How Huang Yi Turned AlexBot Into RoboParty’s Open Humanoid Platform

RoboParty’s story is less about a dorm-room robot becoming a conventional hardware startup than about turning openness into a product strategy. Founder Huang Yi first released AlexBot’s work publicly, found that other people could reproduce and build on it, and then built RoboParty to offer a fuller embodied-AI development layer: robot bodies and modules, control systems, SDKs, and open-source tools. 1944

The path from AlexBot to RoboParty

Huang built the bipedal AlexBot while studying at Harbin Institute of Technology and released the project openly. Reporting says more than 10 companies and universities reproduced or used it. 1824

That early response appears to have shaped the company’s thesis. Huang has said he initially felt the first robot had little practical use, released it publicly, and then began receiving extensive discussion and feedback from people interested in building it themselves. 19 Rather than treat those users only as hobbyists, RoboParty’s model is to serve them with more capable, standardized hardware and a shared development stack.

After graduating early, Huang moved to Shanghai with fellow HIT students and founded RoboParty in February 2025. 1836 The company describes itself as an open-source, full-stack embodied platform, with work spanning joint modules, complete robot platforms, motion control, SDKs, and an open-source development framework. 1344

What RoboParty means by an “Open Body”

RoboParty is positioning itself between robot-model teams and established, more vertically closed hardware vendors. Its stated target is an adaptable robot-body platform for developers, university labs, startups, robot-model companies, and industrial users. 17

The distinction matters: the company is not simply publishing code for a finished robot. Its approach includes the physical body and underlying engineering details, alongside the software needed to train, control, and deploy robotic behavior. Huang has framed the goal as reducing repeated work across embodied-AI development, while retaining a business built around robot bodies and related platforms. 1920

In Huang’s formulation, open source is not useful merely because it is free. He has argued that it must remain technically competitive; otherwise, developers have little reason to adopt it. 1720

RPO today and RP1 next

RPO is RoboParty’s entry-level, globally oriented open-source bipedal platform, designed to help developers progress from an initial idea to a working robot system. 18 The company says the platform achieved a top running speed of about 3 meters per second. 4751

RP1 is the planned next-generation platform. RoboParty has said it is expected in the fourth quarter of 2026 and will focus on improved motion capability, reliability, manufacturing consistency, developer usability, and deployment in more complex real-world environments. 747 A World Robot Conference exhibitor listing describes RP1 as roughly 1.3 meters tall and under 30 kilograms, with a modular body and planetary in-rotor joint modules. 49

Those plans should be read as product targets rather than independently verified deployment results: RP1 had not formally launched in the cited reporting. 58

Party OS: development infrastructure, not a conventional OS

RoboParty calls Party OS an open embodied-intelligence R&D foundation rather than a conventional robot operating system. The purpose is to consolidate the difficult but recurring parts of humanoid development—from motion data and training to policy deployment on real hardware—into reusable infrastructure. 4550

The initial open-source toolchain includes:

• MimicLite, training and deployment infrastructure for general humanoid motion tracking. 2252
• UFO, an unsupervised reinforcement-learning control framework intended to cover the path from training through deployment on physical robots. 2252
• Human-to-Humanoid Tools (hhtools), motion-retargeting tools that map human movement onto humanoid robot structures. 2252

Together, these tools address a practical gap in embodied AI: a capable model is not enough if a team must separately solve robot-specific motion processing, simulation-to-hardware transfer, control, and deployment each time. RoboParty’s platform strategy is to make those layers easier to reuse and extend. 4550

Early adoption and the commercial model

By August 2026, RoboParty’s open-source projects had reportedly surpassed 2,000 GitHub stars. Huang said RPO had delivered more than 300 units since sales began in March, at about 100 units per month. Reported buyers included domestic and overseas universities, robot-model companies, developers, and industrial users. 17

Earlier reports cited almost 10,000 community followers, nearly 1,000 orders, and more than 100 units delivered. The changing totals are consistent with reports from different dates, but they are largely company-reported figures and should not be treated as independently audited sales data. 1846

The commercial logic is therefore straightforward: open source expands the potential developer base and reduces adoption friction, while RoboParty sells standardized bodies and development platforms to users who need a physical system rather than documentation alone. Huang has said open source did not necessarily undermine sales in the education-oriented market, where users may want to reproduce or legally modify the platform. 35

Funding and CATL’s verified role

RoboParty announced nearly RMB 500 million in combined Angel++ and Pre-A financing. CATL exclusively invested in the Pre-A round; returning Angel++ investors included Shunwei Capital, Matrix Partners China, Xiaomi Strategic Investment, and SenseTime Guoxiang. Lightspeed China was identified as incubator and exclusive financial adviser. 1314

Reports also said the company completed six financings in eight months and that its valuation rose more than 20-fold. These are reported company/industry figures, not a disclosed valuation or audited financing record. 218

The company said it would use the capital to scale production of its open-source humanoid platform, improve reliability and manufacturing consistency, build out Party OS, and pursue behavioral and humanoid foundation-model research. 1347

CATL’s investment is notable, but the evidence provided establishes an investment—not a production contract. RoboParty was reported to be building production capability and preparing RP1 for mass-production ramp-up, yet the available sources do not substantiate a CATL manufacturing, battery-supply, contract-manufacturing, factory, volume, or exclusivity agreement. 1346

The bigger bet: shared embodied-intelligence infrastructure

RoboParty’s ambition is to make the robot body a more accessible common layer for embodied-AI development. That is a meaningful departure from a model where each developer or model company starts with a closed machine and limited access to lower-level hardware details.

Whether that model becomes durable will depend on more than community interest. RoboParty must keep its open stack technically strong, manufacture reliable systems at usable cost, and prove that external developers create applications valuable enough to sustain hardware demand. Its early funding, open-source traction, and reported RPO sales show momentum; RP1’s release and broader real-world deployment are the next tests. 171947