Surface layer
Robot skin, e-skin, and tactile surfaces
Terms that explain what covers the robot and what contact signals the surface can capture.
A terminology matrix for robot skin, tactile AI, e-skin, electronic skin, multimodal sensors, ROS 2 tactile pipelines, and source-backed category language.
Definitions are written for practical reading: what the term means, which signals it touches, and where to go next for research context.
Comparing model roles? Use the VLA, VDA, world model and HFM comparison alongside the definitions below.
Updated : added HFM with its paper version and proposed-capability boundary.

Surface layer
Terms that explain what covers the robot and what contact signals the surface can capture.
Signal layer
Terms that connect pressure, shear, slip, multimodal sensing, and signal interpretation.
System layer
Terms that keep public claims conservative while guiding readers toward source-backed pages.
Core topic routes
Reading guide reviewed November 4, 2026. Existing source and research dates remain separate; no hardware measurements.
A sensor specification can use a familiar word in a narrower way than a buyer expects. Before comparing numbers, identify the quantity, unit, reference, conditions and statistic. The JCGM international metrology vocabulary separates accuracy, precision, resolution and sensitivity. The reading questions below apply those distinctions to tactile documentation; they are not a new sensor benchmark.
For tactile systems, also separate physical sensing locations from channels and reconstructed pixels. One position can report several quantities, and a learned map can contain more output cells than there are physical sensing elements. Neither multiplication establishes more independently resolved contacts.
| Term | Practical interpretation | Question for a tactile specification |
|---|---|---|
| Accuracy | Agreement with the quantity being measured; not synonymous with repeatability | Which reference and error statistic support the claim, under what conditions? |
| Precision | Agreement among repeated indications under specified conditions | Were the same sensor, mounting, load history and environment retained? |
| Resolution | The smallest change in the measured quantity that produces a distinguishable change in indication | Does the number describe force, position, digitization or merely image dimensions? |
| Sensitivity | Change in output relative to change in the input quantity | What are the units, loading range and response slope? |
Suppose a hypothetical sensor repeatedly reports a similar force while consistently differing from a reference. Closely grouped readings alone do not establish accuracy. Likewise, a finely digitized output can still have noise, bias or cross-axis effects that prevent useful small changes from being distinguished. These examples illustrate terminology, not measured performance of a named device.
Keep measured, estimated and predicted labels on the signal. A tactile image is an observation; a calibrated force reconstruction is an estimate; a future-contact model predicts an outcome. The downstream controller may use any of these, but their evidential roles remain different. Follow the linked definition to the relevant method and evaluation rather than treating a shared word as a compatibility guarantee.
Term route matrix
Term 01
A tactile sensing surface that helps a robot detect contact, pressure, shear, slip, or interaction events across hands, grippers, arms, or curved body surfaces.
Term 02
Software and sensing workflows that turn touch data into useful robot signals for grasping, contact response, manipulation, or evaluation analytics.
Term 03
A broad term for AI systems that perceive, reason, and act through physical machines using sensors, models, policies, control, actuation, safety, and measured feedback.
Term 04
The use of data or experience to train robot perception, prediction, or action through demonstrations, rewards, self-supervision, simulation, or multimodal feedback.
Term 05
A physical robot whose body plan or capabilities are designed around human-scale environments, often combining mobility, arms, hands or grippers, perception, control, and safety systems.
Term 06
A robot model that uses visual observations and language instructions to produce or condition physical actions; implementations differ in action representation, embodiment, data, and control interface.
Term 07
A broadly trained model intended for reuse or adaptation across multiple robot tasks, environments, or embodiments, with the claimed breadth requiring direct transfer evidence.
Term 08
The use of a robot arm, hand, gripper, tool, or whole body to intentionally change an object or environment through physical action.
Term 09
Electronic skin: a flexible or soft sensor layer designed to measure contact-related signals on non-flat surfaces.
Term 10
Detection of object movement relative to a robot finger or gripper, often using shear, vibration, texture, or event-based tactile signals.
Term 11
A sensor approach that captures more than one stimulus type, such as pressure and temperature, while managing crosstalk and calibration.
Term 12
A robotics software path for recording, replaying, transforming, and consuming tactile sensor data with consistent timestamps, frames, and middleware settings.
Term 13
A conservative explanation of how robot skin terminology, applications, and research context relate without implying product availability.
Term 14
The measurement of physical contact through signals such as pressure, normal and shear force, slip, vibration, temperature, deformation, or tactile images.
Term 15
A device that converts contact at a robot surface into electrical, optical, magnetic, acoustic, or other measurable observations.
Term 16
A spatially indexed tactile sensing element in an array. Unlike a pixel, its physical quantity, response area, crosstalk, and calibration depend on the sensor design.
Term 17
Robot manipulation that uses measured contact to estimate physical state and update grip, pose, trajectory, force, or recovery actions.
Term 18
A manipulation setting in which sustained or repeated physical contact is central to task success, as in insertion, surface following, folding, or in-hand reorientation.
Term 19
Joint processing of visual and tactile observations to estimate material, contact, pose, stability, or other physical state.
Term 20
Robot manipulation in which both visual context and measured touch influence actions such as approach, grasp stabilization, insertion correction, or recovery.
Term 21
Learning compact features from tactile observations for downstream perception, prediction, retrieval, or robot-control tasks.
Term 22
A broadly pretrained tactile or multimodal model intended to transfer across multiple downstream tasks, sensors, objects, or embodiments, with the claimed breadth requiring direct evidence.
Term 23
Haptic foundation model: a proposed reusable model for heterogeneous touch and contact signals. The 2026 HFM framework includes action-coupled sensing and prediction as research objectives; its frozen-encoder benchmark does not establish the full proposed framework. The term overlaps with tactile foundation model rather than defining a standardized separate class.
Source: Wang et al., The Potential of Haptic Foundation Models (v1)
Term 24
An action-conditioned model that predicts future tactile state, contact, force-related signals, slip, images, or learned touch representations.
Term 25
A versioned collection of tactile observations and metadata, potentially aligned with vision, language, force, robot state, actions, objects, or tasks.
Term 26
A reproducible evaluation contract that defines tactile inputs, tasks, data splits, metrics, baselines, hardware conditions, and version.
Term 27
Distributed contact sensing across large robot surfaces such as hands, arms, torso, or other body regions rather than a single fingertip or gripper pad.
Term 28
A tactile sensor with a compliant or flexible interface that can conform to contact geometry while measuring deformation or related physical signals.
Term 29
Force acting tangentially along a contact surface; in robotic grasping it can provide evidence about load, sliding, and incipient slip.
Term 30
The ordered tactile observations from one physical press, grasp, or continuous contact event. Keeping sequences intact helps prevent near-duplicate train-test leakage.
Terminology correction path
Share a research question, source suggestion, or guide route case so the inquiry can be routed clearly.