Researchers from the United States and Switzerland have developed an autonomous 5-ton John Deere 332G skid-steer loader capable of performing complex acrobatic maneuvers. The machine has been observed hopping on two wheels, pirouetting, and balancing while flexing its lift arm. According to a paper titled Heavy Metal Acrobatics, these stunts are difficult enough that "only a handful of elite human operators" have ever been able to perform them.
The technology relies on reinforcement learning (RL) combined with physics modeling and neural network models. Because the machinery is too large and powerful to undergo direct reinforcement learning on the hardware, the research team utilized a simulation programmed with the complex controls of the John Deere 332G. This training data was then applied to the physical machine through a sim-to-real transfer process.
This research aims to establish the foundations for dynamic autonomous control of heavy machinery. The authors stated that the goal is to promote greater automation in sectors such as construction, mining, and material handling. The study demonstrates that reinforcement learning in a simulation can transfer effectively to real-world applications.
Ultimately, the researchers suggest that autonomous control could eventually become safer, faster, and more efficient than human operation. The ability of the AI to execute such highly maneuverable movements suggests that there is more scope for AI-driven automation in heavy industry than previously expected.














