wah inovasi banget, biar makin efisien
ASUS just skipped the stage demo and put a humanoid on a live motherboard line.
Gallop packs 15 soft, bagged accessories in ~20 seconds at a targeted 99.5% success rate—real factory work, not choreography.
What stands out on the hardware and training side:
• Single NVIDIA Jetson AGX Thor onboard, ASUS-built body, and custom tactile hands.
• 5-stage loop: open base models + worker head-cam footage + physics sim + VR glove teleop + on-the-job learning.
• 4-layer AI stack (planner → supervisor → skills → control) so newer models can drop in without rebuilding the robot.
Wheeled version shows at Computex 2027. Customer pilots the same year. Commercial target is H2 2028.
The interesting part for unit economics is the path: factory-tested early, modular brain, and a timeline that assumes the training stack keeps improving rather than waiting for perfect generalist hardware.
If ASUS hits those 20-second cycle times at 99.5% on real lines, how much does that actually change the labor-cost math versus other humanoids still in demo or pilot stage?
41


