TouchScale human visual-tactile dataset
Bimanual tactile gloves, Head RGB-D camera, Wrist RGB cameras
Visual-tactile representation learning, Human contact supervision for later robot adaptation
Primary paper ->Preview the research questions and evaluation structure below, then download the fixed 2026-08-17 PDF edition with its source register.

What this is
This edition demonstrates how RoboSkin.ai organizes robot skin and tactile AI evidence. It does not include invented market sizing, paid rankings, hardware test results, or unsupported forecasts.
Work through the sensing-to-policy chain before shortlisting a technology. These questions organize a review; answers must come from the selected system and its primary evidence.
Name the robot surface, object, motion, and operating conditions. A fingertip experiment does not establish whole-body performance.
Map the robot applicationSeparate raw images, magnetic readings, or array values from derived geometry and force estimates.
Compare sensor recordsRecord the calibration target, reference measurement, mounting, and held-out checks.
Review calibrationCheck actual file access, license, modality, timestamps, and the separation of training and test trajectories.
Inspect dataset availabilityDistinguish representation learning, contact prediction, action selection, and low-level control.
Compare tactile model rolesCompare the same protocol with and without the tactile input. Include failures, interventions, and timing.
Review evaluation benchmarksReading guide reviewed November 4, 2026. The PDF edition and original research dates remain separate; no independent reproduction.
The downloadable PDF remains the August 17, 2026 edition. Its six dataset records and six research signals are a dated sample, not a claim to exhaust the field or describe current availability. The live directory preview below is a separate view. A new review date on this page does not update the PDF or turn its historical access notes into fresh checks.
One practical example is FreeTacMan. At the November 17, 2026 documentation check, the provider dataset card labels the data MIT, while the code repository’s root LICENSE and README identify Apache-2.0; the README qualifies this as unless specified otherwise. Preserve that source distinction: the sample’s short license label is not a single license for every file, dependency, model or downstream use. We reviewed public documentation, not downloaded dataset contents or executed software.
| When using the sample | Record separately | Decision it supports |
|---|---|---|
| Compare a historical entry with today’s directory | PDF edition, live entry review date and the changed field | Whether the original shortlist needs to be reconsidered |
| Follow a data or code link | Exact artifact, revision, access result and applicable terms | Whether further technical and rights review can proceed |
| Carry a research result into a proposal | Reported task, metric, baseline and untested conditions | Which local validation remains necessary |
| Request a focused research brief | Decision, constraints, deliverable and unresolved evidence | A scope to agree before scheduling work |
A list of source URLs helps trace a claim; it does not prove that every hosted artifact is available, that a policy transfers to a different robot, or that a supplier meets an acceptance criterion. Use the sample to structure those questions, then attach current evidence to the specific decision.
For a correction, identify the PDF edition, page or entry, original statement and proposed primary source. The historical file can then remain distinguishable from a later correction or edition. The current directory and editorial policy provide the routes for checking evidence and submitting a correction.
The records below come from the live directory and may differ from the 6-record PDF sample. The historical PDF is preserved unchanged; consult current directory entries for updated access and license evidence.
Bimanual tactile gloves, Head RGB-D camera, Wrist RGB cameras
Visual-tactile representation learning, Human contact supervision for later robot adaptation
Primary paper ->Simulated FlexiTac-style fingertip pads
Scripted simulated lifting, Contact-conditioned imitation learning research
Primary paper ->Active and passive thermal sensors, Normal-force sensing resistor, Contact microphone for vibration, Two IMUs
Material recognition, Compliance recognition
Primary paper ->Manikin pressure skin: 44 readings per frame, RGB-D cameras, Gaze and pose tracking
Caregiving activity perception, Human demonstration analysis
Primary paper ->Discuss a focused evidence package for one technical or commercial question. Scope, deliverables and timing are agreed before work starts.