Recent advances in robot learning for manipulation have increased the importance of collecting real-world demonstration data. However, existing robotic systems primarily focus on end-effector manipulation, making it difficult to teach and execute tasks involving body-surface contact with the arms and chest. We present TWINS, a robotic system composed of a Wearable Dual-Arm Device and an Isomorphic Robot with the same joint configuration and external dimensions. Distributed tactile sensors embedded in the chest and arms measure body-surface contact in sync with joint motion. Demonstrations collected for four contact-rich manipulation tasks are used to train imitation learning policies, enabling manipulation guided by body-surface tactile observations.
TWINS unifies embodied demonstration, isomorphic motion transfer, and distributed body-surface tactile sensing in a single platform for collecting and learning contact-rich manipulation.
01 · Wearable embodiment
The operator inserts their arms into the hollow wearable device and physically interacts with objects through its robot-shaped outer shell, enabling intuitive demonstration of body-surface contact.
02 · Isomorphic design
The wearable device and execution robot share the same joint configuration and external dimensions. Each arm has seven degrees of freedom and a parallel two-finger gripper, allowing demonstrated joint angles to transfer without complex retargeting.
03 · Distributed touch
Pressure and proximity are measured across 219 tactile cells placed on the chest, upper arms, forearms, and grippers—capturing contact beyond the end effectors.
We evaluate TWINS on four contact-rich tasks by collecting 10 demonstrations per task with the Wearable Dual-Arm Device and deploying imitation-learning policies trained with Diffusion Policy on the Isomorphic Robot.
All videos on this page are shown at real-time speed (1×).
Body-surface tactile measurements are visualized for both data collection and policy rollout.
Recorded joint angles and tactile measurements are shown with data collection on the left and the corresponding policy rollout on the right.
@misc{kitamura2026twins,
title = {TWINS: A Tactile Wearable Isomorphic Arm Networked System for Contact-Rich Manipulation Learning},
author = {Takahide Kitamura and Masaki Murooka and Natsuki Yamanobe and Yukiyasu Domae},
year = {2026},
eprint = {2608.01733},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2608.01733v1}
}