Human fingertips continuously gather detailed physical information. They register how strongly an object resists pressure, how its surface feels as it moves, and its temperature. For robots that manipulate cups, fruit, or tools, acquiring similar sensory information is crucial for safe operation.
Image Credit: Vu Van Quy/Shutterstock
Researchers are now developing tactile sensors that combine these signals into a single flexible layer, giving robots a working sense of touch. Early robotic grippers relied on a single force reading, which revealed little about the objects they held. The next generation of tactile sensors pairs pressure detection with vibration, proximity, temperature, and texture cues.1,2
Why Single Signals Fall Short
A sensor that measures a single quantity can mislead a robot when multiple object properties overlap. Softness, texture, material type, and temperature often produce similar changes in a basic pressure reading. A recent study in Advanced Science reported that this coupling significantly reduces recognition accuracy when a robot relies on a single tactile feature. The machine receives a number with no clue about its cause.3
The same team showed how much a second signal helps. Their system identified 26 textiles with 95.49% accuracy when it combined pressure and triboelectric data, compared with 76.62% accuracy from the triboelectric signal alone. When a volunteer stroked fabrics with random force and speed, the gap widened to 92.5% versus 52.23%. Multimodal data holds up better under realistic handling.3
Proximity sensing gives a robot advanced notice before its fingers touch anything. Capacitive sensors generate an electric field that changes when a hand or a liquid container approaches. In the Advanced Science work, twin helical electrodes detected objects more than 5 cm away and achieved their highest accuracy within the final 5 mm. This early warning allows a gripper to slow down and position itself more carefully.3
Proximity data is very important for transparent items, as glass and clear plastic can interact unpredictably with light, which confuses cameras. A capacitive sensor responds directly to the object itself, making it less affected by lighting conditions. During a drink-serving demonstration, the capacitive sensor also noticed water moving past the gripping point inside a cup. An external camera would struggle to capture that detail.3
Pressure and the Direction of Force
Pressure sensing remains the foundation of robotic touch, and newer designs resolve both the magnitude and the direction of a force. Microstructures such as tiny domes, pyramids, and interlocked bumps deform differently under normal and oblique loads. These shapes help a sensor distinguish normal force, which presses inward, from shear force, which acts sideways. This differentiation is crucial for detecting slips and ensuring the gentle handling of delicate objects.1
Incorporating arrays enhances spatial resolution. The Advanced Science system used four pressure elements built from MXene and lotus nanofibre composites, with a sensitivity of 4.34 kPa. By comparing how the load is distributed across the four elements, the sensor tracked gripping pressure, force distribution, and 3D force. When a robot twisted a plastic plate, the array tracked each shift in the direction of the force in real time.3
Vibration and the Early Signs of Slip
Slip generally starts with small vibrations before an object visibly moves. Piezoelectric films convert those tiny mechanical oscillations directly into voltage, which suits them to catching slips early. A recent work published in Frontiers in Robotics and AI placed a 28 μm polyvinylidene fluoride (PVDF) film on a silicone finger layer. As a water-filled bottle slid slowly from the gripper, the film produced a clear jump in voltage.4
Saving this for later? Download a PDF here.
The thickness of the soft layer affected how well this method worked. A 0.4 cm layer produced slip signals of 2.2-2.6 V, while a 0.8 cm layer produced 1.2-1.7 V because the thicker silicone absorbed more vibrations. Heavier bottles created stronger signals. These results show that the materials surrounding a sensing element can strongly influence how well a robot detects a slip.4
Keeping Mixed Signals Apart
Combining modalities introduces a new problem, as a soft material may respond to pressure, heat, and stretching simultaneously. Engineers reduce this crosstalk by assigning each stimulus to a different electrical output. For example, one hybrid sensor measures proximity via capacitance, pressure via light intensity, and temperature via resistance. Since each signal uses its own channel, the outputs remain clear without needing complicated algorithms to separate them.1
Timing offers another route. Triboelectric effects respond very quickly to touch, while pyroelectric signals from temperature change at a different speed. A report published in Soft Science discusses a single ferroelectric structure that measures pressure and temperature by looking at these speed differences. The report also notes that ion migration in gels can help distinguish strain from temperature within a single material.2
Cameras That Feel
Vision-based tactile sensors place a small camera behind a soft skin and read touch from the images it records. A Nature Communications paper introduced FlexiRay, a flexible gripper finger that uses a 12-megapixel camera and a set of mirrors. From those internal images, it estimates force, contact location, texture, temperature, and finger shape, covering five of the seven modalities of human touch.5
FlexiRay senses temperature via thermochromic pigments that change color at set thresholds, which the camera interprets. The finger bends about 15 mm, about four times the deformation of an earlier GelSight-based finger, and keeps 87.2 % average visual coverage. It estimated force with an error of 0.17 N and gently gripped a 2.3-gram chip. Vibration and pain remain outside its current abilities.5
Where Robotic Touch Goes Next
Raw data from many sensors must be turned into useful decisions. Recent advances use pressure and triboelectric signals, which are fed into a transformer encoder. By applying transfer learning, the model can adjust to new users with smaller datasets, achieving about 85% accuracy. Processing data close to the sensor can reduce energy use, delay, and bandwidth needs.2,3
However, several challenges remain before these sensors are widely used. Dense arrays suffer crosstalk as wiring grows tighter, and covering curved robot surfaces with uniform sensors is difficult. Humidity and bending also shift readings in many devices. Progress in structural design, wireless power, and fusion algorithms will determine how quickly service robots, prosthetics, and factory grippers can gain a reliable sense of touch.1,2
References and Further Reading
- Kong, H. et al. (2024). Recent advances in multimodal sensing integration and decoupling strategies for tactile perception. Mater. Futures, 3, 022501. DOI:10.1088/2752-5724/ad305e. https://iopscience.iop.org/article/10.1088/2752-5724/ad305e
- Tu, J. et al. (2023). Electronic skins with multimodal sensing and perception. Soft Sci, 3, 25. DOI:10.20517/ss.2023.15. https://www.oaepublish.com/articles/ss.2023.15
- Jiang, Y. et al. (2024). A Multifunctional Tactile Sensory System for Robotic Intelligent Identification and Manipulation Perception. Advanced Science, 11(41), 2402705. DOI:10.1002/advs.202402705. https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202402705
- Rosle, M. H. et al. (2025). PVDF-based flexible piezoelectric tactile sensor for slip estimation using robotic gripper. Frontiers in Robotics and AI, 12, 1691688. DOI:10.3389/frobt.2025.1691688. https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2025.1691688/full
- Wang, Y. et al. (2025). Flexible robotic hand harnesses large deformations for full-coverage human-like multimodal haptic perception. Nature Communications, 17(1), 458. DOI:10.1038/s41467-025-67148-y. https://www.nature.com/articles/s41467-025-67148-y
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.