Learning Tactile-Aware Quadrupedal
Loco-Manipulation Policies

Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies

Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies

Pokuang Zhou1, Yuhao Zhou1, Quan Khanh Luu1, Seungho Han1, Heng Zhang1,2, Binghao Huang3, Yunzhu Li3, Arash Ajoudani2, Zhengtong Xu1,†, Yu She1,†

1Purdue University, West Lafayette, IN, USA  |  2Istituto Italiano di Tecnologia, Genoa, Italy  |  3Columbia University, New York, NY, USA
† Corresponding Author

Purdue University Istituto Italiano di Tecnologia Columbia University

IROS 2026

arXiv

Learning Tactile-Aware Hierarchical Whole-Body Control Policies for Quadrupedal Loco-Manipulation

Pokuang Zhou1,*, Le Mao2,*, Xiyue Chen3,*, Zhiyuan Zhang1, Joseph Campbell2, Yu She1,†

1Industrial Engineering School  |  2Computer Science School  |  3Electrical and Computer Engineering School
Purdue University, West Lafayette, IN, USA
* Equal contribution  |  † Corresponding Author

Purdue University

Coming Soon

Poster

Your browser cannot display the poster preview. Open the poster PDF.

Open full poster

Code, Pretrained Checkpoints, and Data

Code and Hugging FaceData will be fully open-sourced once the journal article is published.
We have also released the WBC training logs, with the results reproduced across multiple random seeds and GPUs. Weights & Biaseslink
Preview version code: including simulation and real-world deployment. 📂link

Clone code:

bash 
git clone --depth 1 --filter=blob:none --no-checkout https://github.com/PokuangZhou/tactile-aware-quadrupedal-loco-manipulation.git && cd tactile-aware-quadrupedal-loco-manipulation && git sparse-checkout init && git sparse-checkout set "/code/" && git checkout main

Tactile-Guided Quadrupedal Active Loco-Manipulation with Physical 3D Point Clouds

Pokuang Zhou1,*, Le Mao2,*, Yu She1

1Industrial Engineering School  |  2Computer Science School
Purdue University, West Lafayette, IN, USA
* Equal contribution

Purdue University
Robot dog equipped with tactile whisker sensors

This project enables quadruped robots to perceive physical properties of their surroundings, such as stiffness and compliance, in the wild through tactile sensing and construct physically grounded 3D point clouds.

BibTeX

@article{zhou2026learning,
  title={Learning Tactile-Aware Quadrupedal Loco-Manipulation Policies},
  author={Zhou, Pokuang and Zhou, Yuhao and Luu, Quan Khanh and Han, Seungho and Zhang, Heng and Huang, Binghao and Li, Yunzhu and Ajoudani, Arash and Xu, Zhengtong and She, Yu},
  journal={arXiv preprint arXiv:2604.27224},
  year={2026}
}