RoboParty Unveils RP1, an Open-Source Humanoid You Can Kick, at IROS 2026

On September 28, 2026, in a crowded convention hall in Pittsburgh, a humanoid robot invited a crowd of engineers to kick it. Visitors obliged. Pushed, shoved, and kicked, the robot adjusted its posture and recovered its balance again and again, while its makers watched with the calm of people who had seen this work a thousand times in simulation.

The robot was the RP1, built by RoboParty, and the venue was IROS 2026, the IEEE/RSJ International Conference on Intelligent Robots and Systems, which ran from September 27 to October 1 at the David L. Lawrence Convention Center. The unveiling marked RP1's global debut, and it set the tone for a conference where humanoid robots stopped being lab curiosities and started looking like platforms.

The kick heard around the robotics world

The "Kick Me" demonstration was the most talked-about moment of RoboParty's IROS booth, and it was deliberately unscripted. Unlike choreographed demos built on predefined motions, visitors could apply real external disturbances and watch the robot's dynamic stability respond in real time. Repeated high-intensity shoves doubled as a stress test of the body and the control system at once.

Under the hood, RP1's balance comes from PartyOS, RoboParty's open research and development foundation for humanoid robotics. PartyOS integrates a training framework called UFO that discovers humanoid motor skills through unsupervised reinforcement learning, covering skill transitions, disturbance recovery, and fall recovery without relying on predefined motion trajectories. In other words, the robot was not playing back a "getting kicked" animation. It was improvising.

The hardware backing that up is serious. RP1 delivers peak joint torque of up to 160 N-m, built on in-house-developed Romomo actuator modules paired with a real-time motion-control system. RoboParty is integrating the robot body, actuator modules, low-level control, and embodied AI algorithms into one system for continuous joint validation.

The robot was not playing back a "getting kicked" animation. It was improvising.

Open source as a strategy, not a slogan

What makes RP1 genuinely different from the parade of humanoids unveiled in recent years is the business model around it. The robot builds on RPO (ROBOTO ORIGIN), the fully open-source humanoid project RoboParty launched in January 2026, which has since collected more than 2,500 GitHub stars and inspired community builds and reproductions around the world.

"Humanoid robots should not be closed systems that only a small number of teams can build while everyone else can do little more than watch," said Yi Huang, founder of RoboParty. The company wants RP1 to become, in his words, "an open foundation that developers can access, modify, train and continuously improve."

For RoboParty, "full-stack open source" means progressively opening motion-control systems, simulation environments, SDKs, training tools, and the PartyOS foundation itself, so developers can participate from hardware and low-level control up to model training and deployment. RP1 also builds on the dexterous RP Hand, with PartyOS evolving across locomotion, perceptual interaction, whole-body manipulation, and autonomous intelligent humanoids.

The team behind it reads like a robotics all-star roster, with core members from Tsinghua University's Yao Class, Peking University's Zhi Class, Harbin Institute of Technology, Stanford, Carnegie Mellon, and Zhejiang University. RoboParty says it has built a global community of more than 5,000 developers, plans to announce its open-source roadmap in October 2026, and will follow with broader software and hardware releases and a mass-production program later in the fourth quarter.

Astribot brings the T1 to America

RoboParty was not the only humanoid debut in Pittsburgh. Astribot, a Shenzhen-based physical AI company, showcased its T1 humanoid robot to North America for the first time at IROS 2026, and it is already on sale: US pricing starts at $18,000, with immediate delivery available.

Astribot's pitch is "Design for AI," the idea that intelligence, the operating system, and the physical body should be designed as one integrated system rather than bolting AI onto an existing robot. Its Lumo model family connects reasoning and action in physical environments. Lumo-1 introduced a Reasoning-Action Foundation Model approach, while Lumo-2 explores latent world-action modeling for more complex physical tasks. Equipped with Lumo-2, the T1 has demonstrated autonomous tidying of miscellaneous items into a backpack, combining decision-making with dexterous manipulation of deformable objects.

The hardware is distinctive. T1 uses a cable-driven architecture aimed at dexterity, compliant interaction, and high-speed movement. The robot stands approximately 1.55 meters tall, weighs about 66 kilograms, has 23 degrees of freedom excluding end effectors, and supports up to 5 kilograms of payload per arm. It also supports automatic charging and quick battery replacement, with a modular design for swapping end effectors, computing modules, and sensors.

Astribot is also courting developers aggressively. At its second Astribot OS Hackathon, which concluded September 21 in Beijing, 15 teams built more than ten functional applications on the T1 in 36 hours. Notably, the company is honest about the limits: the T1 does not learn autonomously in real time during deployment. Instead, Astribot collects multimodal data from operation and teleoperation, curates it, retrains models like Lumo-2, and ships updated skills through over-the-air updates.

Why IROS 2026 mattered

Zoom out and a pattern emerges. Days after IROS, China saw the deployment of its first humanoid robot built on a fully domestic electronic architecture, a joint effort by LimX Dynamics and Kyland Technology unveiled in Yichang, Hubei province, integrating indigenous operating systems, networking protocols, developer software, and AI compute chips. The geopolitics of humanoids is now as real as the geopolitics of semiconductors.

But the deeper shift on display in Pittsburgh was architectural. The winning demos were not the most choreographed ones. They were the systems: open training frameworks, integrated AI stacks, developer SDKs, hackathons. Humanoid robotics is moving from "watch what our robot can do" to "here is what you can build on our robot." The kickable robot was the perfect symbol. A demo you can attack is a system its makers trust, and a system they are ready to hand to everyone else.