Model library
Robot Tactile Sensor Models: Specs and Selection
A robot tactile sensor is a device that measures contact at a robot surface and exposes a signal for perception or control. Robot tactile sensor selection depends on the contact geometry, force range, sampling path and integration interface. The model library below preserves the source-review date for each record and separates published specifications from missing information.
Last updated:
Part of tactile sensing: compare sensing principles before selecting a model or end effector.
Published | Updated by RoboSkin.ai Editorial Team
November 4, 2026: reviewed model-library interpretation and added a specification-to-interface handoff. Existing 17 records retain their individual source dates; no device testing or live stock check.
- 3
- sections
- 2
- questions
- 3
- next routes
17 sensor models tracked
Tracked manufacturer and research models. Source-reviewed records retain their actual evidence dates. Manual review follows a checked release or correction; no automatic refresh. Unknown values remain Not published.
| Model | Type | Force range | Sampling rate | Interface | Vendor / group | Source / reviewed |
|---|---|---|---|---|---|---|
| Loomia Smart Skin | Flexible capacitive and resistive sensing | Not published | Component-specific; see the dated profile specification table. | Not published | Loomia Technologies | Loomia Smart Skin source |
| Robotiq TSF-85 | Capacitive tactile array with dynamic and inertial sensing | Not published | Pressure array: 60 Hz. Dynamic signal: 1000 Hz. The SDK describes a 1 kHz USB stream per finger; transport rate is not pressure refresh rate. | USB; C++ SDK and Python tools | Robotiq | Robotiq TSF-85 source |
| Melexis Tactaxis | Magnetic tactile sensing with a Triaxis Hall-effect IC | 5 N normal / 2 N shear; extended 15 N range at reduced accuracy (2025 taxel) | Not published. Not publicly specified in the reviewed Melexis e-guide or 2026 module announcement. | Not published | Melexis | Melexis Tactaxis source |
| ViTai VT-GF225 | Marker-based vision tactile sensing with a compliant gel interface | 30 N maximum normal and tangential force (manufacturer-reported) | 30 Hz for the standard version according to the reviewed manufacturer page | USB 2.0 | ViTai Robotics | ViTai VT-GF225 source |
| DIGIT | Vision-based tactile sensing | Not published | The 2020 paper reports 60 FPS at 640 × 480; the official Python interface defaults to 640 × 480 at 30 FPS. Verify the stream supported by the actual unit. | USB camera; Python digit-interface | Meta AI Research / GelSight | DIGIT source |
| Digit 360 | Multimodal vision-based tactile sensing | Not published | Not published. Not publicly stated on the reviewed repository overview | Not published | Meta AI Research / GelSight | Digit 360 source |
| GelSight Mini | Vision-based tactile sensing | Not published | 25 FPS in the reviewed official datasheet | USB; image acquisition and reconstruction software | GelSight | GelSight Mini source |
| ReSkin | Magnetic tactile skin | Not published | Up to 400 Hz reported by Meta | Magnetometer acquisition; version-specific hardware | Meta AI Research / Carnegie Mellon University | ReSkin source |
| AnySkin | Magnetic tactile skin | Not published | Not published. Not publicly stated on the reviewed project page | Not published | New York University / Carnegie Mellon University / Columbia University / Meta | AnySkin source |
| GelSlim 4.0 | Vision-based tactile sensing | Not published | Not published. Not publicly stated in the reviewed abstract | Not published | University of Michigan MMint Lab | GelSlim 4.0 source |
| TacTip | Biomimetic optical tactile sensing | Not published | Depends on the camera and TacTip variant; no single rate is stated for the full family | Not published | Bristol Robotics Laboratory / University of Bristol | TacTip source |
| BioTac | Fluid-filled biomimetic multimodal sensing | Not published | No single cross-modality rate is used in this directory; BioTac signal channels and hardware revisions must be checked separately. | Not published | SynTouch | BioTac source |
| uSkin | Distributed magnetic three-axis force sensing | Not published | Up to 500 Hz for the uSPa 11 model in the reviewed catalog | Not published | XELA Robotics | uSkin source |
| DenseTact 2.0 | Vision-based optical tactile sensing | Not published | Not published. Not publicly stated in the reviewed abstract | Not published | Stanford University ARM Lab | DenseTact 2.0 source |
| 9DTact | Vision-based tactile sensing with translucent gel | Not published | Not published. Not publicly stated on the reviewed project page | Not published | Shanghai Qi Zhi Institute / Tsinghua University / HUST / Shanghai AI Laboratory | 9DTact source |
| Insight | Vision-based haptic sensing | Not published | Not published. Not publicly stated on the reviewed project page | Not published | Max Planck Institute for Intelligent Systems | Insight source |
| AllSight | Vision-based optical tactile sensing | Not published | Not published. Not publicly stated in the reviewed abstract | Not published | Ben-Gurion University research team | AllSight source |
Short answer
What you need to know
- 1
The library contains 17 existing sensor records. It is a tracked set, not a complete market census.
- 2
Specifications are taken from the linked manufacturer documentation or original research sources. No unreported rate, interface or force range is inferred from a related product.
- 3
Review is manual and event-driven: check a record when its source release changes or a correction arrives. There is no automatic refresh; the date on each row records the evidence actually reviewed.
Topic 01
Turn a specification row into a usable interface requirement
The 17 records below are dated source summaries. A page-level revision does not mean every manufacturer specification has been freshly measured or that every device is currently in stock. Begin with the exact model and revision, then follow the linked profile or primary source for the conditions behind a field.
The DIGIT and TSF-85 records illustrate why a single rate column is not enough. DIGIT’s paper and interface describe camera acquisition settings; TSF-85 distinguishes pressure-array refresh, dynamic sensing and USB transport. Neither field alone is the rate of a calibrated force estimate or the bandwidth of an external robot controller.
| Requirement | Record from the exact configuration | Reject this shortcut |
|---|---|---|
| Quantity and units | Raw image/count/field or calibrated force, pressure or geometry | Interpreting a numeric SDK output as newtons without a calibration contract |
| Spatial footprint | Actual active area, element layout, blind regions and contact geometry | Comparing image pixels directly with independent three-axis sensing elements |
| Timing | Acquisition configuration, complete sample interval, timestamps and processing delay | Using interface throughput as full-array refresh or action frequency |
| Mechanical compatibility | Mount, cover, allowable deformation and replacement method | Transferring a flat-sample result to a curved mounted surface unchanged |
| Software and reuse | Supported hardware revision, pinned driver/model and relevant terms | Assuming a public repository licenses every asset or supports every variant |
Resolve unknown fields before making them requirements
Not published in this directory means the reviewed evidence did not establish that field. It is neither a zero value nor proof that a supplier has never documented it. Record the exact question and source revision, and leave the field unresolved until documentation or a measurement addresses it.
For a force-based task, specify loading direction, contact geometry, calibration range and acceptable error before comparing candidates. For a slip task, define an independently labelled slip event and acceptable false alarms and delays. The correct choice follows the required observation and failure cost, not the largest number in a specification table.
Topic 02
Choose the model after defining the measurement
Start with the required contact footprint and signal: a taxel array, a reconstructed surface, an estimated force or a dynamic vibration channel. Compare normal and shear force only when calibration, loading geometry and units match.
An image frame rate, USB transport rate and calibrated force update rate are different values. If the source does not establish the quantity required by your controller, record it as an unresolved integration requirement.
| Field | What to compare | What not to assume |
|---|---|---|
| Type | Transduction principle and contact surface geometry | All sensors in one family have the same spatial resolution |
| Force range | Normal and shear limits with the calibration conditions | A raw pressure count is calibrated force |
| Sampling rate | Sensor refresh, transport and processing rate separately | A fast transport stream proves the same closed-loop bandwidth |
| Interface | Electrical connection, API and timestamp behaviour | A public repository supports every hardware revision |
| Vendor or research group | Exact model, release and available documentation | A laboratory prototype is currently orderable |
Topic 03
Verify a candidate in the intended contact task
Use a reference load and a repeatable fixture to record the unloaded baseline, loading and unloading curves, repeatability and drift. Check timestamp alignment with the robot state before comparing a policy result.
For slip detection, use independently labelled slip events. For insertion, record geometry and contact-force limits. Keep these task measurements alongside the model revision and mounting details so another run can reproduce the conditions.
Common questions
FAQ for this topic
Why do some sensor fields say Not published?
No applicable value was established in the reviewed source. This does not mean zero capability, nor that the manufacturer has no additional private documentation.
Are the model rows ranked?
No. They have different contact areas, sensing outputs and test protocols. The table is a selection record, not a shared performance leaderboard.