DITTO: Dexterous Interface for Transparent TeleOperation

Joaquin Palacios, Katelyn Lee, Cheng Zhang, Zhanpeng He, Matei Ciocarlie · N/A · 2026

Summary

Collecting data for manipulation with high-DOF hands is challenging, as interfaces must capture rich hand motion while rendering the contact interactions essential for precise manipulation.

Abstract Summary

Collecting data for manipulation with high-DOF hands is challenging, as interfaces must capture rich hand motion while rendering the contact interactions essential for precise manipulation. Existing data collection approaches face a trade-off: teleoperation ensures deployment consistency but lacks force feedback, while handheld (in-the-wild) systems provide natural force transparency but introduce a visual embodiment gap at deployment. We present DITTO, a Dexterous Interface for Transparent TeleOperation, which resolves this through the anatomically informed co-design of a dexterous 7-DOF robotic hand and a kinematically equivalent motorized exoskeleton. A 1-to-1 actuator mapping between the exoskeleton and robotic hand enables handheld (in-the-wild) data collection and bilateral teleoperation with joint-level force feedback unified in a single platform. We demonstrate that the DITTO exoskeleton spans the operator's natural index-to-thumb workspace, and showcase DITTO's dexterous capabilities via learned policies on contact-rich tasks.

Key Points

Collecting data for manipulation with high-DOF hands is challenging, as interfaces must capture rich hand motion while rendering the contact interactions essential for precise manipulation.
Existing data collection approaches face a trade-off: teleoperation ensures deployment consistency but lacks force feedback, while handheld (in-the-wild) systems provide natural force transparency but introduce a visual embodiment gap at deployment.
We present DITTO, a Dexterous Interface for Transparent TeleOperation, which resolves this through the anatomically informed co-design of a dexterous 7-DOF robotic hand and a kinematically equivalent motorized exoskeleton.
A 1-to-1 actuator mapping between the exoskeleton and robotic hand enables handheld (in-the-wild) data collection and bilateral teleoperation with joint-level force feedback unified in a single platform.
We demonstrate that the DITTO exoskeleton spans the operator's natural index-to-thumb workspace, and showcase DITTO's dexterous capabilities via learned policies on contact-rich tasks.

DITTO Hardware Design

DITTO aims to bridge the fundamental trade-off between kinematic and force transparency in teleoperation by co-designing a dexterous 7-DOF hand and a kinematically equivalent motorized wearable exoskeleton. This leader-follower system has a 1:1 actuator mapping, where every motor on the follower hand corresponds to a motor on the leader exoskeleton.

This concept has been demonstrated for arm teleoperation [ 13 , 14 ] but not for high-DOF hands in a wearable form factor.

DITTO Hardware Design — DITTO: Dexterous Interface for Transparent TeleOperation — Fig. 2: (a) DITTO aligns each of its 7 motor axes with a finger joint axis. In blue, the 2-DoF CMC and MCP joints have offset, decoupled motors so the motor axes intersect at the anatomical joint center. (b) DITTO has 1:1 kinematic equivalence between the leader exoskeleton and the follower hand. (c) The offset, decoupled motors for the CMC enable dexterous thumb motions like abduction and opposition.

User Study: Dexterous Data Collection

When used in leader-follower teleoperation mode, DITTO aims to provide active force feedback while still allowing rich tip-to-tip movement for dexterous tasks. To assess the importance of these features, we conducted a user study ($n=8$) comparing DITTO teleoperation against two ablations and across two tasks that require coordinated fingertip dexterity and force regulation.

We compare three teleoperation conditions : DITTO exoskeleton: leader-follower teleoperation using DITTO exoskeleton with force feedback enabled.

User Study: Dexterous Data Collection — DITTO: Dexterous Interface for Transparent TeleOperation — Fig. 3: (a) The three bolded lines in the ECDF plot denote the gross average over all $n=8$ subjects per teleop condition, and the faint lines denote the individual subject curves. (b) The stacked bar plot tallies the total number of mahjong tile flips over the three trials, in order, with teleop condition labeled by color and trial number denoted by opacity.

Kinematic Workspace Analysis

Matching human index-to-thumb dexterity is a key design goal for DITTO providing two important benefits. First, from the perspective of the DITTO leader (a wearable exoskeleton), a human-like workspace removes constraints on the teleoperator’s hand to allow natural and intuitive movement without pressure points.

Second, from a follower perspective, human-like tip-to-tip movement implies high dexterity, since the human hand can perform complex manipulation with just index-to-thumb grasps exclusively.

Kinematic Workspace Analysis — DITTO: Dexterous Interface for Transparent TeleOperation — Fig. 4: Top: Kinematic analysis data collection and data processing workflow. Bottom: Example workspace results visualized, targets colored by IK solution quality, from green (within tolerance) to red (error exceeds threshold).

Policy Learning Results

Finally, we evaluate weather DITTO data-collection enables autonomous policies for tasks requiring use of the intrinsic hand DOFs, as well as regulation of contact forces with the environment. We test these abilities for both data collection modes offered by DITTO, teleoperation and handheld, as well as combined datasets from both paradigms.

For the autonomous policies, we use standard Diffusion Policies with a Transformer Architecture [ 17 ] using the relative pose representation from [ 18 ] .

Policy Learning Results — DITTO: Dexterous Interface for Transparent TeleOperation — Fig. 5: Learned policies for dexterous and contact-rich manipulation tasks. For unimanual tasks, we include: USB Drive Insertion , which requires picking up and extending a retractable USB flash drive and then inserting it into a slot, Mahjong Tile Flipping , where the robot reorients a tile through two sequential flips followed by placement into a target box, and Key Rotation to engage a lock. For the bimanual task, we include Rubik’s Cube Rotation , which requires picking up and then rotating two faces of a Rubik’s Cube using thumb-index finger manipulation. Videos at: https://ditto-robotics.github.io/

Sources and Demonstrations

Figures are reproduced from Palacios et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.