South Korean physical AI company Maum AI says a Doosan Bobcat skid loader running its World Action Model has cleared manure on its own in a cattle barn in Pyeongtaek, about five months after the two companies signed an agreement in June. Doosan Bobcat executives watched the demonstrations live from the Bundang Doosan Tower.

In the first run, the loader drove up to a pile, scooped a suitable amount, carried it to a set place and tipped it into a trailer about 2.4 m high, with no one in the driver's seat. In the second, it recognised manure spread unevenly across the floor, chose its own path and pushed the waste together with its bucket close to the ground, then loaded it.

A skid loader is a mobile manipulator: its body of several tonnes moves while hydraulics work the boom and bucket, so pushing into a pile can make it slide, and lowering the bucket too far can hit the concrete floor. Maum AI redesigned the inputs and outputs of its vision-language-action model for heavy machinery. Its researchers earned loader licences and hired an operator with more than 20 years of experience to collect work data, which the company combined with simulations in NVIDIA Isaac Sim.

Maum AI has already put physical AI into an autonomous farm sprayer sold in Korea and Japan, with a production contract for 100 units. Choi Hong-seop, its chief executive for technology, said it would expand the cooperation with Doosan Bobcat, and Doosan Bobcat's chief strategy officer, Park Hyun-chul, called the demonstration the first result of the technical work the two companies began in June.

Source: 지디넷코리아

Background

Maum AI

Physical AI company

South Korean physical AI company that develops the World Action Model.

Doosan Bobcat

Construction equipment maker

Maker of farm and construction equipment under the Bobcat brand, with its American headquarters in West Fargo, North Dakota.

Choi Hong-seop

Technology chief of Maum AI

Chief executive in charge of technology at Maum AI.