publish mechanical drawings without the motion-learning policies needed to actually make a humanoid walk reliably, leaving builders with a static robot and no practical way to get it moving without substantial independent software development. RoboParty's release crosses that boundary deliberately, publishing hardware files, control software and trained motion policies together, for two distinct hardware configurations, a 24-degree-of-freedom build and a 28-degree-of-freedom build.

The project's own stated rationale, drawn from its GitHub documentation, is a diagnostic argument rather than a purely ideological one: a motion failure in a humanoid robot can originate in the learned control policy, in the physical actuator hardware, or in the mechanical structure itself, and a developer who only has access to one of those three layers cannot reliably determine which layer actually caused a given failure. By releasing all three together, RoboParty is betting that giving developers the ability to isolate and test each layer independently will accelerate debugging and iteration across the broader open humanoid-robotics research community, not just for RoboParty's own platform.

What RP1 Can Actually Do, and the One Chart That Does Not Apply

RoboParty states that RP1 has demonstrated continuous walking, recovery from physical disturbance, the ability to get up after a fall, and human-motion tracking, where the robot mirrors a human operator's movements. Those are the standard capability benchmarks the humanoid-robotics field has converged on as proof that a platform's basic locomotion and balance control work, rather than evidence of any advanced manipulation or task-completion skill. RoboParty has not disclosed a retail price or a shipping schedule for RP1 as hardware, which suggests the current release is aimed squarely at the developer and research community capable of assembling a unit from the published design files, rather than at a consumer or enterprise buyer looking for a finished, supported product.

One detail in the release is worth flagging precisely because it is easy to misread. A MimicLite comparison plot included in RoboParty's materials evaluates a Unitree G1 robot, not RP1 itself, and should not be interpreted as a performance claim about RoboParty's own hardware. Readers and potential builders comparing RP1 against established commercial humanoids should treat that chart as a benchmark reference for the comparison methodology, not as data demonstrating how RP1 performs against a G1 or any other platform. RoboParty's own public materials do not currently include a head-to-head performance comparison of RP1 against competing humanoid robots.

The Licensing Details That Decide Who Can Actually Build This

RoboParty's hardware designs are released under CERN-OHL-W-2.0, the CERN Open Hardware Licence's weakly reciprocal variant, a license commonly used for open electronics and mechanical hardware that requires anyone who modifies and distributes the design to release those modifications under compatible terms, while still permitting commercial use. The accompanying hardware documentation carries a Creative Commons Attribution-ShareAlike 4.0 license, which requires attribution and requires derivative documentation to carry the same license. RoboParty's own footer notes that other repositories bundled into the broader RP1 project, which draw on academic research code tied to specific published papers, carry their own separate licensing terms, meaning a developer building a full RP1 system needs to check license compatibility across multiple components rather than assuming one blanket license covers the entire stack.

That licensing structure is permissive enough to allow commercial derivatives, which is the detail most likely to interest small robotics companies and research labs evaluating RP1 as a starting point rather than a hobby project. A startup that wanted to build a commercial humanoid variant on top of RP1's mechanical and electronic design would not be blocked from doing so by the hardware license itself, though it would need to clear the separate academic-code licenses attached to some of the motion-learning components before shipping a derivative product.

The Research Lineage Behind the Motion Policies

The motion-learning components RoboParty bundled into the release trace back to a cluster of academic papers on humanoid locomotion and manipulation, including work described in the project's documentation as P-BFM, I-BFM, PredActor, a table-tennis manipulation paper, INTACT and VIStream, with contributing researchers including Xue Wang, Yitang Li, Hao Zhao, Cewu Lu and several co-authors across the different papers. That research lineage matters for credibility: RP1's motion policies are not a from-scratch implementation built solely for this release, but an aggregation of published, peer-reviewed robotics research, which gives outside developers a documented technical basis to evaluate rather than an unverified black-box control system.

How RP1 Compares With Other Open-Source Humanoid Efforts

RoboParty's release follows a broader pattern this year of commercial humanoid makers open-sourcing parts of their own stack, typically as a strategic move to build developer ecosystems around a platform they still sell as finished hardware. Unitree has published aspects of its own software tooling while continuing to manufacture and sell its G1 and other humanoid lines as commercial products; several other robotics companies have released simulation environments or training datasets without releasing the underlying hardware designs needed to actually build a physical unit. RoboParty's release differs in both direction and completeness: it is not a commercial hardware company cracking open part of its stack to build goodwill around a product it still sells, but a developer-community project releasing the entire buildable design, mechanical, electronic and software, with no commercial hardware product of its own currently for sale.

That distinction matters for who the release actually serves. A commercial maker's partial open-source release is aimed at developers who will buy the company's hardware and build software on top of it, reinforcing a sales relationship. RoboParty's full-stack release removes that commercial dependency entirely: a well-resourced university lab, a small robotics startup, or an individual hobbyist with access to a machine shop and electronics assembly capability could, in principle, build a working RP1 unit without ever purchasing anything from RoboParty, since the project currently has no hardware product to sell. Whether that openness accelerates humanoid robotics research broadly, by letting more teams experiment without needing a large capital outlay for a commercial platform, or simply remains a niche resource for well-equipped labs, will depend on how many outside teams actually attempt a full build rather than just reading the documentation.

What to Watch as the Project Matures

The next marker worth tracking is whether independent developers outside RoboParty's own community successfully build and operate an RP1 unit from the published design files, which would be the first real test of whether the release is complete and accurate enough to replicate, as opposed to a partial reference that still requires undocumented tribal knowledge to actually assemble. A second marker is whether RoboParty or a downstream developer ships any commercial derivative product built on the RP1 hardware license, which would indicate the open-source release successfully seeded a new business rather than remaining confined to academic and hobbyist use. Until either happens, RP1 should be read as a credible, unusually complete open-source humanoid reference platform, not yet as evidence of a broader shift in how humanoid robots reach the market.

This account synthesizes Humanoids Daily's October 15, 2026 reporting on RoboParty's RP1 release, cross-referenced against RoboParty's own GitHub documentation and public announcement, as of the stated publication date.

This article is provided for general information purposes only and does not constitute investment, procurement, legal, or engineering advice. Figures reflect the sources listed above as of the stated information cut-off and may change without notice.