Xingyi Intelligence
Category: Robotics / Embodied AI
Tsinghua-incubated physical AI data infrastructure startup. It builds wearable multimodal egocentric capture hardware and an automated hand-pose annotation engine that produce high-precision training data for embodied AI and robotics. Xingyi Intelligence was founded in 2024. The company is led by Song Zhiheng (宋知珩). Based in Beijing (Haidian), China. Team size: 11-50. Total funding raised: $4.0M. Latest round: Seed. Key investors include Shunxi Fund (顺禧基金), Luster LightTech (凌云光), Zhongguancun Capital (中关村资本), First Capital Investment (一创投资), Shuimu Ventures (水木创投), Quanshi Capital (泉士资本), Yaozhuo Capital (钥卓资本).
- Founded
- 2024
- Headquarters
- Beijing (Haidian), China
- Team size
- 11-50
- Total funding
- $4.0M
Value proposition
Turns real human manipulation into high-quality, reusable, scalable training assets for embodied AI models. Its first-person capture fuses vision, touch and pose, and its automated annotation reaches millimeter-level 3D hand-pose precision. The company says collection costs about one-tenth of teleoperation and annotation costs about 1/100 of manual labeling.
Products and solutions
EgoKit: multimodal egocentric data-collection system with a head-mounted multi-camera rig and tactile sensing gloves. HBR Engine: data engine for hand-pose parsing, multimodal fusion, automated mm-level annotation and quality auditing. Hand-interaction training datasets, with a stated plan to open-source 1,000–10,000 hours of data.
Unique value
High-precision, high-degree-of-freedom hand interaction data with tactile feedback, captured in real-world settings rather than labs. It parses hand pose through gloves of any color and handles occlusion, extreme lighting and high-speed motion.
Target customer
Embodied AI model companies, robot OEMs, universities and research labs, data-collection factories, and eventually end-scenario operators such as factories.
Industries served
Embodied AI / humanoid robotics, robot foundation models (VLA), industrial manufacturing (precision assembly), research institutions
Technology advantage
Millimeter-level 3D hand-pose estimation in long-sequence, contact-rich hand-object interaction. Recognition of gloved hands, which enables fusion of visual, tactile and pose signals. Millisecond-level multi-sensor synchronization. Near-fully automated annotation. The team comes from Tsinghua and Beihang and reports 70+ papers at CVPR, ICCV, ECCV, NeurIPS and IJCAI; industry veterans joined from AgiBot, Megarobo, Efort, Hikvision and Li Auto.
How they differentiate
Rivals mostly rely on two-finger-gripper UMI capture, simulation, or lab-bound egocentric capture. Xingyi Intelligence instead focuses on high-DoF dexterous hand data (21 hand joints) with tactile feedback, captured in real production scenes. It offers mm-level automated annotation at a fraction of manual labeling cost.
Main competitors
LightWheel (光轮智能, simulation/synthetic data plus EgoSuite human video), Mifeng Technology (觅蜂科技, AgiBot-incubated data collection), TARS Robotics (它石智航, human-centric data and embodied models)
Key partnerships
Strategic partnership with Luster LightTech (凌云光) through its wholly-owned subsidiary Yuanke Shijie (北京元客视界). The partnership covers ego capture, multi-view vision and optical motion-capture hardware, plus collection, annotation, quality evaluation and data delivery. Incubated by the Tsinghua University Department of Computer Science. Based in the Haidian AI Origin Community (AI原点社区). Reports project-level work with several leading embodied-model companies and robot makers, none named.
Major milestones
2025: founded in Beijing, incubated by Tsinghua CS. Mar 2026: closed a first round of ¥10M+ led by Shuimu Ventures. Apr 2026: voted 'Most Interested Project' at Haidian TouHai Tech Show. 2026: selected as an Embodied AI Application Benchmark Case at the China Beijing International High-Tech Expo, featured on CCTV Focus Report (焦点访谈). Oct 2026: strategic partnership with Luster LightTech's Yuanke Shijie, closed Seed and Seed+ rounds totaling tens of millions of RMB.
Growth metrics
The team numbers several dozen, with over half from Tsinghua or Beihang. The company reports project-level data deliveries to multiple leading embodied-model companies and robot makers.
Market positioning
Pitches itself as a 'physical-world ImageNet', the data infrastructure layer for embodied AI. It positions beneath model developers and robot hardware makers.
Geographic focus
China (Beijing) primarily, with global embodied-AI data ambitions
About Song Zhiheng (宋知珩)
Ex-AgiBot (智元机器人) product lead for its full-size bipedal humanoid robot, also in charge of data collection and teleoperation systems. Ex-Megarobo (镁伽机器人) early employee (among the first 20), where he built the Innovative Applications BU and served as product lead, delivering a 10,000-unit mass-produced product and over RMB 100M in revenue. Tsinghua University Master's; founding vice chair of the Tsinghua MBA Embodied AI Club. About 8 years of full-stack hardware/software R&D and commercialization experience.
Latest news about Xingyi Intelligence
More Robotics / Embodied AI companies
Official website: https://www.xy-ai.tech/