ViTacFormer: Learning Cross-Modal Representation for Visuo-Tactile Dexterous Manipulation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Heng, Liang, Geng, Haoran, Zhang, Kaifeng, Abbeel, Pieter, Malik, Jitendra
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916005717999616
author Heng, Liang
Geng, Haoran
Zhang, Kaifeng
Abbeel, Pieter
Malik, Jitendra
author_facet Heng, Liang
Geng, Haoran
Zhang, Kaifeng
Abbeel, Pieter
Malik, Jitendra
contents Dexterous manipulation is a cornerstone capability for robotic systems aiming to interact with the physical world in a human-like manner. Although vision-based methods have advanced rapidly, tactile sensing remains crucial for fine-grained control, particularly in unstructured or visually occluded settings. We present ViTacFormer, a representation-learning approach that couples a cross-attention encoder to fuse high-resolution vision and touch with an autoregressive tactile prediction head that anticipates future contact signals. Building on this architecture, we devise an easy-to-challenging curriculum that steadily refines the visual-tactile latent space, boosting both accuracy and robustness. The learned cross-modal representation drives imitation learning for multi-fingered hands, enabling precise and adaptive manipulation. Across a suite of challenging real-world benchmarks, our method achieves approximately 50% higher success rates than prior state-of-the-art systems. To our knowledge, it is also the first to autonomously complete long-horizon dexterous manipulation tasks that demand highly precise control with an anthropomorphic hand, successfully executing up to 11 sequential stages and sustaining continuous operation for 2.5 minutes.
format Preprint
id arxiv_https___arxiv_org_abs_2506_15953
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ViTacFormer: Learning Cross-Modal Representation for Visuo-Tactile Dexterous Manipulation
Heng, Liang
Geng, Haoran
Zhang, Kaifeng
Abbeel, Pieter
Malik, Jitendra
Robotics
Dexterous manipulation is a cornerstone capability for robotic systems aiming to interact with the physical world in a human-like manner. Although vision-based methods have advanced rapidly, tactile sensing remains crucial for fine-grained control, particularly in unstructured or visually occluded settings. We present ViTacFormer, a representation-learning approach that couples a cross-attention encoder to fuse high-resolution vision and touch with an autoregressive tactile prediction head that anticipates future contact signals. Building on this architecture, we devise an easy-to-challenging curriculum that steadily refines the visual-tactile latent space, boosting both accuracy and robustness. The learned cross-modal representation drives imitation learning for multi-fingered hands, enabling precise and adaptive manipulation. Across a suite of challenging real-world benchmarks, our method achieves approximately 50% higher success rates than prior state-of-the-art systems. To our knowledge, it is also the first to autonomously complete long-horizon dexterous manipulation tasks that demand highly precise control with an anthropomorphic hand, successfully executing up to 11 sequential stages and sustaining continuous operation for 2.5 minutes.
title ViTacFormer: Learning Cross-Modal Representation for Visuo-Tactile Dexterous Manipulation
topic Robotics
url https://arxiv.org/abs/2506.15953