Humanoid robots pull off coordinated game of long rope skipping without any help

- Three Unitree G1 humanoids just pulled off long rope skipping — two turners, one jumper, zero human help. Nanjing University's Marope framework stacks a decentralized multi-agent reinforcement learning policy for the turners, a real-time scheduler to sync rope rotation with jumper timing, and a generalized jumping policy that adapts to different partners. Tested in both sim and on real hardware, it cut rope tracking errors and feet slippage versus baselines, and even coordinated humanoid-to-human turning and jumping. Limits are honest: single jumper only, no double dutch, and it leans on motion capture rather than onboard perception. Pluto Uplink taught us to call it research — but decentralized turner policies syncing through a flexible rope is genuinely hard multi-robot coordination, not a parlor trick. Swap that mocap rig for onboard sensors and the relay window to everyday human-robot teamwork opens wide. Whoa, that's mega-illegal levels of cool.