Tashi Zhihang's AWE3.7 Embodied Model Handles Factory and Home Tasks With One Brain

In industrial settings, the AWE3.7-powered robot sorted tiny screws and nuts with different shapes into corresponding boxes, correcting itself when disturbed. It wrapped tape around flexible wire bundles with two hands, adjusting the trajectory in real time to keep the tape tight and unbroken. It also inserted network plugs into sockets with millimeter-level precision, packed a phone into a box and closed the lid accurately, and tightened screws by sensing torque and stopping once they were properly fastened. The tasks involved precision, force control and sustained stability.

On moving production lines, the model grasped soft toys from a conveyor belt and bagged them while following the belt rhythm. It sorted colored blocks by matching each color to the correct frame, detected misplaced blocks and removed them, and tracked frames when they were shifted. In a larger office environment, the robot walked autonomously to a target, picked up an object and dropped it at a delivery point, with end-to-end coordination between navigation and manipulation.

Household tasks tested long-horizon generalization in unstructured spaces. The robot bent to pick up scattered clothes and put them into a basket, switching its center of gravity while moving. It wiped a whiteboard with either hand, gently erasing marks with controlled force, and swapped hands mid-task. It packed pens, erasers and a pencil case into a schoolbag after deciding an order. It took dishes out of cabinets for breakfast, and made a smoothie by holding a soft plastic cup under an ice machine, watching the fill level and stopping before the cup overflowed.

The company attributes the versatility to AWE's methodological loop of native architecture, dual-prior pretraining, world-model-driven post-training, large-scale verification and data feedback, trained with more than one million hours of human-centric data.

Earlier evidence includes a Guinness World Record set in March 2026 by the AWE-powered Tashi A1 robot for assembling the most sub-millimeter wire harnesses in one hour. At the 2026 WAIC, AWE received the SAIL Star as the only embodied foundation model honored. The report says the model has also run continuously on real production lines for automotive wire-harness assembly and collaborative inspection.