Researchers at ETH Zurich have used AI to teach a 5-tonne skid steer loader to balance and drive on its two rear wheels on its own. The work is published in Science Robotics under the title Heavy metal acrobatics: Learning dynamic control for skid-steer loaders.
The team trained the control for John Deere's 332G loader in a simulated environment and then moved it to the real machine. A skid steer loader steers by turning its left and right wheels at different speeds. The researchers chose simulation because the machine's weight and complex hydraulics make repeated tests on the real loader hard.
In video of the tests, the loader lifts its front wheels and balances on the rear pair, drives over uneven ground and turns on the spot. It also swings its bucket back and forth while balancing, then sets down safely on four wheels. The skills were learned in stages, lifting the front, keeping balance, controlling speed and landing, and then joined up, with data from the real machine used to narrow the gap between simulation and reality.
The team said the work shows that reinforcement learning combined with simulation can control heavy and complex machines dynamically.
Background
ETH Zurich
Research organization
Public university in Zurich, Switzerland, founded by the Swiss federal government in 1854, which teaches and researches mainly science and technology.
More stories about ETH Zurich
John Deere
Machinery maker
A US company that makes farm machinery, heavy construction equipment, forestry machines, diesel engines and drivetrain parts. It is headquartered in Moline, Illinois.



