With the rapid development of personalized healthcare¹⁻³, virtual reality (VR) / augmented reality (AR)⁴⁻⁶, and humanoid robots⁷⁻⁹, optical tactile sensors have attracted intensive attentions due to their high sensitivity, high precision, fast response, and anti-electromagnetic interference¹⁰⁻¹⁴. Typically, an optical tactile sensor consists of a light source, a packaged sensing element, and a detector. By monitoring the change of output intensity, shift of resonant peak or interference spectra, a great number of high-performance tactile sensors have been demonstrated using silica optical fibers¹⁵⁻¹⁸, polymer optical waveguides/fibers¹⁹⁻²², hydrogel optical fibers²³⁻²⁵, and optical micro/nanofibers (MNFs)²⁶⁻²⁸. Among them, MNFs possess excellent optical and mechanical properties, including strong evanescent field, low optical loss, wavelength scale diameter, small bending radius, various structures, and seamless connection with standard optical fiber²⁹⁻³². The use of MNFs in tactile sensors can offer many exciting capabilities unavailable from other optical tactile sensors. For example, ultrahigh pressure sensitivity was achieved by using an optical nanofiber thanks to its prominent evanescent field and low stiffness²⁸. MNFs can be mass-produced with controllable diameter, high transmission, high reproducibility, and very low cost by the fiber drawing technology³³⁻³⁵. Various flexible and stretchable MNF structures have been designed for measuring pressure, temperature, strain, bending, and so on. Furthermore, MNF tactile sensors show great potential for simultaneous detection of multi-parameters, integration with various substrates, and spatial multiplexing. Recent progresses in MNF tactile sensors have been well reflected by the numerous high-quality publications, especially in the applications of healthcare, human machine interface, and robotics.

On the other hand, MNFs have been used to assemble miniaturized photoactuators³⁶, which can produce a reversible mechanical deformation under light stimuli. In the recent years, photoactuators have attracted tremendous interest owing to their potential for soft robot and gripper application³⁷⁻³⁹. Most of photoactuators rely on free-space illumination, which requires a line-of-site low-loss optical path. While waveguide photoactuators can overcome this limitation⁴⁰⁻⁴², their actuating performances are fundamentally restricted by the nature of the standard optical fibers. For example, the relatively large size of the plastic optical fiber (e.g., 500 μm in diameter) results in a thick active layer, which may mitigate its bending angle⁴⁰. Optical fiber taper, a kind of MNF with a tip diameter less than 1 micron, makes it possible to reduce the thickness of active layer and overcome the size mismatch between optical fiber and photo-responsive material. Moreover, the fiber taper provides a higher energy density than that of standard optical fiber due to the strong confinement of the MNF. Although the optical fiber enabled photoactuator is still in its infancy, the optical fiber taper offers an effective strategy to design high performance miniature photoactuators.

Several excellent review articles have been published on optical tactile sensors⁴³⁻⁴⁶ and photoactuators^47,48, however, the role of MNFs has been underestimated. In this review, we focus on the recent advances in MNF enabled tactile sensors and actuators, and highlight the uniqueness of MNF for ultrasensitive and multifunctional sensors and miniaturized high-performance actuators. Figure 1 summarizes the on-going research in the four application areas. Our purpose is to introduce the key concepts and key papers for each topic. Firstly, we will discuss the fabrication and packaging of the MNF device. Secondly, we will summarize the sensing principle of the MNF enabled tactile sensors and their typical applications. Thirdly, we will discuss the MNF enabled photoactuators. Finally, we will give a brief summary and future prospect of MNF devices.

Since the demonstration of low-loss optical waveguiding in subwavelength-diameter nanofiber⁴⁹, flame-heated, laser-heated, and electrically heated taper drawing techniques have been widely used to fabricate MNFs⁵⁰. In 2008, we fabricated a 680-nm-diameter nanofiber using an alcohol burner (Fig. 2(a-i) for fast response humidity sensing⁵¹. Although this humidity sensor achieved fast response time and high sensitivity, the alcohol burner based taper drawing technique faced great challenge in large-scale manufacturing of uniform and low optical loss MNFs. In order to address this issue, the flame-heated taper drawing setup evolved from hand-held drawing, stage assisted drawing (Fig. 2(a-ii) to automatic drawing (Fig. 2(a-iii). Fang and Tong developed a series of methods³³⁻³⁵ by real time monitoring optical transmission to precisely control the diameter of the MNF. Basically, there are several clearly observed abrupt drops of transmission intensity, and each drop indicates a cutoff of an optical mode, corresponding to a specific MNF diameter (Fig. 2(b-i). Thus, by accurately measuring the time interval between two drops, one can precisely determine the time to stop taper drawing based on a target MNF diameter. Experimentally, the diameter accuracies can be 2 nm and 5 nm in the MNF diameter range of 360–680 nm³⁴, and 800–1300 nm³³, respectively. For a 10-cm-long, 1.05-µm-diameter MNF, the standard deviation over the whole MNF is 19 nm, yielding an excellent diameter uniformity as ∆d/L < 1.6×10⁻⁷, where ∆d is standard deviation over the whole MNF, and L is the length of the MNF³⁵. Due to the smooth surface and uniform diameter (Fig. 2(b-ii), the transmittance can be as high as 99.4% for a 10-cm-long MNF with a diameter of 1.2 μm. The low optical loss and high flexibility of the MNF guarantee the MNF devices with a high transmittance (Fig. 2(b-iii), which is very important for flexible sensors and photoactuators. Recently, Fang et al. proposed a parallel-fabrication approach (Fig. 2(c-i) to simultaneously drawing multiple (up to 20) MNFs with almost identical geometries⁵². Alternatively, a home-made electric heater was used to generate a large hot zone with uniform temperature distribution. During the taper drawing, the multimode-induced oscillation for each fiber starts almost simultaneously and is suppressed at approximately the same time (Fig. 2(c-ii), indicating a nearly identical taper drawing environment. As a result, the as-fabricated MNF array shows high uniformity in terms of diameter and transmittance (Fig. 2(c-iii). This parallel fabrication approach may open a door towards large scale fabrication of MNF devices.

In addition to large scale fabrication, a reliable package plays critical role in MNF devices, particularly in flexible MNF tactile sensors and actuators. Inspired by the PDMS microfluidic chip embedded MNF (Fig. 2(d-i)⁵³, almost all the reported MNF tactile sensors and actuators were packaged by PDMS, owing to its high transparency, high flexibility, low Young's modulus, low refractive index (~1.40), and excellent biocompatibility. Typically, a thin layer of PDMS film (e.g., 80 μm in thickness) can effectively isolate the evanescent fields, while maintaining high mechanical flexibility and low optical loss of the MNF. With the package of PDMS, an MNF tactile sensor can be bent in free space (Fig. 2(d-ii)²⁸, attached on human skin (Fig. 2(d-iii)²⁸ or curved surface (Fig. 2(d-iv)⁵⁴. With reliable package, a sensitive MNF device shows high stability, robustness, and compact size, making it attractive for various applications with unique advantages.

Since the basic model for microfiber sensors was proposed in 2005⁵⁵, a great number of physical, chemical or biological MNF sensors have been designed and fabricated based on evanescent-wave guiding properties of MNFs⁵⁶⁻⁵⁸. By measuring optical phase shift or changes in output intensity of the guided light, these MNF sensors offer excellent properties including high sensitivity, fast response, low power, and small footprint. MNF sensors based on intensity demodulation usually employ photodetectors to detect the light intensity changes caused by scattering, absorption, or re-emission of the guided light. Intensity demodulation has the advantages of simple system and good maneuverability, which has been widely used in chemical materials detection and biosensing. On the other hand, the intensity demodulation method directly detects the light intensity, the relative light intensity noise in the light component is relatively large, which may restrict its detection limit and resolution. Alternatively, the MNF sensors that adopt wavelength demodulation can be divided into resonators (e.g., whispering gallery mode (WGM) resonator⁵⁹⁻⁶¹, Fabry−Perot cavity^62,63, gratings (e.g., fiber Bragg grating⁶⁴⁻⁶⁶, long period grating⁶⁷⁻⁶⁹), and interferometers (e.g., Mach-Zehnder interferometer)^70,71. They measure external stimuli by detecting the wavelength shift, so the demodulation results are less affected by the light intensity noise. These wavelength demodulation MNF sensors have demonstrated high sensitivity, high resolution, and large dynamic range, however, the sensing system is more complex and expensive than that of intensity demodulation MNF sensors. Along with the rapid progress in wearable sensors⁷², and increasing demands on flexible optical sensors, extensive research has been devoted to MNF-based optical flexible and wearable sensors, in which MNFs were employed as an artificial never sensing pressure, strain, bending angle, temperature, and relative humidity. Here we summarized the typical sensing mechanisms and the advantages and/or potentials of these miniaturized optical sensors categorized by MNF structures including biconical MNF, MNF ring resonator, MNF probe, MNF coupler, and microfiber Bragg grating (µFBG).

One of the most straightforward approaches to an MNF tactile sensor is using a biconical MNF, which has been successfully applied in physical^73,74 and chemical sensors^75,76. Typically, a biconical MNF was embedded in a thin layer of flexible and stretchable low refractive index PDMS film²⁸. The host of PDMS not only isolate the evanescent field of the MNF, but also transfer the exerted stimuli to the MNF with high fidelity (Fig. 3(a-i). As shown in the inset of Fig. 3 (a-i), the well confined symmetric mode of a straight MNF will evolve into an asymmetric profile with clear optical leakage when it is slight bent, making it a highly sensitive to micro deformation. Within the PDMS host film, the MNF can be straight-, U-, or wavy-shape (Fig. 3(a-ii), offering additional flexibility for pressure, strain, or bending sensing. The engineerable MNF diameter and the tunable wavelength of the probing light offer an opportunity to achieve ultrahigh sensitivity or a wide detection range. For example, a suspended MNF tactile sensor (50-μm-thickness, 780-nm-diameter MNF) can feel the extreme low pressure down to 0.1 Pa with a high signal-to-noise ratio (Fig. 3(a-iii) corresponds to a detection limit of about 7 mPa. Moreover, by taking advantage of PDMS's large negative thermo-optic coefficient, temperature measurements, which plays a key role in tactile sensing, has been demonstrated for healthcare or robotics applications. In brief, the biconical MNF enabled tactile sensors pave the way toward a new category of optical tactile sensors due to its high sensing performance, controllable sensing structure (e.g., MNF diameter, bending radius, PDMS thickness), potential for large scale manufacturing, and simplicity in signal processing.

Based on evanescent coupling at the overlapping area, an MNF can be assembled into a closed loop, knot, or coil, forming a high-quality whispering gallery mode (WGM) resonator^77,78. Although a free-standing circular MNF resonator is easy to fabricate, it is mechanically fragile, and the MNF is susceptible to environmental disturbance. Enclosing a free-standing MNF resonator inside a low refractive index polymer is an effect approach to improve the robustness of an MNF resonator^79,80. Benefitted from the high Q-factor and the miniaturized structure, the MNF resonators have been intensively investigated and applied for weak signal detection^81,82. In 2018, Xu et al. proposed a hybrid plasmonic microfiber knot resonator embedded in a PDMS membrane (Fig. 3(b-i, ii) for healthcare monitoring²⁷. By measuring the shift of the resonance peak (Fig. 3(b-iii), the flexible and wearable MNF sensor achieved an ultrahigh gauge factor of 13700 and a pressure sensitivity of 0.83 kPa⁻¹. The WGM resonator provides a sensitive approach for designing wearable sensors for vital physiological monitoring applications. In addition to WGM resonator, MNF theta-shaped resonator⁸³, MNF Saganc resonator⁸⁴, multimode MNF inline interferometer⁸⁵ and MNF Mach-Zehnder interferometer⁷⁰ have been widely used for refractive index, force and ultrasonic sensing, providing more options for developing high performance flexible and wearable MNF sensing systems. However, its application in robotics or human-machine interface might be limited, because the MNF resonator sensing system often requires wavelength-swept laser and a high-resolution optical spectrum analyzer.

An MNF probe is a half biconical MNF, which is fabricated by nonadiabatic fiber tapering, followed by fiber cleaving. In order to increase the reflectivity of the MNF tip with a diameter of a few microns, a thin layer of gold can be coated on the enface of the MNF tip⁸⁶. To work as a flexible tactile sensor, a low refractive index polymer package is necessary. Similar to the biconical MNF, the bending deformation of the MNF probe will lead to distortion of waveguide structure, resulting in modification of index profile in MNF probe⁸⁷. When the effective refractive index of fundamental mode is lower than the cladding index, the fundamental mode will partly leak into the cladding and gradually transform into radiation modes, leading to the loss of light. For an MNF probe with a diameter of 2.1 µm, bending or pressure caused deformation can be obtained with a high sensitivity and flexibility by measuring the change of the reflected light. As a finger motion sensor, it achieved a bending angle sensitivity of −3.42%/(°) in the range of 5°–30° and −0.74%/(°) in the range of 30°–60°, respectively.

MNF coupler provides an ultrasensitive sensing approach for weak strain or pressure detection, because the coupling efficiency is strongly dependent on the gap between the two MNFs and the coupling length^84,88,89. Thus, any displacement between the two MNFs will be reflected upon the change of coupling efficiency. Note that, the change of ambient temperature may lead to a change of the coupling efficiency, because the MNF cladding usually possesses a giant negative thermal refraction coefficient (e.g., PDMS, −1.0 × 10⁻⁴ /°C)⁹⁰. Fortunately, the impact of temperature on the weak strain sensing can be mitigated⁹¹. On the other hand, the temperature dependent coupling efficiency offers a promising approach to realize subtle change of ambient temperature⁹². Although the working range of this kind of sensors is limited, the ultrahigh sensitivity, fast response, and small footprint make them attractive for vital sign and micro-displacement monitoring. For example, Yu et al. proposed a flexible sensor with two evanescently coupled optical MNFs embedded in a PDMS film (Fig. 3(c-i)⁹³. As shown in Fig. 3 (c-ii), the gap between the two MNFs is only about 50 nm. The coupler sensor achieved a gauge factor of 64.5 and a strain resolution of 0.0012% which corresponds to elongation of 120 nm on a 1 cm long device (Fig. 3(c-iii). Moreover, the fast temporal frequency response (up to 30 kHz) and high sensitivity (102 kPa⁻¹) bestowed the coupler sensor a capability for sound detection.

In addition to the abovementioned sensing mechanisms, the fiber Bragg grating represents the most successful sensing element with subwavelength or nano-engineered periodical structures in a single optical fiber^66,94,95. The Bragg wavelength (λ_B) of the fiber Bragg grating is deduced from: λ_B= 2n_effΛ, where Λ is the grating period and n_eff means the effective refractive index of the core mode. λ_B can respond to external parameters mediated by the variation of n_eff and Λ. When a FBG is bent, the interior of the structural unit is shortened by compression and the exterior of the structural unit is stretched by tensile forces. The length of the neutral line always remains constant between compression and extension. Therefore, the curvature of the neutral line can be used to represent the shape change of the fiber units. In general, there is a linear relationship between the bending curvature k and the Bragg wavelength drift Δλ_B. Thus, by simply measuring the Bragg wavelength drift of the FBG, one can obtain the curving degree. For example, Xu et. al developed a microfiber Bragg grating enabled wearable sensor for healthcare monitoring⁹⁶. Due to the strain amplification effect of the microfiber, the sensor achieved enhanced sensitivity in pressure, bending angle and temperature.

The pulse is a key biomedical signal containing various human physiological and pathological information highly related to cardiovascular diseases⁹⁷. Typically, a flexible pressure sensor is attached onto the wrist (Fig. 4(a-i) to detect the subtle vibration caused by the blood flow in radial artery. A typical pulse waves contains three peaks, named as percussion wave (P-wave), tidal wave (T-wave), and diastolic wave (D-wave) (Fig. 4(a-ii)²⁷. The pulse waveform (e.g., the amplitude, time intervals between the peak are related to the systolic and diastolic blood pressure, the ventricular pressure, and the heart rate. For example, continuous and cuff-less blood pressure (BP) monitoring has been realized by combining the electrocardiogram and pulse transit time (PTT)⁹⁸. However, a wrist pulse sensor often exhibits unstable signals under hand motion because of the concave surface of the wrist⁹⁹. By contrast, fingertips have a convex surface and therefore show great promises in stable and long-term pulse monitoring (Fig. 4(a-iii, iv)²⁷. Nevertheless, the fingertip pulse signal is weaker than the wrist pulse, calling for highly sensitive sensors, such as MNF resonator sensor or MNF coupler sensor.

In order to develop a stable wrist pulse sensor, Li et al employed a flexible soft liquid sac to eliminate the misalignment and sensor position drift because the pressure change at any point in the incompressible static fluid caused by the pulse can be transmitted to upper PDMS film embedded with an MNF¹⁰⁰. The high-fidelity pulse signal in the area of 20 × 20 mm² is successfully detected with uniform waveform.

On the other hand, the pulse waveforms are affected by body temperature, which may change the blood vessels vasodilation and vasoconstriction¹⁰¹. Therefore, it is necessary to measure temperature together with pulse monitoring to realize an accurate analysis of pulse waveforms. Yao et al. developed an MNF sensor that can detect both temperature and pressure by measuring the shift of a high-order mode cutoff wavelength in the short-wavelength range and the change of transmittance in the long-wavelength, respectively (Fig. 4 (b-i)¹⁰². As shown in Fig. 4 (b-ii), when the skin temperature increases 1.1 °C, the measured transmission peak presents a 1.3-nm wavelength shift. As a result, the pulse waveforms are quite different at cold (22.5 °C) and warm (40.0 °C) conditions in terms of the round-trip time (Fig. 4 (b-iii).

Despite of the success in real time wrist pulse monitoring¹⁰³, it is difficult to achieve systemic hemodynamic monitoring due to the lack of multichannel and time-synchronized the whole-body sensing system. Recently, Zhu et al. developed a continuous systemic hemodynamic measurement technique to monitor the whole-body blood circulation by using three flexible microfiber Bragg grating patches attached on neck, wrist, and foot, respectively (Fig. 4(c-i)¹⁰⁴. By detecting the pulse wave at different superficial artery sites, the PTTs of three different cases (heart to carotid artery (CA), heart to radial artery (RA), and heart to pedal artery (PA)) can be detected and calculated. The waveforms and details of the synchronous ballistocardiogram (BCG) signal and pulse wave shows artery lengths and pulse wave propagation velocity dependent characteristics (Fig. 4(c-ii). Relying on the multichannel and time-synchronized operation capability, the wearable MNF sensing system is capable of detecting the all-mechanical process of pulse wave propagation, making it a promising device for clinical medical diagnosis and daily health management.

In addition to pulse, respiration^91,105, body temperature^54,96,102, and body motion^54,87,106 sensing, the highly sensitive and miniaturized MNF have been used for ultrasound sensing and photoacoustic imaging. Combined with the coherent detection technology, Yang et al. proposed an ultrasound sensor by embedding a straight MNF into a thin layer of PDMS¹⁰⁷. Ma et al. proposed an ultrasound sensor by sandwiching a microfiber loop between a pair of in-line Bragg gratings, forming a Fabry-Perot cavity that allows free delivery of ultrasound/ light beams and unique needle shaped ultrasound focusing along the penetration depth¹⁰⁸. Benefiting from the large evanescent field characteristic of the microfiber, the proposed sensors realized highly sensitive ultrasound detection and demonstrated excellent performance in high-resolution photoacoustic imaging, which is a promising tool for noninvasive biomedical imaging and diseases diagnosis¹⁰⁹.

We are already entering the era of the metaverse, in which virtual life and reality interact via various human machine interfaces^110,111, such as data glove^112,113, smart ring¹¹⁴, wristband^115,116, socks¹¹⁷, arm sleeve¹¹⁸, clothes¹¹⁹, and so on. Among them, data gloves have been intensively investigated due to their great potential for personalized healthcare, remote operation, motion capture, sign language recognition and VR space bidirectional communication. Compare to their electronic counterparts, MNFs enable data glove (Fig. 5(a-i) offers unique properties including high bending angle resolution (Fig. 5(a-ii), fast response, low power consumption, and immunity to electromagnetic interference²⁸. The MNF enabled data gloves have been successfully used for controlling a virtual hand (Fig. 5(a-iii) or a robotic hand (Fig. 5(a-iv).

Although the data gloves can be light weight and desirable comfortability, there is an increasing requirement for imperceptible, convenient to use, and inexpensive human machine interface to facilitate natural and continuous interactions in the real physical world and online virtual ones. Inspired by the gesture-recognition wristband equipped electronic pressure sensors, Wang et al. reported an MNF pressure sensor enable wristband (Fig. 5(b-i)¹²⁰. The key to the success of the optical wristband lies in three aspects: (1) the liquid sac can effectively mitigate the impact of sensor position on the detection of the movement of finger related tendons without sacrificing sensitivity (Fig. 5(b-ii); (2) highly sensitive MNF pressure sensor on the skin can detect the subtle movement of tendon under the skin (Fig. 5(b-iii); (3) The support-vector machine (SVM) machine-learning model can effectively decode the signals from the MNF sensors. For a three-sensor wristband (Fig. 5(b-iv), the maximum hand-gesture-recognition accuracy is about 94%, which is higher than that of electronic wristband using more sensors. To demonstrate its potential for human machine interaction, an MNF enabled wristband was used to remotely control a robotic hand (Fig. 5(b-v).

In addition to the hand gesture recognition devices, significant efforts have been made to develop conductive textile to achieve lightweight, conformable, intuitive, and seamless interaction between human and machines¹²¹. Basically, the current strategies for manufacturing smart textiles are weaving functional fibers into textiles or depositing electric sensing materials on the fabric surface. However, sewability and washability represent two major challenges, which may hinder their practical applications. Ma et al. proposed an MNF enabled smart textile by integrating a PDMS patch embedded with MNF array with a piece of textiles (Fig. 5 (c-i)¹²². The optical driven smart textile (Fig. 5(c-ii) can be attached to cloth for working or detached from cloth during washing, offering an alternative manner to weaved-fibers based electronic HMIs. Benefiting from the warp and weft structure of the textile, the MNF enabled smart textile can feel slight finger touch or slip along the MNF, and classify the touch manners with the help of a SVM model. As proof-of-concept demonstrations, logic control of a robotic hand (Fig. 5(c-iii) and communication with a virtual “xiaozhi” were realized (Fig. 5(c-iv).

In addition to abovementioned pressure or bending sensor-based HMIs, humidity sensor provides a unique approach to realized proximity sensing^123,124, which is essential for human−machine interactions. Inspired by the somatosensory system of human skin, Liu et al. developed a hierarchically designed dual-mode interactive sensor (Fig. 5(d-i)¹²⁵ which consists of two designated MNFs, a Nafion−crystal violet (CV) layer, a stiff PDMS ring, and three PDMS packaging layers (Fig. 5(d-ii). Thanks to the highly sensitive humidity and pressure sensing ability, together with a smart data processing unit, the integrated device achieved continuous detection of the full-contact events, including finger approaching, contacting, pressing, releasing, and leaving (Fig. 5(d-iii). As a proof of concept, an HMI switch system shows collaborative on/off response to humidity and/or pressure change caused by the bare/gloved fingers (Fig. 5(d-iv).

It is essential for a humanoid robot to equip a skin-like tactile sensor with multifunctional sensing ability to realize dexterous manipulation and tactile feedback^126,127. The epidermis with high elastic modulus and the fingerprint patten plays critical roles in amplifying and transferring external stimuli to sensory receptors. By mimicking the structural characteristics of human finger skin, Jiang et al. designed and fabricated an MNF tactile sensor with fingerprint-like parallel ridges (Fig. 6(a-i)¹²⁸. Benefiting from fingerprint structures, a robotic gripper installed with the MNF tactile sensor can detect and discriminate grasping force as well as contact slippage when performing object manipulation tasks (Fig. 6(a-ii). Moreover, slipping related information can be derived by applying wavelet transform to the output signal (Fig. 6(a-iii).

In addition to slip detection, a multitude of tactile sensors with a capability of measuring force, deformation, temperature, and hence allowing for the discrimination of hardness, texture, and thermal conductivity is now playing a vital role in humanoid robots. To this end, Tang et al. designed and fabricated a multimodal (force and thermosensitive) robotic skin by embedding MNF into PDMS films with special modules as shown in Fig. 6¹²⁹. The MNF tactile sensor can enable a commercial robot to emulate human behaviors, including adaptive grasping of a cup of coffee (Fig. 6(b-ii) and recognition of object texture (Fig. 6(b-iii).

Recently, robot assistants, which share working workspace with human users, are becoming a cutting-edge application of smart robots. To improve the safety and avoid collision, a robot should able to detect humans, be aware of risk, and make fast responses to minimize the risk of injury. Typically, the human body temperature is higher than the ambient temper, it is possible for a robot to detect humans by real time measuring the subtle temperature variation near the robot. To this end, Song et al. developed an ultrasensitive temperature sensor based on twisted MNFs (Fig. 6(c-i)⁹². Because the coupling efficiency between the twisted MNFs is highly sensitive to the refractive index of the PDMS cladding, which has a high thermo-optic coefficient, the sensor achieves a sensitivity of −30 nm/°C and a resolution of 0.0012 °C. As illustrated in Fig. 6 (c-ii), the output intensity curve shows a distance dependent response to a human hand in a range of 0–15 cm. When a robotic arm equipped with the temperature sensor, it can effectively detect the existence of a human, and trigger avoiding collision feedback (Fig. 6(c-iii).

With the rapid development of minimally invasive surgery and surgy robots, there is an increasing demand for miniaturized tactile sensors than can provide tactile information during the surgery. To this end, Tang et al. developed an MNF enabled compact tactile sensor with a diameter of 1.5 mm (Fig. 6(d-i)¹³⁰. The sensor can be operated in either scanning or tapping mode, achieving a hardness resolving ability even beyond the human hands. For example, it can discriminate different materials by analyzing the slope of the response curve (Fig. 6(d-ii, iii). As a proof-of-concept demonstration, the hardness discrimination of pork liver and an adductor muscle was experimentally realized (Fig. 6(d-iv). Such MNF-enabled compact tactile sensors may pave the way for hardness sensing in tissue palpation, surgical robotics, and object identification.

Typically, photoactuators convert light energy into thermal (expansion/contraction, molecule adsorption/desorption, and phase transition) or chemical energy (photoisomerization and photodimerization) in actuating materials such as carbon-based composites¹³¹, hydrogels¹³², liquid-crystal polymers¹³³, and shape-memory polymers (SMPs)¹³⁴. Among them, most photoactuators are triggered by free-space illumination, which requires a line-of-site low-loss optical path between the light source and the actuator. Thus, the utility of these photoactuators requires direct and constant line-of-site access. Besides, the intensity of free-space light may decrease significantly during long-distance transport, especially for the environment with strong absorption and scattering. Moreover, for the free-space light-driven actuators to execute locomotion or handle objects, the control light generally needs to intentionally follow the motion of the actuator. The use of optical fiber can overcome the limitation of the free-space illumination, because light can be transmitted over a long distance with low optical loss through flexible and bendable optical fibers. For example, Zhou et al. introduced a strategy for waveguiding photoactuators based on photothermally addressable nanocomposite hydrogels¹³⁵. Kuenstler et al. demonstrated a photoactuator made of liquid crystal elastomer, achieving reversible bending with a bending angle of 15° and response time of longer than 5 s⁴⁰. These reports shed light on the development of waveguide photoactuators, however, all of them suffered from limited bending amplitude and long response time owing to three possible reasons: (1) large thickness of the active layer due to the relatively large size of the commercially available optical fiber (typically >100 μm in diameter); (2) low energy density due to the non-focused light beam; (3) low optical coupling efficiency due to the size mismatch between the optical waveguide and the photo-responsive material.

To address these issues, Xiao et al. employed optical fiber tapers with enhanced optical intensity for high-performance photoactuators with large deformation and fast response³⁶. As shown in Fig. 7(a), an optical fiber taper is embedded in a PDMS/Au nanorod (AuNR)-graphene oxide (GO) photothermal film. When a control light is launched into the photothermal film via the fiber taper, photothermal heating induced by the AuNR will cause a significant expansion of the PDMS/AuNR layer due to the high coefficient of thermal expansion (CTE) of PDMS. On the other hand, GO layer undergoes negligible thermal expansion due to its low CTE. Thus, the mismatch between the CTE and deformations of two layers leads to a dramatic bending of the photoactuator. When the light is switched off, the photoactuator recovers back to its initiate state. Benefiting from the special geometric features of fiber taper, i.e., micro/nanoscale diameter and long taper region, the photoactuator features a thin active layer, high energy density and optical coupling efficiency, which enable a fast response of 1.8 s (Fig. 7(b) and much larger bending angle of 270° (Fig. 7(c) than that of waveguide actuators reported before. Table 1 provides a performance comparison of the MNF actuator with other types of photoactuators. Owing to its large deformation, the one-arm MNF actuator can grasp an ant (~20 mg) stayed on a tip immediately with the switch on of control laser and then take it off from the tip (Fig. 7(d-i). The two-arm MNF actuator can capture and lift a group of small balls with a maximum weight of 27 mg (Fig. 7(d-ii). In addition, the grasped object can be moved with high stability, which is particularly valuable or accurate control when executing a task that requires dynamic location. For example, the two-arm MNF actuator can pick up, transfer, and release a cuboid into a container ~44 cm away from the original place within 25 s (Fig. 7(d-iii, iv). These results demonstrate the advantages of the MNF actuator in accurately handling objects in a wide operating area, while free-space light-driven actuators suffering from the difficulties in accurately controlling the illuminating spot to follow the moving of actuators during operation.

In summary, the recent advances in MNF enabled tactile sensors and photoactuators were reviewed. Looking ahead, we believe that the following issues that may be worth further investigation. Firstly, scalable fabrication of MNF enabled tactile sensors with same performance remains a grant challenge. Although the fabrication of the MNFs can be well controlled, it is still difficult to automatically package MNFs to form uniform tactile sensors. With the rapid development of advanced manufacturing, we believe this issue can be overcome using programable taper drawing machine with the help of highly efficient and dexterous micromanipulation instrument. The reported MNF sensors have shown excellent flexibility, stretchability, and robustness, because the tensile strength of the tiny MNF is typically higher than 5.5 GPa⁴⁹. We found that a well packaged MNF sensor could survive after multiple cycles of a destructive loading of 14.15 MPa, which is almost 140 times atmospheric pressure¹²⁹. On the other hand, light-emitting diode and photodiode can be used as the light source and detector of the MNF sensor, which can dramatically reduce the cost of the sensing system, making it practical in wearable and portable applications. Secondly, human skin is a large area tactile organ with high spatial resolution. Meanwhile, the MNF enabled tactile sensors can only detect the stimuli at one or a few positions. How to realize distributed sensing along the MNF should be a promising field. Thirdly, for soft actuators, how to realize complex deformation and achieve more potential applications, such as searching in unstructured environment, deep-water sampling, and in vivo diagnosis/therapy, need special designs of the actuators and implementation of advanced materials. We believe that the integration of MNF photoactuator and MNF tactile sensor can dramatically enhance the actuator’s capability by a closed-loop control. Finally, to meet the increasing demand for novel applications in healthcare, robotics, and metaverse, novel sensing mechanisms or actuating strategies should be introduced into tactile sensors or soft actuators. For example, the fusion of AI algorithm with the flexible opto-electronic devices has significantly improved the sensing ability and haptic feedback experience. In terms of MNF sensors, with the assistance of AI algorithm, one can obtain complex guided modes evolution information along the tapered region with inhomogeneous diameter accommodating more higher-order modes, which can provide more sensing information. It is possible to achieve an unprecedented spatial resolution for a tapered optical fiber enabled distributed sensor with the help of advanced deep learning technology. When the above-mentioned issues are well addressed, the gap between the lab work and the application of the MNF devices can be significantly mitigated and the MNF sensors or self-sensing MNF photoactuators might be used for biopsy, measuring cell stiffness, manipulating small model organisms (e.g., elegans, larval zebrafish), or detecting pico- to femtonewton force (e.g., radiation pressure), which are difficult to realize with conventional optical sensors or photoactuators.

We are grateful for financial supports from the National Natural Science Foundation of China (No. 61975173), the Key Research and Development Project of Zhejiang Province (No. 2022C03103, 2023C01045).

The authors declare no competing financial interests.