Chinese embodied intelligence sets global record: beats Figure AI's 45% efficiency at 30% cost, smart embodied brain key

In a live logistics sorting test, the Chinese company Zibian (自变量机器人) achieved a sorting rate of 1816 packages per hour using dual robotic arms and standard grippers, surpassing Figure AI's record of 1248 packages per hour by over 45%. The hardware cost is 70% lower than Figure AI's humanoid solution with five-finger dexterous hands. Sorting accuracy reached 98%. The key is their proprietary WALL-B world unified model (WUM), which integrates vision, language, touch, and action in a single network, enabling real-time perception, spatial reasoning, and adaptive strategies. Unlike traditional VLA models, WALL-B reduces information loss between modules. The robot handled randomly mixed packages of various shapes, sizes, and materials, adjusting actions such as using dual-arm coordination for heavy boxes or sliding for awkward items. The model's zero-shot generalization allows it to adapt to unseen objects. The company previously tested WALL-B in home environments, accumulating experience for industrial applications. This demonstrates a cost-effective path for embodied intelligence with simpler hardware and stronger models.