CyberDemo: Augmenting Simulated Human Demonstration for Real-World Dexterous Manipulation

Jun Wang1*, Yuzhe Qin1*, Kaiming Kuang1, Yigit Korkmaz2, Akhilan Gurumoorthy1, Hao Su1, Xiaolong Wang1
1UC San Diego; 2USC
CVPR 2024
*Indicates Equal Contribution
Paper Code Video

We introduce CyberDemo, a novel approach to robotic imitation learning that leverages simulated human demonstrations for real-world tasks.

Abstract

By incorporating extensive data augmentation in a simulated environment, CyberDemo outperforms traditional in-domain real-world demonstrations when transferred to the real world, handling diverse physical and visual conditions. Regardless of its affordability and convenience in data collection, CyberDemo outperforms baseline methods in terms of success rates across various tasks and exhibits generalizability with previously unseen objects. For example, it can rotate novel tetra-valve and penta-valve, despite human demonstrations only involving tri-valves. Our research demonstrates the significant potential of simulated human demonstrations for real-world dexterous manipulation tasks.

                Motivation

                Video Presentation

Method

MY ALT TEXT

CyberDemo Pipeline. First, we collect both simulated and real demonstrations via vision-based teleoperation. Following this, we train the policy on simulated data, incorporating the proposed data augmentation techniques. During training, we apply automatic curriculum learning, which incrementally enhances the randomness scale based on task performance. Finally, the policy is fine-tuned with a few real demos before being deployed to the real world.

Real World Deployments

Pouring Task

Rotating Task

Pick And Place Task

BibTeX

@inproceedings{wang2024cyberdemo,
        title={CyberDemo: Augmenting Simulated Human Demonstration for Real-World Dexterous Manipulation},
        author={Wang, Jun and Qin, Yuzhe and Kuang, Kaiming and Korkmaz, Yigit and Gurumoorthy, Akhilan and Su, Hao and Wang, Xiaolong},
        booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
        pages={17952--17963},
        year={2024}
      }

Acknowledgement

We gratefully acknowledge support from the Technology Innovation Program (20018112, Development of autonomous manipulation and gripping technology using imitation learning based on visual and tactile sensing) funded by the Ministry of Trade, Industry & Energy (MOTIE), Korea.