Elder-care scenarios (companionship, health monitoring) accumulate data for the home; home scenario product planned for 2027
Core product; a consumer product focused on home scenarios
First new product after listing and Unitree's first robotic arm under RMB 10K: 7-axis bionic joints, 0.1 mm repeat positioning accuracy, joint speed over 180°/s, 650 mm reach (with dexterous hand), 2 kg rated payload, 5.5 kg self-weight, supporting dual-arm material sorting/assembly and force control
全栈自研具身大模型，具备多模态情绪感知、三维空间理解与主动行动、自主学习与长期进化能力
22-DoF tendon-driven dexterous hand, 50 actuators, tactile fingertip sensors; 1.73 m tall, 57 kg carbon-fiber body, 10-hour endurance, 20 kg max payload
仿人脑五层体系架构（S1本能反射-S5自我成长），2Hz语义推理与100Hz运动控制解耦，融合WAM与VLA后长程动作成功率超98%
Released at WRC 2026, exploring training and data collection solutions combining BCI with humanoid robots
BrainCoSoftware / Data / ModelsGeneral body on a wheeled base (not bipedal), pursuing stability, reliability and cost-effectiveness
Demonstrated 1,816 items/hour at 98%+ accuracy in August 2026, at ~30% the cost of comparable U.S. solutions; already deployed in real logistics scenarios
Uses an 'inner skeleton' architecture with a soft-touch safety design, led by VP of R&D Qin Hailong
World-action model (WA); the world's first to break through on real-time performance, training efficiency, and success rate simultaneously; beat NVIDIA GR00T N1.5 and pi-0.5 for No.1 on the RoboCasa365 home scenario benchmark
Core growth engine at over 70% of revenue; the CC1 has shipped over 20,000 units globally; the BG1 is the world's first AI-native large floor-scrubbing robot; the ET1 covers 100-800 sqm small stores
「皮肤」品类：视触觉极致小型化用于灵巧手指尖，MEMS方案解决漂移与迟滞
旗舰视觉-语言-动作（VLA）模型，首次实现人形机器人全身控制（行走、平衡、蹲伏、五指手操作），单一策略驱动整具身体
具身推理模型（VLM），支持视频理解、任务规划、人机/多机器人协作，已上线 Google AI Studio 并在 Gemini Enterprise Agent Platform 私测
轻量化本地版本，无网络/低时延环境运行，原生多本体，可快速适配新机器人形态
General embodied foundation model; the world's first embodied-native model with the complete 'pre-training + post-training' paradigm; won the WAIC SAIL Star (the only embodied foundation model); built on over 1 million hours of human-centric real data with rich visual-tactile information
Wheeled dual-arm robot, 720×618×1160-1650 mm, about 160 kg, 29 degrees of freedom on the body (excluding end-effectors), 4-hour battery life with hot-swappable battery replacement; innovative long-arm narrow-shoulder design targeting high-payload tasks such as assisted work and eldercare
Industrial wheeled humanoid robot, global premiere
Light-industrial quadruped robot, global premiere
First mass-produced all-scenario robot: 7-DoF modular force-controlled arm (±0.1 mm precision / 6 kg payload / 1,000 mm stroke), automotive-grade E/E architecture + 2,070 TFLOPS FP4 central compute with offline local closed-loop operation, 360° perception with active obstacle avoidance, quick-swap gripper/three-finger hand/dexterous hand; top trim RMB 229,800 with payback in about 1.5 years
Embodied world model; tops WorldArena at 62.34 (the only one above 60), leading across physics adherence, 3D accuracy, and visual quality; technical report, model, data, and benchmark all open-sourced
Fully in-house software-hardware stack for commercial service scenarios (hotels, apartments); confirmed it will not enter homes directly
Embodied-native foundation models breaking the single-machine independent-training mode with multi-source joint training and cross-machine general pre-training