Contact-Staged Stand-up Learning for Humanoids under Shifted Center-of-Mass Conditions

Sangbeom Park, Seunghyun Lee, PAUL KIM, Daewan Kim, Haechang Chung, Juwon Kim, Jaehyuk lee, Minho Oh, Euigon Jung, Dongha Chung, Byeongho Yu, Dongkyu Lee

Keywords: Humanoid robots, Contact-aware control, Reinforcement learning, Stand-up, Fall Recovery

TL;DR: A contact-staged multi-critic RL framework enables humanoid stand-up under rear-shifted mass conditions caused by external payloads.

Abstract: Autonomous stand-up is essential for humanoid robots but remains challenging due to complex contact transitions and coordinated whole-body weight transfer. The challenge becomes more severe when customized hardware shifts the center of mass (CoM) and alters the nominal mass distribution. To address this, we propose a contact-staged stand-up framework for humanoid robots under biased CoM conditions. The proposed method decomposes the stand-up motion into four sequential stages defined by contact and body-state conditions. It further assigns dedicated critics to distinct reward groups and adopts a two-phase curriculum to improve motion smoothness and transferability. Experiments on a customized Unitree G1 with a modified head module and a rear payload demonstrate that the proposed framework outperforms ablated variants in both simulation and real-world experiments.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 27