RPL: Learning Robust Humanoid Perceptive Locomotion on Challenging Terrains
Industry
Source: arXiv cs.ROPublish time unverified
arXiv:2602.03002v2 Announce Type: replace Abstract: Humanoid perceptive locomotion has made significant progress and shows great promise, yet achieving robust multi-directional locomotion on complex terrains remains underexplored. To tackle this challenge, we propose RPL, a two-stage training framework that enables multi-directional locomotion on challenging terrains, and remains robust with payloads.