A new flexible sensor could help machines see, feel and make decisions with a level of coordination that has long challenged conventional electronic skins. In a study published in Nature Sensors, researchers report a vertically stacked bimodal device that combines photoelectric and pressure sensing in the same compact location while keeping the two electrical signals largely separate. The architecture is designed to imitate a basic feature of biological perception: multiple types of information are collected from one physical point, then interpreted together rather than being measured by distant or independently operated components. The researchers say their sensor can recognize objects, guide robots through simulated fire environments without preloaded maps and monitor both soil moisture and light intensity, suggesting a route toward more perceptive autonomous systems.
Flexible sensors are increasingly being developed for robotics, wearable electronics, environmental monitoring and human–machine interfaces. Yet many multimodal systems still rely on physically separated sensing elements or on different devices that acquire signals independently. That arrangement creates a fundamental problem. When light, pressure, temperature or other stimuli arrive at the same time, the system must determine which response belongs to which stimulus. Signals can interfere with one another, while differences in location and timing make data fusion more difficult. In a robotic skin, for example, a light detector may be mounted beside a pressure sensor rather than directly beneath it, forcing software to reconcile measurements that do not originate from exactly the same point. The new device addresses this challenge by placing both functions into a vertically integrated structure and engineering their signal pathways to remain intrinsically decoupled.
At the heart of the photoelectric channel is a SnSeₓSᵧ/PTAA heterojunction. SnSeₓSᵧ is a tin selenide–sulfide semiconductor whose composition can be represented by the variables x and y, allowing its electronic and optical properties to be tuned through the relative amounts of selenium and sulfur. PTAA, or poly[bis(4-phenyl)(2,4,6-trimethylphenyl)amine], is an organic semiconductor commonly used as a hole-transport material. When the two materials are brought together, the interface forms a heterojunction, meaning that their different energy levels can assist the separation and movement of photo-generated charge carriers. Light absorbed by the active semiconductor produces electrical changes that can be read as a photodetection signal. According to the researchers, this pairing provides broadband light response, useful responsivity and low detection limits in a mechanically flexible format.
The second sensing channel uses a covalently interlocked network made from polypropylene and functionalized carbon nanotubes. Polypropylene provides a lightweight, flexible polymer framework, while the carbon nanotubes create electrically conductive pathways throughout the material. Functionalizing the nanotubes improves their interaction with the surrounding polymer, and covalent interlocking helps stabilize the resulting network. When pressure is applied, the structure deforms, changing the distances and conductive connections between neighboring nanotubes. That change is translated into an electrical response. Because the network is designed to deform in a controlled way, it can produce a highly linear pressure signal, allowing the strength of an applied force to be estimated more reliably than in systems with strongly nonlinear or unstable responses.
The vertical arrangement is central to the device’s operation. Instead of placing photoelectric and pressure elements side by side, the researchers stack them so that the sensing functions occupy the same footprint but respond through different physical mechanisms and electrical routes. The photoelectric component reacts primarily to incident light and the behavior of charge carriers at the semiconductor heterojunction. The pressure component responds primarily to mechanical deformation within the polymer–nanotube network. Separating these mechanisms at the material and architecture levels reduces the risk that a force-induced structural change will be mistaken for a light signal, or that illumination will significantly alter the pressure readout. The reported cross-channel interference is below 1%, a level that the researchers identify as evidence of strong decoupling between the two modes.
This distinction matters because simultaneous sensing is more valuable than simply collecting two measurements. In real environments, visual or optical information can be ambiguous without contact information, while pressure alone may reveal that an object is present without identifying its shape, surface or position. A fused signal can combine complementary clues. The researchers use deep learning to interpret these synchronized data streams, allowing the sensor to move beyond raw electrical outputs and toward task-level recognition. Rather than treating the photoelectric and pressure channels as isolated instruments, the computational model learns relationships between them. That approach can help distinguish objects or events that might produce similar responses in only one modality.
In demonstrations of intelligent recognition, the device supplied multimodal information for identifying objects. Although the sensor’s two channels operate independently at the hardware level, their outputs can be analyzed jointly by a learning system. Optical response can contribute information about illumination or an object’s interaction with light, while pressure response can describe contact, force and mechanical texture. Combining these signals gives the classifier a richer representation than either channel could provide alone. The result is a sensor platform intended not merely to detect stimuli, but to support recognition under conditions in which one source of information may be incomplete or noisy. Such an approach is especially relevant to flexible robotic skins, where contact and environmental light often change together.
The researchers also tested the system in a simulated fire-navigation scenario, where a robot used fused sensory information to navigate without relying on a pre-existing map. Mapless navigation is demanding because the robot must infer its surroundings while moving, rather than following a stored representation of the environment. In a fire-related setting, optical conditions and physical interactions can both change rapidly, making a single sensing modality vulnerable to confusion. A photoelectric signal can provide environmental information related to light, while pressure feedback can indicate contact or interaction with nearby surfaces. Feeding both into a deep-learning framework gives the robot additional context for making movement decisions. The demonstration points toward autonomous machines that can respond to unfamiliar environments rather than simply execute predetermined routes.
Beyond robotics, the sensor was used to track soil moisture and light intensity for environmental monitoring. These measurements are important in agriculture because plant growth depends on the availability of water and the amount of incoming light, yet the two variables can fluctuate independently. A flexible, co-located sensor could potentially be placed on or near agricultural surfaces, where it would monitor optical conditions while also registering mechanical changes associated with moisture. The reported demonstration does not by itself establish a complete field-ready farming system, but it illustrates how a single platform might gather multiple environmental signals in a coordinated way. In precision agriculture, that combination could eventually support more targeted irrigation, crop monitoring and resource management when integrated with suitable wireless electronics and control systems.
The researchers describe the device as a material and architectural paradigm for synergistic bimodal sensing. Its significance lies not only in the specific semiconductor, polymer and carbon-nanotube components, but also in the strategy of assigning each modality a distinct physical pathway before using computation to combine their outputs. That hardware–software division can simplify data fusion and reduce the crosstalk that has limited many flexible multimodal sensors. Challenges remain before such systems become widespread, including long-term durability, manufacturing at large areas, calibration across devices, energy consumption and reliable operation under changing temperature and humidity. Even so, the combination of vertically stacked sensing, intrinsic signal separation and deep-learning interpretation offers a compelling blueprint for future electronic skins. By allowing machines to detect light and pressure from the same point, the technology moves flexible sensors closer to the integrated perception needed for embodied intelligence.
Subject of Research: A flexible bimodal sensor integrating co-located photoelectric and pressure sensing for intelligent recognition, robotic navigation and environmental monitoring.
Article Title: A bimodal sensor with deep learning-enhanced synergistic sensing for intelligent recognition and navigation
Article References: Zhao, B., Gao, D., Hu, X. et al. “A bimodal sensor with deep learning-enhanced synergistic sensing for intelligent recognition and navigation.” Nature Sensors (2026). https://doi.org/10.1038/s44460-026-00127-y
Image Credits: AI Generated
DOI: https://doi.org/10.1038/s44460-026-00127-y
Keywords: Flexible sensors, bimodal sensing, photoelectric sensing, pressure sensing, SnSeₓSᵧ/PTAA heterojunction, carbon nanotubes, deep learning, robotic navigation, environmental monitoring, precision agriculture
Tags: autonomous robotic navigationbio-inspired perception systemsenvironmental monitoring sensorsflexible bimodal sensorsintegrated photoelectric and pressure sensingintelligent object recognitionmulti-functional electronic skinmulti-signal data interpretationmultimodal sensor technologysensor architecture for perceptionsensor signal separation techniqueswearable electronics sensing
