Marope: A MARL Framework for Cooperative Long-Rope Jumping by Multiple Humanoids

According to the paper abstract posted on marope-dev.github.io, Marope uses hierarchical reinforcement learning to teach two robots to turn a rope while another jumps. The proposed method was evaluated with Unitree G1 robots in both simulation and the real world, where it achieved more efficient results than several baselines.

For humanoid robot researchers, the important point is that the study tested a task involving multiple robots sharing roles and adapting to differences in a partner’s jumping rhythm—not just locomotion or individual actions. It used a structure in which a higher level coordinated the overall task while lower-level policies controlled each robot’s handling of the rope. The authors say the method was evaluated with Unitree G1 robots in both simulation and the real world and was more efficient than several comparison methods.