Tactile sensing

What it means

Tactile sensing provides information about physical contact at a robot's surface. A sensor may help locate a contact, estimate loading or detect changes suggesting slip. The information complements vision because it describes what happens where the robot and an object touch. The DIGIT research paper identifies contact-force sensing as a central difficulty in precise manipulation. [2]

An optical tactile sensor can use a camera without being a scene camera. GelSight measures deformation of an elastomer with a reflective membrane to recover contact geometry. Its improved design also investigates slip detection using relative motion and shear distortion of the sensing surface. [1]

How it applies to humanoids

Fingertip sensing can provide feedback during grasping and object manipulation. DIGIT was designed to fit on multi-finger hands, and its authors demonstrate learned control for manipulating glass marbles in a hand. This is a particular research result, not a general dexterity guarantee for every hand fitted with touch sensors. [2]

Figure's Figure 03 announcement describes internally developed tactile fingertip sensors and softer fingertips. 1X's July 2026 NEO hand update describes sensing of normal force, contact location and shear. Both are manufacturer accounts of their own systems and should be read with that attribution. [3, 4]

Terminology and counting traps

A camera in a palm and a camera inside an optical touch sensor serve different roles. Figure discusses palm vision and fingertip tactile feedback separately, while the GelSight paper explains imaging a deformable contact surface. Do not count an ordinary camera as tactile sensing solely because it is mounted near a hand. [1, 3]

Sensor count, spatial resolution and physical coverage are different measures. DIGIT's work emphasizes miniaturization for hand mounting, while GelSight studies geometric accuracy. A dense measurement at one fingertip does not establish sensing over every finger, the palm or the rest of a humanoid. [1, 2]

Slip detection is also a system behavior rather than just a sensor label. The GelSight study evaluates specific contact-motion cues; an unspecified touch sensor does not prove that the robot's controller uses those cues to recover a grasp. [1]

How to read it on humanoidversus.com

Read sensors together with hand_type and the source notes. Figure 03 and NEO are linked because their robot records cite the relevant hand or sensor descriptions. The hand hub can help find these documented configurations, but missing tactile information remains unknown. [3, 4]

This entry has no live table: hand DoF is not a touch-sensor measurement. Compare the disclosed sensing location, measured quantity and evaluation task. A null sensors field does not mean that the robot has no tactile hardware; it means the record lacks a sourced value.

Further reading

GelSight explains optical touch geometry and slip sensing. DIGIT focuses on a compact hand-mounted design. Figure and 1X show how those broader sensing questions appear in current humanoid product descriptions. [1–4]