WB-WAM: Heterogeneous Body-Hand Pre-training for Humanoid Loco-Manipulation

Loading video
Loading videoWB-WAM injects explicit body, root and dexterous-hand supervision into generative video pre-training, jointly learning video and actions from 1,880.2 h of heterogeneous data; 81.9% in HumanoidArena sim and 84.0% mean success across five real humanoid tasks.
Category: humanoid
Author: @ShaotingZ38103
Date: 2026-10-14T00:00:00
Duration: 17.879s
Reference: https://arxiv.org/abs/2609.34199





