Boston Dynamics took the feet off Atlas and bolted skateboard wheels to the ankles. The robot then did a full 360-degree spinning jump and landed it on the first attempt on physical hardware.
The research video went out on 9 November. The wheeled build rolls at speed, turns hard, slides sideways, balances on a single wheel and gets itself up off the floor. The company describes it as a research platform built to find the limits of the motors and of whole-body control, not as a product.
The custom part count is one
Structurally almost nothing changed. Commercial wheels attach to the existing ankle actuators and the only bespoke component is the connector between wheel and ankle. That sparseness is the point. If the machine behaves differently, it is because the actuators and the control policy behaved differently, not because the hardware was redesigned.
Rolling loads are harsher than walking loads in specific ways. The motors have to carry the robot’s mass while accelerating, steering and stopping, holding high torque at high speed. The spinning jump was learned from human motion capture data and executed through reinforcement learning with a simulation that includes the measured physical characteristics of the real motors. Control engineer Andy Wu said the robot worked out on its own how to drive itself with the wheels.
The date that matters is 2028
Hyundai Motor Group intends to feed this control work back into the bipedal Atlas. Hyundai Mobis plans to start prototype production of Atlas actuators this month at its Uiwang research centre in Gyeonggi Province, with supply targeted for 2028. Hyundai plans to deploy Atlas for parts sequencing at its new Georgia plant in 2028 and extend it to assembly processes by 2030.
A research video is cheap to publish and easy to dismiss. An actuator prototype line with a 2028 supply date and a named plant is the part European tier-one suppliers should be reading, because it converts a demonstration into a bill of materials with a schedule attached.
The robot is also being trained for a failure mode that factories care about. Boston Dynamics wants the control stack to adapt when an actuator fails in an unpredictable environment, which is the difference between a machine that runs a demo and one that holds a shift.
Editor’s note: This is an adapted translation of the original Gasgoo report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://auto.gasgoo.com/news/202611/11I70477573C101.shtml.