RoboParty Releases Fully Open-Source Humanoid Robot RP1

“

Not only open source code, but co-evolve with humanoid developers

At the RoboParty exhibition booth of IROS 2026, onlookers were invited to do something rare:

Lift your foot and kick the newly unveiled RoboParty RP1.

After being pushed or kicked, RP1 does not fall over, but quickly adjusts its posture to stand up again, even if it does fall, it can quickly get up.

This action may seem simple, but it actually tests the entire system from joint actuation to full-body control.

The motion capability of humanoid robots is the result of the combined efforts of mechanical structure, actuators, control systems, and learning algorithms.

At the moment of being kicked, the joints need to output sufficient torque and respond to commands in time; the entire machine needs to coordinate multiple joints to redistribute body movement; the control algorithm must continuously adjust actions based on the current state to avoid local adjustments causing new imbalance, which ultimately leads to the result people see.

On October 26, RoboParty officially released this high-performance, fully open-source humanoid robot RP1.

Unlike robot platforms that only open partial application interfaces or algorithm code, RP1's full-stack open-source approach,

based on self-developed ROMOMO joints, integrates the entire machine structure, motion control, simulation training, and real machine deployment into a unified development system, extending the scope of openness to mechanical, electrical, joint, low-level control, simulation, training, and deployment development stages, targeting robot research and embodied intelligence development.

The Robotics Lecture Hall found that the questions it needs to answer are not just “how to create a robot with better performance,”

But as humanoid robot development becomes increasingly reliant on multidisciplinary collaboration, the question is whether there can be a more open technical foundation, allowing more teams to avoid starting from scratch with the underlying system and instead focus their resources on new algorithms, new movement capabilities, and new embodied tasks.

01.

The first open-source humanoid robot to fully integrate the development pipeline

The level of openness determines what researchers can actually do.

The robot industry is not lacking in open-source projects.

Opening code, providing SDKs, and publishing algorithm models already help developers access some technical capabilities. However, for humanoid robots that directly interact with the physical world, different levels of openness lead to completely different research spaces.

Suppose a team wants to validate a new bipedal control algorithm. If the platform only opens high-level interfaces, the team can send movement commands but cannot modify the underlying control logic. Once the new algorithm requires adjusting joint parameters, sensor configurations, or control strategies, the research will be constrained by the platform's original design.

Embodied model development faces similar issues: a model may learn an action in simulation, but when deployed on a real robot, it may experience action lag, posture deviation, or inability to adapt to disturbances. If researchers cannot inspect the control state and execution process, it will be difficult to determine whether the problem lies in the model, the control, or the hardware.

True full-stack openness is not just about letting developers see the code; the core is allowing them to trace the location of problems along the development pipeline and attempt to change the conditions that cause the problems.

Robot Academy noticed that

RP1 extends the openness to mechanical, electrical, joints, low-level control, simulation, training, and deployment, and provides structural documentation, 3D models, circuit documentation, physical description models, and software tools.

Developers can either reuse existing resources or modify the system according to their research goals, and then test the results through simulation and real-world experiments.

This extends the value of open-source beyond code reuse, fundamentally expanding into collaborative research between hardware and algorithms.

For example, in motion control research, developers can combine joint and whole-body documentation to analyze the relationship between actuator response and whole-body motion; in hardware modification, they can adjust sensors or end-effector configurations based on the open mechanical and electrical documentation; in embodied model development, they can use simulation and deployment resources to validate strategies on real robots.

Traditional open-source often stops at the application layer's usability, while full-stack open-source for research is fundamentally about the ability to modify at the lower level.

As the competition in humanoid robot development enters deeper waters, evolving to a stage where research efficiency and system iteration speed are key, the value of open-source depth is undoubtedly growing.

The level of openness essentially determines how high a research team can stand on the shoulders of giants.

RoboParty's RP1 full-stack open-source truly upgrades robots from pre-fabricated tools into experimental objects that can continuously grow with research topics, potentially redefining a new research paradigm.

02.

The challenges of embodied intelligence are not only in the model

The bottlenecks of embodied intelligence are not only in the model, but in the entire pipeline from the model to the real robot.

A model outputting a motion command and a robot accurately performing the motion in the physical world are two different things. Once entering the real world, the actuator response, mechanical structure, sensor observations, control latency, and external disturbances may all change the final outcome. Even if the model's predicted motion is semantically correct, it may not be stably executed by the robot.

This creates an awkward situation for model teams, where if they can only see the input and output, but cannot inspect the underlying control and execution states, then when the real robot's performance does not meet expectations, the troubleshooting may become a process of repeatedly adjusting model parameters.

Even if the final improvement is achieved for a certain task, it may not be possible to explain whether the improvement comes from the model, the control, or hardware adaptation.

The open development pipeline provided by RP1 creates conditions for identifying such issues.

Developers can use physical description models and simulation training resources to conduct experiments, then deploy the strategy on the real robot for verification, and combine the open control and hardware documentation to inspect the execution process, adjust interfaces, and analyze deviations.

The model determines what the robot can attempt, the control and hardware determine how these attempts are implemented, and the complete development pipeline determines whether the team can identify deviations and validate improvements.

Data collection is equally critical.

Embodied model research often requires establishing associations between observations, actions, and the robot's state. If different data streams have time misalignment, or if the observation conditions during training and deployment are inconsistent, the model may learn unstable action strategies.

Robot Academy noticed that RP1 emphasizes millisecond-level synchronized sampling, and

the design philosophy of aligning the data acquisition head with the robot's perspective, which undoubtedly helps improve data time alignment and observation consistency under specific configurations, providing a clearer reference for action learning and real-world verification.

Because the 'hallucinations' of embodied intelligence often occur during the phase when simulation transitions to the physical implementation. When we discuss large models integrating with robots, we often overlook the objective laws of the physical world. If there is a lack of a transparent end-to-end chain from simulation models to low-level control and then to physical execution, the iteration of embodied intelligence will fall into the trap of 'blind men touching an elephant'.

It can be found that RP1's open development pipeline essentially breaks the data black box between models, control, and hardware.

Only when algorithm engineers can clearly identify the physical source of each execution deviation can the closed-loop of embodied intelligence truly be established. This is why full-stack open source is the necessary infrastructure for developers to access general embodied intelligence research.

03.

High performance and modifiability can be achieved on a single robot.

For researchers and developers, openness is indeed important, but openness cannot replace the robot's own capabilities.

A humanoid robot with insufficient hardware capabilities and unstable motion control, even if it provides complete code and blueprints, may not be able to meet the research needs for high-dynamic motion, full-body control, or complex embodied tasks. Conversely, if a robot has strong motion capabilities but does not allow researchers to deeply modify the underlying system, its research value will also be limited.

Therefore, a humanoid robot platform for research needs to simultaneously answer two questions:

whether it can provide sufficient physical capabilities for research, and whether researchers can truly change the way these capabilities are implemented.

RP1 attempts to combine these two aspects.

In terms of hardware configuration, the open-source version has 24 active degrees of freedom, mainly targeting bipedal motion, balance control, and Sim2Real research; the open-source Pro version is equipped with 30 degrees of freedom (including 2 for the head, 7 for each arm), targeting demands such as full-body tasks, human-robot interaction, and embodied data collection, allowing the research platform to match different experimental goals.

Modular design allows researchers to replace head components, end-effectors, and computing configurations according to their research topics; dual battery design supports hot-swapping

. Combined with modular disassembly and easy on-site part replacement design, it can provide engineering support for continuous experiments.

These details may not be as intuitive as dynamic actions, yet they are closely related to the actual efficiency of the research platform.

In terms of motion capabilities, RP1's self-developed ROMOMO joint module has a peak torque of over 160 N·m, the parallel ankle configuration moves the actuator upwards, transmitting force through linkages, reducing the mass at the distal end, and improving the repeatability of the center of mass trajectory; the mechanism parameters are traceable, providing first-hand variables for Sim2Real debugging.

On the algorithm side, RP1 runs its self-developed UFO/MimicLite motion control cerebellum, UFO trains the underlying motion strategies through unsupervised reinforcement learning, covering motion switching, disturbance recovery, and getting up after falling. Thanks to RP1's open control/training deployment pipeline and simultaneous hardware openness at the lower level, it achieves alignment between the simulation actuator model and the real-world actuator characteristics.

High performance makes open source convincing, and open source makes high performance researchable.

Therefore, the Robot Academy believes that RP1's differentiation lies precisely in

integrating hardware motion capabilities, lower-level modifiability, and development toolchains into a single platform design, allowing developers to truly enjoy the benefits of open source.

04.

From working separately, to co-evolving

RoboParty's reason for developing RP1, a high-performance and full-stack open-source robot platform, is not only to solve the problem of how a single team can conduct research and development.

Because humanoid robot development involves mechanical structures, joint drives, simulation environments, basic control, and embodied models, different teams may have completely different research goals, but before innovation, they often face similar underlying work.

For example, university teams need to build experimental platforms, model companies need to complete robot adaptation, education teams hope students can understand the complete system from mechanics to algorithms. When building infrastructure from scratch becomes the norm, repetitive investments are inevitable.

If an open platform can provide relatively complete technical foundations, different teams will have the opportunity to focus more on the issues they truly care about. Researchers can explore different control algorithms, model teams can adjust data collection and deployment processes around specific tasks, and hardware developers can study the synergies between joints, mechanisms, and control strategies.

This differs from traditional robot product logic.

Traditional products mainly solve predefined tasks, and users care about whether the device can stably complete work; research platforms also need to answer another question, which is whether users can continue to create new capabilities on top of existing capabilities.

RoboParty hopes to build such a platform.

It can be found that RP1, with its self-developed ROMOMO joints and whole-machine motion capabilities as the hardware foundation, connects different development stages through PartyOS and related simulation, training, and deployment resources, forming an open system from the body to algorithms, and back to real-world validation.

Thanks to RoboParty's previous RPO open-source project and its community accumulation, RP1 can further address more complex humanoid robot research and embodied intelligence development needs.

Extending the scope of openness to mechanical structures, joints, control, and training deployment stages allows developers to delve into the system and conduct their own experiments, reducing the repetitive, complex, and difficult-to-share foundational work before innovation, enabling more teams to continue exploring on the existing technological basis.

When the basic system can be shared, the industry's innovation resources have the opportunity to shift from redundantly building underlying infrastructure to exploring new mobility capabilities and embodied tasks. This is the true meaning of full-stack open source.

05.

Conclusion and Future

Today, humanoid robots are moving from demonstrating mobility capabilities toward more complex control, learning, and exploration of embodied tasks.

The industry no longer needs just stronger single-machine performance; it also requires a technically foundational platform that is easier to study, modify, and verify.

How many actions a robot can perform defines its current capability boundary;

How many people can deeply modify, verify, and extend a robot influences the technical boundary it can reach in the future. From building their own underlying systems to continuously exploring on an open foundation, there is still a long way to go toward a truly mature open-source ecosystem.

But involving more people in development, reusing results, and verifying new ideas is providing the humanoid robotics industry with another promising development path. On the foundation of technical equity, collectively reshaping the boundaries of humanoid robots.

🔧 GitHub:https://github.com/Roboparty/RP1

Reading Material / Technical Whitepaper / Documentation: https://roboparty.com/documents/roboto-01

END

Industrial robot companies

Eston Automation

|

Efte Robot

|

Fao Robot

|

Yuexiang Robot

|

Jieka Robot

|

Songling Robot

|

Luoshi Robotics

|

A Tongmu Robot

|

JiZhiJia

|

Hikvision Robot

|

Yifei Technology

|

Ailite Robot

Service and special robot companies

Yijiahe

|

Jingpin Tieshang

|

Qitem Robotics

|

Shihe Robot

|

Pudu Robot

|

Shiluo Robot

|

Kuma Technology MAMMOTION

Humanoid robot companies

UBTech Robotics

|

Unitree

|

Yunchedu

|

Xingdongjinyuan

|

Weijing Robot

|

Zhujidongli

|

Leju Robot

|

Daxiang Robot

|

Magic Atom

|

Zhongqing Robot

|

Pasi Ni Perception

|

Cyborg Robot

|

Digital Huaxia

|

Fourier Intelligent

|

Tianlian Robot

|

Kepler Humanoid Robot

|

Lingbao CASBOT

|

Qingbao Robot

|

Zhejiang Humanoid Robot Innovation Center

|

Dongyi Technology

|

Zhishi Technology

|

PNDbotics

|

Zhuoyide Robot

|

Qinglang Intelligent

|

Galileo GALILEO

|

Songyan Power

|

Tianji Intelligent

|

Woaan Robot

|

LiGong Huahui

|

Jiace Jinhua

Embodied intelligence companies

Cross Dimension Intelligent

|

Galaxy Universal

|

Qianxun Intelligent

|

Lingxin Qiaoshou

|

Ruierman Intelligent

|

Microintelligence

|

Tuisheng Technology

|

Zhongke Silicium Records

|

Shutu Technology

|

Lingqiao Intelligent

|

Xingchen Intelligent

|

Qiongche Intelligent

|

Fangzhou Infinite

| iFLYTEK |

Beijing Humanoid Robot Innovation Center

|

National and Local Co-built Humanoid Robot Innovation Center

|

Daimeng Robot

|

Shibitu Robot

|

Xinghaitu

|

Yuequan Bionics

|

Zeroth Power Robot

|

Zhongke ShenGu

|

Zhi Fangping

|

Daka Robot

|

Haocun Technology

|

Ju Shi Intelligent

|

Xynova Xiner Future

|

Feixi Technology

|

Weilu Power

|

Boden Intelligent

|

Qianjue Technology

|

Lingsheng Technology

|

Ji Cui Intelligent Manufacturing

|

Xinbeit Technology

|

Chen Hunxian Technology

|

Dexmal Original Power

|

Youliqi

|

Self-variable

|

Ruiyan Zhikong Lingqiao Hand

|

Qiwu Technology

|

RoboScience Machine Science

|

Zhongke Fifth Dynasty

|

Critical Point

|

Danghong Technology

|

Qiaojie Shu Wu

|

Vbot Veta Power

|

Tashan Technology

|

Ju Brain Pan Shi

|

Youai Zhijhe Robot

|

Zhixing Jian

|

Aimiao Robot

|

Lu Ming Robot

|

Quantum Intelligent

|

Ruoyu Technology

|

Tiehu Robot

|

Zhi Wuji

Medical robot companies

Yuanhua Intelligent

|

Tianzhihang

|

Sizhe Rui Intelligent Medical

|

Jingfeng Medical

|

Tuo Dao Medical

|

Zhenyida

|

Shuru® Robot

|

Luosenbot

|

Shuimu Dongfang

｜

Kangnuo Sistem

|

Disight Medical

Upstream industry chain companies

Lvde Harmonic

|

Yinshi Robotics

|

Kunwei Technology

|

Muta Intelligent

|

Qingpou Vision

|

Benmo Technology

|

Lan Dian Touch Control

| Xinjingcheng Sensor

|

BrainCo Strong Brain Technology

|

Yuli Instrument

|

Jiya Jingji

|

Sila Technology

|

Shenyuan Sheng

|

Feipu Navigation Technology

|

Inks

|

Juxi Intelligent Drive

|

Lingyun Guang Yuanke Vision

|

Xuanji Power

|

Yiyou Technology

| Ruiyuan Precision |

Lingzhu Era

|

HIT Huawei Ke

|

Xinghui Sensor

|

Lingdi Technology

|

Quan Zhi Bo

|

CubeMars Robot Power

|

Wanglong Robot Elevator

|

Hangkai Microelectronics

|

Wutong Sensing and Control