UDEXREAL UDCAP DataGlove
A flexible-sensor capture system whose documented coordinate conventions matter when translating demonstrations into robot commands.
DROID RESEARCH · SOURCES CHECKED
Why can UDCAP angles look correct in a viewer but map incorrectly to a robot?
HandDriver’s documentation distinguishes version-specific axes and distinguishes angle streams from quaternion streams. Establish the release, coordinate system and representation before retargeting.
What the sources establish
Manufacturer documentation is attributed below. These values are not independent measurements by Droid.
| Field | Reported value | Conditions and evidence |
|---|---|---|
| Tracking approach | Flexible sensors | Manufacturer description; exact sensor revision must be confirmed for the ordered unit. UDEXREAL ↗ |
| Reported minimum resolution | ≤0.1° | Resolution claim, not validated absolute accuracy or repeatability. UDEXREAL ↗ |
| Data representations | Joint angles and quaternions | The documentation assigns different coordinate conventions to these streams. UDEXREAL ↗ |
| Version boundary | HandDriver 2.1.2 hand model | The page contrasts this model with legacy conventions. Preserve version tags in collected datasets. UDEXREAL ↗ |
| Host software | HandDriver: Windows and Linux Ubuntu | The robotics guide recommends HandDriver Web. Exact OS release and processor support remain to be confirmed. UDEXREAL ↗ |
| Robot-hand mapping | Matching feature in v2.2.1 Windows / Linux Web | Supplier-provided hand identifiers select adapted output. A robot motion-control program is not included by default. UDEXREAL ↗ |
| ROS 2 interface | Not established in reviewed sources | A C++/Python SDK heading does not by itself establish a supported ROS 2 package. Evidence status: unknown |
| Finger force feedback | Not established in reviewed sources | No resistive finger-force capability established by the sources reviewed for this record. Evidence status: unknown |
What this means for an integration
Droid’s analysis of the cited documentation:
- Before recording a large dataset, capture a small pose fixture: open hand, closed hand, thumb opposition and isolated finger motions. Save the expected sign and axis for each motion.
- Treat an SDK or HandDriver change as a dataset migration. Replay the fixture before combining old and new sessions; a visually plausible pose can hide a frame or sign mismatch.
- Keep angular resolution out of rankings against millimeter fingertip accuracy. They measure different quantities and cannot establish which glove yields better robot demonstrations.
Still to establish
- The exact glove revision supplied through Droid’s offer
- Native ROS 2 distribution and package support
- Delivered price, lead time and complete tracking accessories
- Independent accuracy and end-to-end latency under load
Before a larger capture run
- Record HandDriver and SDK versions with every episode.
- Validate both hands against a known pose fixture.
- Verify the coordinate transform at the robot boundary.
- Confirm the supplied hardware matches the quoted specification.
Source record
- Hand model and data specification ↗UDEXREAL · HandDriver 2.1.2 onward and legacy conventions · Checked 2026-11-10
- UDCAP DataGlove introduction ↗UDEXREAL · UDCAP; undated documentation · Checked 2026-11-10
- Robotics workflow and HandDriver adaptation ↗UDEXREAL · Includes HandDriver v2.2.1 Windows / Linux Web matching feature · Checked 2026-11-10
Make the next decision.
START WITH THE WORKFLOWBring us the task.
Request a project review Bring us the task.
Let’s find the fit.
Share your environment, the job to be done, and the outcome you need to measure.