Robotics

Full-body tactile sensing

ActivePublished: 31 July 2026Updated: 31 July 2026Published

Key innovation

Covering a robot's entire surface with a distributed network of tactile sensors (taxels), so it can detect contact, pressure and their location on any part of the body - not only at the end-effectors (hands).

How it works

1) A flexible tactile skin (e-skin) with many taxels (usually capacitive) covers the robot's body surface in interconnected modules. 2) Taxels measure pressure/contact at many points simultaneously; the data is aggregated over a distributed bus (e.g. serial/CAN). 3) The system localizes contacts and estimates the force distribution over the surface (a tactile map). 4) The control layer uses this information to react to contact (safe stop/yield), for whole-body contact manipulation, and as an input modality for learning.

Problem solved

Robots that sense touch only in the hand are blind to contact with the rest of the body - risking unsafe collisions and preventing whole-body contact manipulation. Full-body sensing gives contact awareness everywhere, for safety and richer manipulation.

Components

Distributed tactile skin (e-skin)Touch detection

Flexible, interconnected modules with many taxels covering the body surface.

Official

Taxels (pressure sensors)Point measurement

Individual sensors (usually capacitive) measuring pressure/contact at a point.

Official

Bus and data aggregationData collection

Collecting signals from many taxels and delivering them to control (a distributed bus).

Official

Tactile map processingContact interpretation

Localizing contacts and estimating the force distribution over the body surface.

Official

Implementation

Reference implementations

iCub skin (IIT)

Istituto Italiano di Tecnologia (IIT)

Official

Implementation pitfalls

Scaling wiring and dataHigh

Covering the whole body with thousands of taxels creates wiring, bandwidth and processing challenges.

Fix:Modular interconnected boards with local aggregation, serial/CAN buses, compression/event-based readout.

Skin durability and calibrationMedium

Flexible skin wears out, and taxels require calibration (drift, temperature, hysteresis).

Fix:Replaceable modules, auto-calibration, drift/temperature compensation.

Fusing touch with vision and proprioceptionMedium

Integrating the touch modality with other sensors into coherent perception is non-trivial.

Fix:Shared representation/multimodal learning, spatio-temporal calibration.

Evolution

2011

iCub robot skin (IIT)

Inflection point

IIT develops a distributed capacitive tactile skin covering the iCub robot's body - a milestone for whole-body sensing.

2020

Scalable e-skins and learning from touch

Development of scalable, flexible e-skins and inclusion of touch as a modality in robot learning.

2025

Whole-body touch in humanoids

Modern humanoids (e.g. Generative Bionics, iCub descendants) integrate distributed touch for safe physical interaction.

Hyperparameters (configurable axes)

Body coverageCritical

Which body parts are covered by the skin (only hands/arms vs whole body).

Taxel density (resolution)High

Number of taxels per unit area - spatial resolution of touch.

Sensing modalityMedium

Capacitive, resistive, optical or vision-based sensing.

Computational complexity

Time complexity: Nie dotyczy (paradygmat sensoryczny, nie algorytm). Space complexity: Nie dotyczy.

Compute bottleneck

Real-time processing of many taxels

As the taxel count (hundreds-thousands) grows, so does the data to read, aggregate and interpret at low latency; the bottleneck is often the bus and tactile-map processing, not the sensing itself.

Depends on

Liczba taxeli i przepustowosc magistraliLatencja reakcji na kontakt