PDF Paper arXiv arXiv github Project

J A E G E R

Dual-Level Humanoid
Whole-Body Controller


PKU    BAAI    BeingBeyond

PDF Paper arXiv arXiv github Code

J A E G E R
Dual-Level Humanoid Whole-Body Controller

Ziluo Ding*   Haobin Jiang*   Yuxuan Wang*   Zhenguo Sun   Yu Zhang  
Xiaojie Niu   Ming Yang   Weishuai Zeng   Xinrun Xu   Zongqing Lu§


PKU     BAAI     BeingBeyond

§ Corresponding author

Paper arXiv Project

Pose-based Whole Body Control

2.5 Years Dance

GYM

Snake

Pose-based Walk

Y.M.C.A.

Squat

Taichi

Walk Sideways

Elbow Swing


Root-based Whole Body Control

Root-based Teleoperation 1

Root-based Teleoperation 2

Root-based Teleoperation 3

Abstract

    This paper presents JAEGER, a dual-level whole-body controller for humanoid robots that addresses the challenges of training a more robust and versatile policy. Unlike traditional single-controller approaches, JAEGER separates the control of the upper and lower bodies into two independent controllers, so that they can better focus on their distinct tasks. This separation alleviates the dimensionality curse and improves fault tolerance. JAEGER supports both root velocity tracking (coarse-grained control) and local joint angle tracking (fine-grained control), enabling versatile and stable movements. To train the controller, we utilize a human motion dataset (AMASS), retargeting human poses to humanoid poses through an efficient retargeting network, and employ a curriculum learning approach. This method performs supervised learning for initialization, followed by reinforcement learning for further exploration. We conduct our experiments on two humanoid platforms and demonstrate the superiority of our approach against state-of-the-art methods in both simulation and real environments.

J A E G E R

JAEGER

    [Dual-level Humanoid Whole-Body Controller]: Retargeting Human Poses to Humanoid Poses. We utilize an efficient real-time retargeting neural network to convert human poses into humanoid poses. The network learns the mapping between human and humanoid pose pairs which are obtained through optimization, rather than simply copying human joint angles.

    [Dual-Level Whole-Body Control (WBC) Architecture]: We decouple the upper and lower bodies into two independent controllers, allowing each to focus on its specific tasks while alleviating the curse of dimensionality in the action space.

    [Curriculum Learning Approach]: Initially, the lower-body controller is trained independently. Next, we implement a supervised initialization for the upper-body controller, followed by whole-body control training using reinforcement learning (RL).

    [Real-World Deployment]: Finally, we deploy the policy directly on the adult-sized H1-2 robot in the real world.

BibTeX

      @article{ding2025jaeger,
        title={JAEGER: Dual-Level Humanoid Whole-Body Controller},
        author={Ziluo Ding, Haobin Jiang, Yuxuan Wang, Zhenguo Sun, Yu Zhang, Xiaojie Niu, Ming Yang, Weishuai Zeng, Xinrun Xu, Zongqing Lu},
        journal={arXiv preprint arXiv:2505.06584},
        year={2025}
      }