强化学习让5吨装载机完成杂技动作
美国和瑞士研究人员在仿真环境中用强化学习训练 John Deere 332G 滑移装载机,随后将系统迁移到真机。这台自主装载机可双轮平衡、通过崎岖地面、旋转、操控铲斗并安全落地。相关结果由论文《Heavy Metal Acrobatics》介绍,该研究旨在为重型机械的动态自主控制奠定基础。
该演示表明,在仿真环境中训练的强化学习能够迁移到全尺寸工业机械,为重型机械的 sim-to-real 控制提供了一个实例。
Videos of an incredibly acrobatic, nimble, and graceful John Deere 5-ton skid-steer loader have been shared by a group of researchers from the U.S. and Switzerland. A recently published paper titled Heavy Metal Acrobatics provides the reasoning behind this eyebrow-raising autonomous robotics feat. It is hoped this research will be useful in the future automation of heavy machinery in industries like construction, mining, and material handling. It also demonstrates that reinforcement learning in a simulation can transfer remarkably well to the real world.
Researchers have used reinforcement learning to teach a 5-tonne John Deere skid-steer loader to perform complex acrobatic manoeuvres autonomously. Trained in simulation and transferred to a real machine, the system can balance on two wheels, navigate rough terrain, spin, and control its bucket before landing safely. The research demonstrates the potential of AI-driven automation for heavy construction and industrial machinery.Video Credit: @leggedrobotics / @ETH Paper: https://t.co/MT7fyZ54Hk#engineering #technologyNovember 17, 2026
If you unfurl the video embed above, you can see a hulking 332G skid-steer do some astonishing tricks. It hops back and forth on two wheels, pirouettes, and balances while flexing its lift arm. Moreover, it does these moves while never looking like it will teeter over. The researchers claim that “only a handful of elite human operators” have ever been able to pull off stunts as you see in the demo videos.