AI Swings 5-Ton Loader Onto Two Wheels and Balances It

Zurich, Switzerland – A 5-ton skid-steer loader has learned to swing itself up onto two wheels, balance, spin and land again on its own, using artificial intelligence trained in computer simulations.

Researchers from ETH Zurich's Robotic Systems Lab, Blue River Technology and Intel trained the control programs for a John Deere 332G loader before running them on the real machine, according to their study published in the journal Science Robotics on Wednesday, November 11. Video released with the study shows two loaders balanced on their rear wheels coming down side by side. The one guided by the AI lands gently on its front wheels and settles on all four, while the other, with its commands cut, slams its bucket into the dirt, bounces its rear wheels off the ground and is left tipped forward on its bucket.

The researchers said the system could lift the loader onto two wheels and keep it balanced while it drove over rough ground, followed set paths, spun around and swung its bucket back and forth, before setting it back down on all four wheels. It pushed the machine to its limits, at full speed and maximum control inputs.

Such moves "have only been previously accomplished by a small number of elite human operators in official world-record performances," the authors wrote. They said the technique could one day let heavy machines load themselves onto trucks without a ramp or find their way around obstacles on rough ground.