Zhiyuan Robotics (Beijing Zhiyuan Robotics Technology) is one of the less widely covered but well-funded Chinese humanoid companies, with approximately CNY 1 billion ($138 million) in disclosed funding as of 2025. Its humanoid robot, STAR1, is designed for service environments, hospitals, hotels, retail, and elder care, rather than industrial manufacturing. Zhiyuan’s differentiation from the manufacturing-focused competitors (AgiBot, Unitree, UBTECH) is its focus on unstructured human environments and its embodied AI approach, which emphasizes natural language interaction and adaptive behavior rather than optimized production cycle times. This article covers STAR1, Zhiyuan’s AI approach, and its target market.
STAR1 Specifications
STAR1 is a full-size bipedal humanoid robot: height 168 cm, weight 65 kg, 32 degrees of freedom. The hands are 5-finger dexterous with tactile sensors, 10 DOF per hand. Walking speed: 1.2 m/s, lower than industrial humanoids but sufficient for hospital ward and retail floor speeds. The design aesthetic is deliberately approachable, smoother surfaces, no exposed actuators, a stylized face with LED eye displays, and a voice interface optimized for Mandarin customer interaction. This design philosophy reflects Zhiyuan’s target deployment environment: a STAR1 operating in a hospital ward must not alarm patients or staff, and its physical appearance is part of the user acceptance calculus.
Sensors: forward stereo RGB-D camera (face-mounted for human interaction gaze direction), 360-degree LiDAR for navigation, body cameras for task perception, microphone array for voice detection in noisy environments, speaker with speech synthesis. Compute: Qualcomm Snapdragon for edge inference, cloud connection for large model inference. The cloud-connected AI is a deliberate design choice: STAR1 offloads complex reasoning and language tasks to cloud models, keeping local compute requirements lower and enabling model upgrades without hardware changes.
Embodied AI Approach
Zhiyuan’s AI stack centers on a proprietary embodied large model trained on service scenario data, not the manipulation-heavy data that AgiBot and Unitree train on, but the interaction-heavy data specific to service contexts: patient intake conversations, prescription pickup verification, product location queries, room service delivery confirmation. The model is multimodal (visual, audio, language) and fine-tuned on Chinese service industry vocabulary and social conventions, a significant domain adaptation from general-purpose large models trained on internet data.
The 10 million data entry target announced by Zhiyuan covers this service context data, interactions recorded from pilot deployments in Beijing hospital and hotel environments, synthetic scenarios generated for edge cases (patient in distress, medication verification conflict, navigation in crowded elevator lobby), and licensed service industry interaction data from partner institutions.
Hospital Deployment Pilots
Zhiyuan has conducted pilot deployments in two Beijing hospital environments (one disclosed: a 1,000-bed teaching hospital in Beijing). Tasks in hospital pilots: medicine delivery from pharmacy to ward (STAR1 carries a sealed medication tray, handles to ward, confirms patient identity verbally, delivers tray to nurse), equipment cart retrieval (locating and returning equipment carts from hallways to storage rooms between shift changes), and patient orientation (guiding patients and visitors from reception to their destination, the robot accompanies and provides verbal navigation). Tasks not piloted: patient physical interaction (lifting, repositioning), excluded from initial pilot due to safety certification requirements not yet established.
Retail Deployment
In retail, Zhiyuan has piloted STAR1 in a Beijing convenience chain (5 stores, 30-day pilot in 2025): product restocking (STAR1 handles storage room, picks products from shelves by planogram specification, carries to store floor, places on shelf), inventory scanning (walking aisles, scanning shelf barcodes and noting out-of-stock positions), and customer direction (responding to customer queries about product location). The retail pilots generated the most user interaction data, convenience store customers interact with STAR1 spontaneously, providing unstructured interaction data that hospital pilots do not generate at the same volume.
Full ranking: Top 40 China Robot Rankings 2025. Service vs industrial humanoid positioning: Fourier GR-3 emotional AI.
Sources
- 36Kr: Zhiyuan Robotics funding and STAR1 launch (2025)
- TechNode: Zhiyuan hospital deployment pilots Beijing (2025)
- Zhiyuan Robotics: STAR1 product documentation (2025)
- CRIA: Chinese humanoid service robot sector report (2025)
