DIGIT tactile sensor: from RGB frames to robot touch

The paper provides a compact hardware reference. Use the manufacturer documentation for the revision you acquire, and record the image format, lighting, gel condition, and capture timestamps used in your experiment.

The official package is named digit-interface. Its documented workflow lists attached sensors using DigitHandler.list_digits(), connects a Digit instance by the serial number printed on the device, and retrieves an image with get_frame(). The example viewer uses show_view(). Follow the repository installation and device-access instructions for your operating system.

Before collecting a dataset, inspect Digit.STREAMS and choose a supported resolution and frame rate explicitly. Save the serial number, stream format, illumination settings, unloaded reference image, gel replacement history, and robot timestamps with the recordings. This makes later differences between sensors and sessions interpretable.

The interface and design repositories are archived as checked on September 12, 2026. Their instructions remain useful reference material, but compatibility with a current Python, operating system, camera stack, or replacement component needs local verification. We have reviewed the documentation rather than run a connected-sensor test.

The image can reveal where the gel contacts an object and how its appearance changes as contact moves. A depth map, force vector, slip decision, or grasp-success estimate is a separate model output. Evaluate that model using independent reference measurements appropriate to the claimed quantity.

For geometric reconstruction, specify the reference objects and the part of the gel included in calibration. For force estimation, synchronize a force reference with the images. For a manipulation policy, compare trials on the same robot, objects, controller budget, and success criterion. Camera FPS is not the latency of that complete feedback loop.

The digit-design repository contains enclosure, elastomer, electronics, and firmware-binary resources, including a manufacturing quick-start document and component-substitution notes. Check those notes against the exact board revision before ordering parts.

Both the reviewed design and Python-interface repositories carry Creative Commons Attribution-NonCommercial 4.0 terms. Publicly available files should not be treated as unrestricted commercial-use software or hardware designs. Commercial sensor purchase and reuse of repository material are separate questions.

The official project also links TACTO, a tactile simulator. Simulated observations can support development, but their appearance and contact dynamics do not validate a physical sensor calibration or a robot policy. Keep simulation and physical results separate in your records.

DIGIT is relevant when a compact visual fingertip and an established research interface fit the robot geometry. GelSight Mini offers a separate commercial optical-sensor workflow and replaceable gel cartridge. ReSkin uses magnetic measurements from a replaceable elastomer, producing a different observation space.

Choose using mounting geometry, signal requirements, calibration effort, component access, and maintainable software. A camera pixel count cannot be compared directly with a magnetic channel count as a universal measure of tactile quality. The linked sensor guides preserve these distinctions.

The calibration workflow separates no-contact image baselines, reference geometry for depth reconstruction and independently measured force targets. Use the blank record to retain units, repeats, environmental conditions and invalid samples. The workflow is source-based; no new hardware measurements are claimed.