Archon Robotics

Archon Robotics (Chinese: 源策未来; legal name 上海源策未来智能科技有限公司, Shanghai Yuance Weilai Intelligent Technology Co., Ltd.) is a Chinese embodied AI startup based in Shanghai that develops foundation models for humanoid robots. The company calls its approach "Whole-Body Intelligence" (WBI): models pretrained on large amounts of human whole-body motion data and then post-trained on real-robot data, intended to let a humanoid coordinate its legs, torso, arms, hands and gaze as one system rather than treating locomotion and manipulation as separate problems ^[1][2][7]. It was founded in April 2026 by Hongyang Li, an assistant professor at the University of Hong Kong and the lead of the UniAD autonomous-driving project, together with co-founders Tianyu Li (CEO) and Li Chen (head of AI) ^[2][3][5]. In June 2026 it reported a seed round of several hundred million yuan from investors including ZhenFund, Gaorong Ventures, IDG Capital and 5Y Capital ^[2][3][4].

Its public demonstrations have run on humanoids made by other Chinese companies, including the AgiBot A3, Booster T2 and modified Unitree G1 robots ^[11][12][13].

Overview

History

Founding and seed round

Chinese business outlets reported on 29 June 2026 that Archon Robotics had been founded in April 2026 and had closed a seed round worth several hundred million yuan (the exact amount was not disclosed). The investors named were ZhenFund (真格基金), Gaorong Ventures (高榕创投), IDG Capital and 5Y Capital (五源资本), together with the joint fund of Gobi Partners and the University of Hong Kong, MiraclePlus (奇绩创坛) and the Shanghai Innovation Institute (上海创智学院). Lighthouse Capital (光源资本) acted as the company's exclusive financial adviser ^[2][3][4][6]. The same investors, plus the University of Hong Kong, appear in the investor section of the company's website ^[1]. ChinaVenture described the round as the company's first financing round ^[5].

According to the reports, the money was to go to research on a whole-body humanoid foundation model, collection of multimodal whole-body motion data, hiring, and building research centers in several cities and an industry partner network. The company said it aimed to release an open-source humanoid foundation model within 2026 ^[2][4][5]. 36Kr reported that Archon planned to release its first "humanoid native" foundation model in the second half of 2026 ^[3].

One trade outlet, Gasgoo, gave the founding date as April 2024 ^[6]; 36Kr, PEdaily and ChinaVenture all give April 2026 ^[2][3][5].

Public research and demonstrations

Founders

Hongyang Li (李弘扬) is the founder and chief robot officer. At the time of the funding announcement he was an assistant professor at the University of Hong Kong and a mentor at the Shanghai Innovation Institute ^[2][3]. He founded OpenDriveLab, a research group that describes itself as working on embodied AI and autonomous driving at HKU, and Archon and OpenDriveLab held a joint program at RSS 2026 ^[2][8][18]. Li was the project lead of UniAD ("Planning-oriented Autonomous Driving"), an end-to-end autonomous driving model from OpenDriveLab and OpenGVLab at Shanghai AI Laboratory, Wuhan University and SenseTime Research that unified perception, prediction and planning in one network ^[9]. UniAD won a Best Paper award at CVPR 2023 ^[10]. At RSS 2026 he gave an Early Career Spotlight talk titled "Whole-body Intelligence with Human-centric Data at Scale" ^[1][8].

Tianyu Li (李天羽) is co-founder and CEO. Chinese press reports describe him as a Fudan University PhD, one of the first graduates of the Shanghai Innovation Institute, and a core developer of the world-engine component of Huawei's production ADS 4.0 driver-assistance system ^[2][3]. He spoke at the RSS 2026 workshop on long-horizon whole-body control on "Scaling Whole-Body Humanoid Skills with Human Demonstration" ^[8].

Li Chen (陈立) is co-founder and head of AI. He studied at Shanghai Jiao Tong University's Zhiyuan Honors program and held an HKU Presidential PhD Scholarship, according to the funding coverage ^[2][3]. He was one of four equal-contribution authors of the UniAD paper and, with Hongyang Li, one of its two project leads ^[9]. The company and PEdaily both state that he is an area chair for ECCV 2026 ^[1][2].

The company says its founding team includes people from the University of Hong Kong, Tsinghua University, Shanghai Jiao Tong University and technology companies, among them former autonomous-driving engineers, robotics researchers and large-model engineers ^[1].

Whole-Body Intelligence

Archon's July 2026 essay argues that most humanoid progress still comes from separate components (whole-body control, behavior foundation models, motion generation, vision-language-action policies, world models, mission-level agents and runtime systems) and that none of them alone gives humanoids a scalable path to general-purpose capability. The company defines whole-body intelligence as "the ability of a humanoid model to learn reusable full-body priors from heterogeneous human and robot experience, then use those priors to produce safe, adaptive, executable behavior in the real world" ^[7].

Four-layer model stack

The essay describes a humanoid model stack with four layers ^[7]:

The essay names SONIC-style whole-body trackers as examples of the S0 layer and says that "a humanoid foundation model is not a single checkpoint. It is a compounding loop" of pretraining, deployment, failure collection and post-training ^[7].

Data strategy

Archon's position is that human data provides scale, robot data provides grounding and failure data teaches recovery ^[7]. In interviews with 36Kr and PEdaily, CEO Tianyu Li argued that existing embodied datasets carry little information useful for whole-body humanoid training: egocentric video lacks body pose, arm-and-gripper data records mainly end-effector trajectories, and real humanoid data is scarce and costs several hundred to nearly a thousand yuan per hour to collect ^[2][3]. The company calls its approach "Human Body Learning", meaning learning human whole-body posture and coordination rather than only end-effector paths, and said it would add touch and higher-precision body and hand capture to its data collection ^[2][3]. In the 36Kr interview the founders described a three-level design in which a "brain" handles task understanding and planning, a "midbrain" learns body-independent whole-body motion representations, and a "cerebellum" handles real-time pose tracking and balance ^[3].

Demonstrations

Debut demonstration (October 2026)

Archon's first public demonstration video, titled "Building Whole-Body Intelligence for the Real World. We're Just Getting Started.", runs a little over three minutes. Its description lists bending to pick up objects, climbing steps, clearing tables, operating dishwashers and organizing refrigerators, and says the framework combines "large-scale human whole-body data pretraining, context-aware behavioral foundation models, and efficient post-training with real-robot data" ^[11].

Coverage by Humanoids Daily, Interesting Engineering and Origin of Bots identified an AgiBot A3, Booster T2 robots and modified Unitree G1s in the footage, including a scene with three G1s and two T2s working in the same kitchen and dining area ^[12][13][14]. The video focused on two layers: S1 for task understanding and behavioral decisions, and S0, which Archon compared to a cerebellum that turns targets into executable whole-body movement ^[12][13]. Archon said its data platform collects human demonstrations in several Chinese cities, that the data goes through automated checks, human spot checks and review of flagged samples, and that it also trains on DAgger corrective data showing recovery from mistakes ^[12][13]. According to the video's closing disclosure as reported by Humanoids Daily, the full task sequences were autonomous at normal speed, while the S0 explanation and DAgger segments showed teleoperation ^[12].

The outlets noted several limits. Humanoids Daily said that robots sharing a room does not show that they run identical model weights or coordinate on a shared task, and that the video's two-layer presentation differs from the four-layer stack in the July essay without explanation ^[12]. Interesting Engineering reported that Archon had not disclosed dataset sizes or published quantitative results on skill transfer ^[13].

Tactile sensing and deformable objects

On 29 October 2026 the company posted a 75-second video on X titled "Building Whole-Body Intelligence for Tactile Sensing and Deformable Object Manipulation". The accompanying text says that touch supports grasping AirPods and that coordinated hands handle changes in fabric shape ^[15]. Archon did not publish an accompanying paper or numbers with the post. See tactile sensing for background on touch in robot manipulation.

Research

PingpongDuo

"PingpongDuo: Agile Humanoid Table Tennis Doubles with Safe Team Coordination" (arXiv:2610.20579, submitted 14 October 2026) is a nine-author paper led by Haoran Jiang, with Tianyu Li and Li Chen among the authors. Listed affiliations are the Shanghai Innovation Institute, Archon Robotics, the University of Hong Kong and NVIDIA ^[16]. The project page describes it as "the first humanoid table tennis doubles system" ^[17].

In doubles, two teammates on the same side must strike in strict alternation while sharing a small workspace. PingpongDuo uses a hierarchical design: a team-level planner picks positioning targets for both robots and filters risky proposals using learned risk estimates, while each robot has separate footwork and striking skills guided by human motion references (including footwork recorded from human doubles play). A racket-arm residual refines strikes using feedback from a simulator calibrated with real ball-flight and contact measurements ^[16]. The real-world experiments used two 29-degree-of-freedom Unitree G1 robots with external motion capture for robot and ball state, and simulation ran in Isaac Lab ^[16].

The abstract summarizes the simulation result as a 13.2% improvement in ball-return success over the state-of-the-art baseline ^[16].

Partners

The company's website lists Unitree Robotics, Booster Robotics, Lightwheel and HumanPlus as partners ^[1]. Archon also lists accounts on GitHub, Hugging Face, YouTube, LinkedIn and X ^[1].

Continue reading

• WWhole-body controlWhole-body control (WBC) is a class of robotics control techniques that coordinates all of a robot's degrees of freedom at once to achieve multiple prioritized tasks, such as balancing, reaching, and…Mentioned twice above
• EEmbodied AIEmbodied AI is an area of artificial intelligence concerned with agents whose perception, action, learning, and reasoning are coupled through interaction with an environment.Mentioned twice above
• AgiBotAgiBot, styled AGIBOT and branded in Chinese as Zhiyuan Robotics (智元机器人), is a Chinese robotics company founded in Shanghai in February 2023.Mentioned twice above
• Humanoid Robot"Humanoid" redirects here. For the London-based robotics company, see Humanoid (company).Listed in See also
• TTeleoperationTeleoperation is the operation of a machine by a human at a distance, with commands carried over a communication link and the machine's cameras, microphones and force sensors standing in for the operator's own…Listed in See also
• TTactile sensingTactile sensing is the technology that gives robots a sense of touch: the ability to measure the physical contact between a robot and the world, including contact force, pressure distribution, shear, slip…Mentioned 4 times above

More to read

• Unitree G1
• AgiBot A3
• Booster Robotics
• Unitree Robotics
• Huawei
• SenseTime

Topics:Chinese AIEmbodied AIHumanoid RobotsRobotics Companies

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14Archon Robotics. "Archon Robotics 源策未来" (company homepage). archon.tech (accessed 30 October 2026).
2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18投资界 (PEdaily). "通用全身具身大脑公司「源策未来」获数亿元种子轮融资." 29 June 2026. news.pedaily.cn/...565638.shtml
3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^1536Kr (Hard Krypton). "HKU Professor LI Hongyang Launches Startup Developing General Full-Body Embodied Brain, Secures Hundreds of Millions in Seed Round from ZhenFund, Gaorong Capital, IDG Capital, 5Y Capital." 29 June 2026. eu.36kr.com/...3868055841641476
4. ^1 ^2 ^3 ^4 ^536Kr. "'Archon Robotics' has completed a seed financing of several hundred million yuan." 29 June 2026. eu.36kr.com/...3873554111796232
5. ^1 ^2 ^3 ^4 ^5 ^6投中网 (ChinaVenture). "源策未来获数亿首轮融资." 29 June 2026. m.chinaventure.com.cn/...114-20260629-392047
6. ^1 ^2 ^3Gasgoo. "Seeds | Archon Robotics, a General Whole-Body Embodied Intelligence Developer, Raises Hundreds of Millions in Seed Round." 29 June 2026. autonews.gasgoo.com/...d-round-2071597085466865665
7. ^1 ^2 ^3 ^4 ^5 ^6Archon Robotics. "Whole-Body Intelligence: The Pretraining Path to Large Humanoid Models." 13 July 2026. archon.tech/...whole-body-intelligence
8. ^1 ^2 ^3 ^4OpenDriveLab. "Archon & OpenDriveLab at RSS 2026." opendrivelab.com/rss2026 (accessed 30 October 2026).
9. ^1 ^2Hu, Yihan; Yang, Jiazhi; Chen, Li; Li, Keyu; et al.; Li, Hongyang. "Planning-oriented Autonomous Driving." arXiv:2212.10156, CVPR 2023. arxiv.org/...2212.10156
10. ^CVPR 2023. "CVPR 2023 Awards." cvpr2023.thecvf.com/...Awards
11. ^1 ^2 ^3Archon Robotics. "Building Whole-Body Intelligence for the Real World. We're Just Getting Started." YouTube, 9 October 2026. youtube.com/watch
12. ^1 ^2 ^3 ^4 ^5 ^6 ^7Humanoids Daily. "Archon Robotics Shares Debut Demo of Its Whole-Body Humanoid Intelligence." 12 October 2026. humanoidsdaily.com/...ebut-whole-body-intelligence
13. ^1 ^2 ^3 ^4 ^5 ^6Interesting Engineering. "Smart system helps humanoid robots coordinate their whole bodies for chores." 13 October 2026. interestingengineering.com/...le-body-intelligence
14. ^1 ^2Origin of Bots. "Shanghai Startup Archon Shows One AI Brain Doing Household Chores on Three Different Humanoids." 12 October 2026. originofbots.com/...s-on-three-different-humanoids
15. ^1 ^2Archon Robotics (@Archon_Robotics). Post on X, 29 October 2026. x.com/...2115686252225991071
16. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Jiang, Haoran; Zhong, Yichao; Zhang, Kai; Liu, Cheng; Pan, Yixuan; Yang, Zhenjie; Wang, Jingbo; Li, Tianyu; Chen, Li. "PingpongDuo: Agile Humanoid Table Tennis Doubles with Safe Team Coordination." arXiv:2610.20579, 14 October 2026. arxiv.org/...2610.20579
17. ^1 ^2 ^3 ^4 ^5 ^6Archon Robotics. "PingpongDuo: Agile Humanoid Table Tennis Doubles with Safe Team Coordination" (project page). archon.tech/...PingpongDuo (accessed 30 October 2026).
18. ^OpenDriveLab. "OpenDriveLab | Embodied AI and Autonomous Driving at HKU" (lab homepage). opendrivelab.com (accessed 30 October 2026).

Show all 18 references