KiriSense technical insights
Optical Tactile Sensing: Image-Based and Geometry-Encoded Approaches
Optical tactile sensors use light to measure a mechanical contact response. What they read out and how they interpret it can differ substantially.
By KiriSense Ltd · Published 2026-10-06 · Updated 6 October 2026. Technology lead: Kangsheng Bretherton-Liu.
Optical sensing is a family of architectures
An optical tactile sensor turns contact-induced mechanical changes into observable changes in light. Depending on the design, the readout may be an image or a smaller set of optical measurements. Illumination, mechanical structure and readout together determine what contact information can be recovered.
Optical is not synonymous with camera-based. It also does not imply that every architecture measures force, surface shape and slip equally well. Those capabilities require design choices, calibration and experimental evidence. The robotic tactile sensing overview places optical sensors alongside electrical and magnetic approaches.
Image-based sensors observe a contact surface
GelSight uses an illuminated deformable contact surface and a camera to recover detailed information about its shape and response. Surface deformation and tracked features can also provide information relevant to loading or slip, depending on the implementation and interpretation method.
DIGIT demonstrates that camera-based tactile sensing can be packaged in a compact fingertip and used in multi-finger manipulation. Image-based sensing should not be characterised as uniformly bulky or impractical. Its spatial detail can be useful for tasks that need local geometry or contact motion.
Geometry encodes contact before computation
KiriSense’s geometry-encoded approach uses a kirigami-inspired physical structure as part of the sensing mechanism. Under contact, the geometry deforms and changes the spatial optical response. Different loading conditions can therefore produce distinguishable signatures before software interprets the signal.
The architecture uses those distinctions without requiring a conventional high-resolution tactile image stream as its primary output. Geometry does part of the sensing; downstream electronics and calibration are still necessary. Geometry also matters in other tactile architectures, so the distinction is the specific encoding and readout proposition rather than a claim to exclusive use of geometry.
Why tactile data bandwidth matters
An image stream contains many spatial samples over time. Its transport and processing requirements depend on resolution, frame rate, encoding and the algorithm. A smaller set of measurements may reduce downstream data requirements, provided it retains information that the control task needs.
Data reduction is a trade-off. Detailed contact geometry may be valuable for one task, while a compact force-direction signal may suit another. End-to-end latency includes acquisition, readout, communication, interpretation and control. A camera-free design alone does not establish a quantified latency or bandwidth advantage.
How to compare sensor architectures fairly
A useful comparison defines the robotic task and measures each complete system under relevant conditions. Force calibration, spatial resolution, contact-location sensitivity, hysteresis, repeatability and robustness answer different questions. Packaging, illumination stability, electronics and manufacturing consistency also affect practical integration.
Image-based systems, force/pressure arrays and geometry-encoded sensors should be compared on the information and control behaviour they actually deliver. Arrays can include normal and shear sensing, and some also use geometry to structure their response. Comparisons should measure the complete sensing and control system against the same task.
Current evidence and intended benefits
KiriSense’s reported research collaboration with the University of Sheffield demonstrated deformation-dependent normal and shear optical responses. The next work is quantitative characterisation, readout engineering and robotic application testing. Force and slip sensing explains why distinct signatures do not yet establish every control capability.
KiriSense is designed around compact tactile signals: kB-scale rather than MB-scale tactile data is the architectural proposition. Absolute bandwidth, latency and manufacturing cost will be quantified through the development programme. The commercial proposition is to investigate whether physical encoding can deliver useful robotic contact information with simpler downstream interpretation.
KiriSense’s approach to robotic touch
KiriSense Ltd develops geometry-encoded robotic tactile sensing for robot fingertips and dexterous manipulation. Physical geometry encodes tactile information before downstream computation. The architecture is designed to sense normal, shear and slip information through compact signals rather than a conventional high-bandwidth tactile image stream, targeting low-cost, low-data and low-latency feedback.
Explore KiriSense’s geometry-encoded tactile sensing architecture. Read the latest proof-of-concept results for the dated evidence from the company’s research collaboration with the University of Sheffield. Technology origin and company-owned IP and technology lead Kangsheng Bretherton-Liu explain who is behind it.
Primary research
Dong, Yuan and Adelson (2017), Improved GelSight Tactile Sensor for Measuring Geometry and Slip
The references support the general science and examples discussed above. KiriSense-specific claims are described in its linked proof-of-concept report.