Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2509.14688 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866909795109306368 |
|---|---|
| author | Xu, Yue Wei, Litao An, Pengyu Zhang, Qingyu Li, Yong-Lu |
| author_facet | Xu, Yue Wei, Litao An, Pengyu Zhang, Qingyu Li, Yong-Lu |
| contents | Tactile-aware robot learning faces critical challenges in data collection and representation due to data scarcity and sparsity, and the absence of force feedback in existing systems. To address these limitations, we introduce a tactile robot learning system with both hardware and algorithm innovations. We present exUMI, an extensible data collection device that enhances the vanilla UMI with robust proprioception (via AR MoCap and rotary encoder), modular visuo-tactile sensing, and automated calibration, achieving 100% data usability. Building on an efficient collection of over 1 M tactile frames, we propose Tactile Prediction Pretraining (TPP), a representation learning framework through action-aware temporal tactile prediction, capturing contact dynamics and mitigating tactile sparsity. Real-world experiments show that TPP outperforms traditional tactile imitation learning. Our work bridges the gap between human tactile intuition and robot learning through co-designed hardware and algorithms, offering open-source resources to advance contact-rich manipulation research. Project page: https://silicx.github.io/exUMI. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_14688 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | exUMI: Extensible Robot Teaching System with Action-aware Task-agnostic Tactile Representation Xu, Yue Wei, Litao An, Pengyu Zhang, Qingyu Li, Yong-Lu Robotics Tactile-aware robot learning faces critical challenges in data collection and representation due to data scarcity and sparsity, and the absence of force feedback in existing systems. To address these limitations, we introduce a tactile robot learning system with both hardware and algorithm innovations. We present exUMI, an extensible data collection device that enhances the vanilla UMI with robust proprioception (via AR MoCap and rotary encoder), modular visuo-tactile sensing, and automated calibration, achieving 100% data usability. Building on an efficient collection of over 1 M tactile frames, we propose Tactile Prediction Pretraining (TPP), a representation learning framework through action-aware temporal tactile prediction, capturing contact dynamics and mitigating tactile sparsity. Real-world experiments show that TPP outperforms traditional tactile imitation learning. Our work bridges the gap between human tactile intuition and robot learning through co-designed hardware and algorithms, offering open-source resources to advance contact-rich manipulation research. Project page: https://silicx.github.io/exUMI. |
| title | exUMI: Extensible Robot Teaching System with Action-aware Task-agnostic Tactile Representation |
| topic | Robotics |
| url | https://arxiv.org/abs/2509.14688 |