Glove captureSetup 2Premium signal

Glove Hand Capture

A higher-value setup for collecting precise hand, finger, and manipulation data using instrumented data gloves. Designed for robotics teams that need grasp sequences, finger pose, wrist orientation, and object interaction data for dexterous manipulation and imitation learning pipelines.

Equipment

What the setup includes

  • Data gloves (both hands)

    finger joint and wrist angle sensors

  • Hand and finger motion capture

    per-joint DOF at 30–100 Hz

  • Task and object tagging

    via companion app or annotator

  • Optional validation camera

    RGB cross-reference for QC

  • Structured export pipeline

    URDF-compatible or custom schema

Best for

Ideal data types

  • Finger pose estimation datasets
  • Hand pose and wrist orientation
  • Grasp sequence collection
  • Repetitive manipulation episodes
  • Object-specific interaction tasks
  • Teleop-compatible hand data
  • Dexterous task demonstrations
  • Imitation learning priors

Industry applications

Where glove capture is deployed

Glove Hand Capture is used in environments where fine-grained hand and finger motion is the primary signal — assembly, manipulation, teleoperation, and dexterous task research.

Packaging & assembly

lid closing, part insertion, fastener tightening

Manipulation research

grasp taxonomy, in-hand re-orientation, transfer

Repetitive hand workflows

folding, wrapping, stacking, sorting by hand

Repair-like tasks

connector insertion, screw driving, cable routing

Teleoperation training

bimanual control, trajectory priors, recovery

Dexterous task pilots

tool use, precision grasp, compliant contact

Data structure

Episode metadata and frame schema

Each episode contains structured metadata and per-frame joint data. The schema can be extended with buyer-specific fields, object taxonomies, and annotation layers.

FieldTypeDescription
episode_idstringunique episode identifier
task_labelstringtask type from defined taxonomy
outcomeenumsuccess | fail | partial
duration_msintegerepisode duration in milliseconds
hand_dataarray<frame>per-frame joint angles, positions, velocities
joint_dofintegerdegrees of freedom captured (glove-dependent)
sample_rate_hzintegerglove sampling rate for session
object_idstring | nulltagged object interacted with
validation_video_pathstring | nullRGB cross-reference if used
qc_statusenumaccepted | rejected | review
qc_notesstring | nullrejection reason if applicable
custom_fieldsobjectbuyer-specific metadata schema

Frame arrays contain per-joint angle, position, and velocity. Sampling rate is configurable per program (30–100 Hz).

Integration

Robotics framework compatibility

Glove capture data is exported in formats compatible with common robotics training pipelines. Custom delivery structures are supported per program.

URDF export

Joint angle data exportable in URDF-compatible format

ROS-compatible

Episode streams compatible with ROS message formats

LeRobot / OpenPI

Dataset format alignment available on request

HDF5 / NumPy

Raw per-frame arrays in open scientific formats

Custom schema

Buyer-defined fields and delivery structure supported

Tradeoffs

Advantages and limitations

Advantages

  • Precise per-joint angle and pose data
  • High-value signal for manipulation models
  • Stronger prior for imitation learning
  • Compatible with URDF and robotics export formats
  • Ideal for dexterous and teleop datasets
  • Premium batch potential for robotics buyers

Limitations

  • Higher equipment cost than video-only setups
  • Longer contributor onboarding and calibration
  • Setup more sensitive to environment constraints
  • Not required for workflow or task video use cases

For task workflows not requiring fine-grained hand data, see Video Task Capture.

Quality control

Default acceptance criteria

Each glove capture episode passes a multi-point QC review before inclusion in the accepted batch. Criteria are extended per buyer spec.

  • Glove correctly calibrated before session
  • All required joints tracked throughout episode
  • Task fully captured start to finish
  • Object tag correctly applied
  • No tracking loss or sensor dropout
  • Per-frame data within expected DOF range
  • Outcome label validated against episode
  • Validation camera angle correct (if used)
  • Sample rate consistent throughout session
  • Metadata fields complete and schema-valid

Start a glove capture pilot

Tell us the target manipulation task, object types, and required joint resolution. We will scope a pilot and propose a collection protocol.