From October 22, 智元机器人（AGIBOT） will run seven consecutive days of embodied AI releases, and will operate six G2 robots in a 24-hour, four-shift store deployment. Separately, 宇树科技（Unitree） founder Wang Xingxing placed cumulative G1 output at 10,000 to 20,000 units since May 2024. Release cadence on one side, shipped volume on the other, frame the day's record.
The G1 figure is among the few quantified shipment disclosures in a sector that still mostly reports prototypes. At that scale Unitree holds a volume lead among humanoid makers, while Western progress remains contractual: AGCO and Agility signed a commercial agreement to place Digit on assembly and material-handling tasks at a Kansas plant from 2027. The gap sits in installed base.
The tooling layer is converging in parallel. 英伟达（Nvidia） added a humanoid reference framework, simulation workflow and data tools to Isaac, lowering the engineering cost of building general-purpose robot stacks. At the component end, 越疆科技（Dobot） launched PULSA at 790 degrees per second joint speed with 7kg and 12kg payload options, evidence that cobot throughput is still being pushed upward.
Uptime is becoming the differentiating metric. Unitree's H2-X is specified at roughly eight hours of endurance for service and guarding duty, and AGIBOT's four-shift store trial functions as a duty-cycle test. Wang's argument that robot-AI alignment, matching model inputs and outputs to real hardware, is the binding constraint fits this shift from capability demos to continuous operation.
Two observation points follow. AGIBOT's October 22 disclosures should be read for pricing and deployment specifications, since seven releases in seven days carries limited signal without unit economics. For Unitree, the 10,000 to 20,000 unit figure needs fielded uptime data before it becomes a defensible lead; University of Illinois work combining reinforcement learning with control theory to guarantee stability across wheeled, flying and six-legged platforms indicates where certification pressure will land.