Skip to main content

Tactile Sensing: GelSight, DIGIT, and BioTac

Duration: 55 min · Level: Advanced · Module: 6. Dexterous Manipulation · Focus: tactile, sensing, GelSight, DIGIT

Learning objectives

By the end of this lesson you will be able to explain and apply:

  • GelSight (MIT, Tao et al. 2012)
  • DIGIT (Meta AI Research, Lambeta et al. 2020)
  • BioTac (SynTouch)
  • Slip detection
  • Manipulation capability enabled

Why this matters​

Tactile sensing is the missing modality in most robot hands — and it is critical for healthcare applications where gentle contact, detecting slip, and measuring contact forces are essential.

Overview​

Tactile sensing is the missing modality in most robot hands — and it is critical for healthcare applications where gentle contact, detecting slip, and measuring contact forces are essential. GelSight-class sensors use a gel-covered surface illuminated from within to image contact geometry at sub-millimeter resolution.

Key concepts​

Key idea

GelSight (MIT, Tao et al. 2012): gel elastomer + internal LEDs + camera; images the deformed gel surface; recovers 3D geometry of contact at 0.1mm resolution

  • DIGIT (Meta AI Research, Lambeta et al. 2020): compact GelSight redesigned for fingertip mounting; 20mm × 20mm × 24mm; 60 FPS tactile images; open-source design
  • BioTac (SynTouch): fluid-filled silicone fingertip with 24 electrodes; measures distributed pressure, vibration, and temperature; FDA-cleared for medical applications
  • Slip detection: tactile images reveal incipient slip (micro-motion before gross slip) 100-200ms before grasp failure; enables proactive grasp adjustment
  • Manipulation capability enabled: threading a needle (requires <0.5mm contact resolution), detecting full medication bottle vs empty (mass from grip force), gentle patient contact (<2N force)
  • Learning from tactile data: convolutional neural networks on tactile images predict contact forces, object geometry, and slip onset with >95% accuracy

Check your understanding​

Try to recall each answer before expanding it.

Q1. What do you know about GelSight (MIT, Tao et al. 2012)?

gel elastomer + internal LEDs + camera; images the deformed gel surface; recovers 3D geometry of contact at 0.1mm resolution

Q2. What do you know about DIGIT (Meta AI Research, Lambeta et al. 2020)?

compact GelSight redesigned for fingertip mounting; 20mm × 20mm × 24mm; 60 FPS tactile images; open-source design

Q3. What do you know about BioTac (SynTouch)?

fluid-filled silicone fingertip with 24 electrodes; measures distributed pressure, vibration, and temperature; FDA-cleared for medical applications

Q4. What do you know about Slip detection?

tactile images reveal incipient slip (micro-motion before gross slip) 100-200ms before grasp failure; enables proactive grasp adjustment

Q5. What do you know about Manipulation capability enabled?

threading a needle (requires <0.5mm contact resolution), detecting full medication bottle vs empty (mass from grip force), gentle patient contact (<2N force)

References​

  • DIGIT: A Novel Design for a Low-Cost Compact High-Resolution Tactile Sensor with Application to In-Hand Manipulation — Lambeta et al. (2020). IEEE Robotics and Automation Letters 2020

← Previous: 6.1 Hand Design: DOF, Actuation, and the 22-DOF Target · Next: 6.3 Grasp Planning with Foundation Models →

Part of Module 6: Dexterous Manipulation.