Vision Based Tactile Sensor Selection Record

Vision Based Tactile Sensor Selection Record

Selection rule

Choose a vision-based tactile sensor for a defined contact and decision, not by global product ranking. The sensor must physically reach the contact, survive the load and environment, expose a signal with enough information, and fit the timing and maintenance budget.

Evidence matrix

Sensor-family evidence

The DIGIT paper describes a compact, high-resolution design intended for robotic in-hand manipulation and makes its design open source. That makes it relevant when fingertip packaging and research reproducibility matter.

GelSight Mini appears in the Sparsh benchmark for force estimation and pretraining data. GelSight 2017 appears in grasp-stability and textile tasks. These benchmark uses show compatibility with specific datasets and tasks. They do not rank the devices for every robot.

The current GelSight Mini product sheet is a vendor source for current packaging and product claims. Specifications, pricing, availability, and support must be checked on the day of procurement.

The DigiTac comparison is useful because it tests different optical tactile constructions in a shared robot system. Its central lesson is that similar pose-prediction results can still lead to different servo-control behavior because physical construction matters.

Pilot rule

Use the same mount quality, objects, contact paths, load envelope, sampling method, calibration effort, model capacity, metric, and failure record for every finalist. Report setup time, failed runs, cleaning, recalibration, and surface damage along with model scores.

The final decision should state why the selected sensor fits this task and which task, environment, or failure remains outside the evidence.

No sensor selection record is procurement approval. Current vendor terms, rights, electrical and mechanical integration, safety, and representative testing remain required.