A flexible foam sensor built from silver selenide detects temperature and pressure simultaneously, enabling a robotic gripper to identify nine different materials with 96% accuracy.
(Nanowerk Spotlight) Pick up a coffee mug without looking and your hand instantly reports back: ceramic, warm, smooth, light. That snap judgment fuses thermal and mechanical information so seamlessly that it barely registers as a cognitive act. Yet for a robotic gripper on a factory line, even distinguishing an aluminum block from a rubber one of the same shape and size remains a serious challenge. The gap is not in motors or control software but in sensing.
Most robotic tactile systems stack separate sensors for temperature and pressure into layered assemblies, but these designs introduce fabrication complexity, higher costs, and a persistent technical headache known as signal crosstalk, where the output of one sensor contaminates the reading of another.
An alternative approach uses a single sensing material to detect both stimuli, but these monolithic designs have typically sacrificed sensitivity. Materials such as conductive polymers generate weak voltage signals in response to temperature differences, while high-performance inorganic thermoelectric compounds tend to be rigid and brittle, unsuitable for wrapping around a robotic fingertip. The result is an unsolved engineering tension between sensitivity, flexibility, and the ability to cleanly separate two overlapping signals.
A study published in Advanced Energy Materials (“Three-Dimensional Ag2Se Thermoelectric Network Advances Multimodal Intelligent Perception”) tackles that tradeoff by building both sensing functions into one flexible material. A research team based primarily at the Harbin Institute of Technology in Shenzhen reports a sensor constructed from a three-dimensional network of silver selenide (Ag₂Se) that detects temperature and pressure through two physically distinct mechanisms, with minimal interference between the two channels.
When paired with a machine-learning classifier, the sensor enabled a robotic gripper to identify nine different materials with 96% accuracy, an 11 percentage-point improvement over systems relying on a single sensing mode.
(a) Schematic diagram of the dual-mode sensing applications and mechanism of 3D Ag2Se network-based sensors. (b) Comparison between 3D Ag2Se network-based sensors and previously reported sensors. (Image: Reproduced with permission from Wiley-VCH Verlag) (click on image to enlarge)
The sensor fabricartion begins as a common melamine foam, the same lightweight, porous material sold commercially as a cleaning sponge. Through a two-step chemical process, the researchers first coat the foam’s internal skeleton with metallic silver using a well-established plating reaction, then convert that silver layer into Ag₂Se through exposure to a selenium solution.
The result is a highly porous, mechanically resilient block in which the thermoelectric compound forms a continuous three-dimensional network threaded through the foam scaffold.
The Ag₂Se constitutes roughly 82 wt% of the final composite. The material retains structural integrity under compressive strains as high as 80%, with negligible plastic deformation after repeated cycling, and it withstood aging tests under elevated heat and humidity as well as repeated bending and torsion.
The two sensing modes exploit different physical effects. Temperature detection relies on the Seebeck effect: when one side of the sensor contacts a hot or cold object, charge carriers migrate along the resulting temperature gradient, generating a measurable voltage. The Ag₂Se network exhibits a Seebeck coefficient of −139 µV K⁻¹, comparable to bulk Ag₂Se, while its porous architecture reduces thermal conductivity to approximately 0.033 W m⁻¹ K⁻¹, far below that of the dense material. That low thermal conductivity sharpens temperature gradients inside the sensor, boosting sensitivity.
The experimentally determined temperature sensitivity reached −122.7 µV K⁻¹ with a linearity of 0.999, and the sensor resolved temperature differences as small as 0.05 K. Its thermal response time of approximately 0.14 seconds is fast enough to register the warmth of a briefly touching hand.
Pressure detection operates through the piezoresistive effect, a change in electrical resistance caused by mechanical deformation. When the foam is compressed, its internal pore structure deforms, changing the number and area of electrical contact points within the Ag₂Se network and thereby altering the sensor’s resistance. Peak pressure sensitivity reached −2.94% kPa⁻¹ in the 3 to 18 kPa range, with a mechanical response time of roughly 0.08 seconds. The sensor detected pressures as low as 37 Pa and maintained stable resistance changes over 1,000 compression-release cycles at 15 kPa.
A critical requirement for any bimodal sensor is that its two channels do not interfere with each other, and clean separation is especially important when both signals must feed into downstream decision-making. The team demonstrated this decoupling systematically. Under varying pressures, the voltage output driven by temperature remained stable, fluctuating by only about 4%.
Conversely, the resistance signal driven by pressure held steady across different temperature conditions, varying by less than 5% within the primary operating range.
Real-time tests reinforced the point: when only a temperature stimulus was applied, resistance stayed flat, and when only pressure was applied, voltage did not drift. Voltage arises from the Seebeck effect along the thermal gradient, while resistance changes arise from mechanical restructuring of conductive pathways. Because the two phenomena operate through largely separate physical mechanisms, their signals remain cleanly separated.
That independence is what makes the sorting demonstration work. The researchers mounted sensors on the five fingertips of a robotic gripper and connected them to a multichannel data acquisition module. The gripper grasped standardized blocks of nine materials, spanning metals such as iron, aluminum, and copper, polymers such as rubber and polymethyl methacrylate, and everyday solids such as glass and wood. Each grasp produced simultaneous resistance and voltage time-series data across three sensor channels.
The team normalized and filtered these signals, then fed them into a support vector machine, a type of machine-learning algorithm that finds optimal boundaries between categories in high-dimensional data. Using 540 samples split 80/20 for training and testing, the classifier achieved 96% average recognition accuracy when it received fused three-channel input combining both pressure and temperature data. Single-channel approaches, using either resistance or voltage alone, yielded lower and more variable accuracy.
The reason is that the two channels capture fundamentally different properties. The temperature channel records transient heat-transfer dynamics at the moment of contact, effectively fingerprinting a material’s thermal character, while the pressure channel encodes mechanical characteristics such as stiffness and deformation behavior. Each channel compensates for noise or ambiguity in the other, which explains why fusion outperforms either mode alone.
The practical significance of this work extends beyond materials science into a domain researchers call embodied artificial intelligence, the branch of AI concerned with agents that physically interact with their environment rather than merely processing abstract data. Such systems depend on high-fidelity sensory input to make autonomous decisions.
A single sensor that reliably extracts two independent physical parameters from one contact event, without requiring complex multilayer fabrication, lowers a meaningful barrier to deploying intelligent robotic systems in unpredictable real-world environments such as waste sorting facilities, food processing lines, and flexible manufacturing cells.
The accuracy gain from fusing thermal and mechanical data illustrates a broader principle: multimodal sensing does not merely add information but qualitatively changes what a system can distinguish. Whether this particular foam architecture scales to commercial production remains an open question, but the underlying strategy of embedding high-performance thermoelectric materials into porous three-dimensional scaffolds points toward a concrete path for building tactile sensors that begin to rival the integrative sensitivity of human skin.
For authors and communications departmentsclick to open
Lay summary
Prefilled posts
Nanowerk Newsletter
Get our Nanotechnology Spotlight updates to your inbox!
Thank you!
You have successfully joined our subscriber list.
Become a Spotlight guest author! Join our large and growing group of guest contributors. Have you just published a scientific paper or have other exciting developments to share with the nanotechnology community? Here is how to publish on nanowerk.com.
