Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Unleashing Humanoid Reaching Potential via Real-world-Ready Skill Space

Zhikai Zhang, Chao Chen, Han Xue, Jilong Wang, Sikai Liang, Yun Liu, Zongzhang Zhang, He Wang, Li Yi

TL;DR

This work tackles the challenge of enabling humanoid robots to reach a large 3D workspace by decoupling multi-skill control from end-to-end optimization. It introduces Real-world-Ready Skill Space (R2S2), a framework that builds a library of primitive skills, ensembles them into a unified latent space via a CVAE, and uses a high-level planner to sample skills for robust sim2real transfer. Extensive experiments on autonomous WBC tasks and real-world transfer demonstrate superior task success and stability over baselines, with a teleoperation system leveraging the latent skill space for large-workspace interaction. The approach offers a scalable, real-world-transferable prior for complex humanoid control, reducing reward engineering and improving coordination across limbs.

Abstract

Humans possess a large reachable space in the 3D world, enabling interaction with objects at varying heights and distances. However, realizing such large-space reaching on humanoids is a complex whole-body control problem and requires the robot to master diverse skills simultaneously-including base positioning and reorientation, height and body posture adjustments, and end-effector pose control. Learning from scratch often leads to optimization difficulty and poor sim2real transferability. To address this challenge, we propose Real-world-Ready Skill Space (R2S2). Our approach begins with a carefully designed skill library consisting of real-world-ready primitive skills. We ensure optimal performance and robust sim2real transfer through individual skill tuning and sim2real evaluation. These skills are then ensembled into a unified latent space, serving as a structured prior that helps task execution in an efficient and sim2real transferable manner. A high-level planner, trained to sample skills from this space, enables the robot to accomplish real-world goal-reaching tasks. We demonstrate zero-shot sim2real transfer and validate R2S2 in multiple challenging goal-reaching scenarios.

Unleashing Humanoid Reaching Potential via Real-world-Ready Skill Space

TL;DR

This work tackles the challenge of enabling humanoid robots to reach a large 3D workspace by decoupling multi-skill control from end-to-end optimization. It introduces Real-world-Ready Skill Space (R2S2), a framework that builds a library of primitive skills, ensembles them into a unified latent space via a CVAE, and uses a high-level planner to sample skills for robust sim2real transfer. Extensive experiments on autonomous WBC tasks and real-world transfer demonstrate superior task success and stability over baselines, with a teleoperation system leveraging the latent skill space for large-workspace interaction. The approach offers a scalable, real-world-transferable prior for complex humanoid control, reducing reward engineering and improving coordination across limbs.

Abstract

Humans possess a large reachable space in the 3D world, enabling interaction with objects at varying heights and distances. However, realizing such large-space reaching on humanoids is a complex whole-body control problem and requires the robot to master diverse skills simultaneously-including base positioning and reorientation, height and body posture adjustments, and end-effector pose control. Learning from scratch often leads to optimization difficulty and poor sim2real transferability. To address this challenge, we propose Real-world-Ready Skill Space (R2S2). Our approach begins with a carefully designed skill library consisting of real-world-ready primitive skills. We ensure optimal performance and robust sim2real transfer through individual skill tuning and sim2real evaluation. These skills are then ensembled into a unified latent space, serving as a structured prior that helps task execution in an efficient and sim2real transferable manner. A high-level planner, trained to sample skills from this space, enables the robot to accomplish real-world goal-reaching tasks. We demonstrate zero-shot sim2real transfer and validate R2S2 in multiple challenging goal-reaching scenarios.
Paper Structure (18 sections, 10 equations, 4 figures, 4 tables)

This paper contains 18 sections, 10 equations, 4 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Works
  3. Unleashing Humanoid Reaching Potential via Real-world-Ready Skill Space
  4. Primitive Skill Library
  5. Heterogeneous Skill Ensembling
  6. High-Level Planning in Real-World-Ready Skill Space
  7. Experiments
  8. Performance of $\text{R}^{2}\text{S}^{2}$ on Whole-Body Control Tasks
  9. Experiment Setting
  10. Experiment Metrics
  11. Baselines
  12. Experiment Results
  13. Evaluation of Real-world-Ready Skill Space
  14. Experiment Setting and Metrics
  15. Baselines
  16. ...and 3 more sections

Figures (4)

  • Figure 1: (a) The humanoid showcases multiple real-world-ready primitive skills, including locomotion and body-pose-adjustment. (b) The humanoid autonomously accomplishes various WBC tasks. (c) The humanoid performs various tasks under our proposed teleoperation system.
  • Figure 2: We present $\text{R}^{2}\text{S}^{2}$, a structural skill prior that helps autonomous WBC task execution in an efficient and sim2real transferable manner.
  • Figure 3: We compare the motor behaviors with baseline methods. Our method significantly helps sim2real transfer.
  • Figure 4: We compare the normalized skill command tracking accuracy.