Nanomaterial-based flexible sensors for metaverse and virtual reality applications

Nanomaterial-based flexible sensors (NMFSs) can be tightly attached to the human skin or integrated with clothing to monitor human physiological information, provide medical data, or explore metaverse spaces. Nanomaterials have been widely incorporated into flexible sensors due to their facile processing, material compatibility, and unique properties. This review highlights the recent advancements in NMFSs involving various nanomaterial frameworks such as nanoparticles, nanowires, and nanofilms. Different triggering interaction interfaces between NMFSs and metaverse/virtual reality (VR) applications, e.g. skin-mechanics-triggered, temperature-triggered, magnetically triggered, and neural-triggered interfaces, are discussed. In the context of interfacing physical and virtual worlds, machine learning (ML) has emerged as a promising tool for processing sensor data for controlling avatars in metaverse/VR worlds, and many ML algorithms have been proposed for virtual interaction technologies. This paper discusses the advantages, disadvantages, and prospects of NMFSs in metaverse/VR applications.


Introduction
In the era of massive information collection and utilization, the mobile Internet has emerged as a ubiquitous tool that has 3 These authors have contributed equally on this work. * Authors to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. made our lives more convenient. The metaverse/virtual reality (VR) concepts have been introduced to naturally integrate mobile Internet into people's work and personal lives to satisfy daily needs such as travel, meeting, shopping, etc. In particular, the metaverse represents a multi-sensor, immersive, realtime, and interactive world based on VR devices. Conventional VR systems are based primarily on silicon sensors, which have a high resolution and low noise for high-quality imaging. However, most of the existing conventional VR devices are rigid, stiff and involve cumbersome machinery and bulky systems [1]. These devices are associated with wearing discomfort, limited skin perception, and low biocompatibility, among other problems [2]. In experimental scenarios, such devices cannot be comfortably worn by the experimenters for long periods, so human physiological data cannot be effectively collected. Consequently, it is an inevitable trend to replace bulky, rigid silicon materials in VR systems with flexible nanomaterials characterized by high biocompatibility, sensitivity, conductivity, and stretchability. In general, flexible nanomaterials have changed how we make and use electronics, and their development has thus attracted significant attention in industry and academia [3,4]. Nanomaterial-based flexible sensors (NMFSs) can overcome the problems associated with conventional silicon metal oxide sensors by ensuring user comfort, skin integration, versatility, and remote intelligent detection when combined with deep learning algorithms. Flexible and lightweight electronic devices can be placed in stable and close contact with human skin without any discomfort, allowing even subtle movements to be accurately captured. In this review, the performance of flexible sensors is evaluated in terms of the mechanical, physical, chemical, and electrical aspects. The existing NMFSs are compared in terms of mechanical and electrical parameters such as conductivity, stretchability, stability, sensitivity, energy consumption, and biocompatibility. The typical strategies for developing multifunctional nanomaterial electronics and systems are discussed.
In general, the emergence of NMFSs as an alternative to silicon sensors has expanded the opportunities for metaverse/VR systems. Owing to their higher sensitivity [5,6], conductivity [7,8], stretchability [9], and biocompatibility [10] than those of silicon sensors, NMFSs can be used to develop innovative VR systems for machine vision [11,12] or medicine [13]. However, the use of NMFSs in practical VR-based systems requires improvements in certain domains such as the stability of flexible sensors, self-healing ability, self-cleaning ability, self-powering ability, and large-area and batch preparation ability. Recent reviews of VR system electronics have covered many areas, such as materials (e.g. active [14], textile materials [15,16]), interface devices and technologies (e.g. elastic tactile devices [17], wearable friction devices [18], thermal tactile devices and technologies [19,20], vibro-haptic interfaces [21], bionics and structures [22,23] Notably, these reviews focused on only limited properties of materials and the functionality of VR equipment and related technologies. In contrast, this research is aimed at providing a comprehensive review of flexible nanomaterials with multimodal features [39,40], and integrated systems [1], combining artificial intelligence (AI) and nanomaterial interfaces for advanced metaverse/VR systems. A conceptual illustration of future multimodal perceptual interface interactions is shown in figure 1.
Compared with traditional flexible sensors, NMFSs have the following advantages [41,42]: (1) high sensitivity: since nanomaterials have a high surface-to-volume ratio, they can interact more effectively with the external environment, thus increasing the sensitivity of the sensor. (2) Low power consumption: nanomaterials can operate at low power, which is useful for systems such as mobile devices and VR headmounted displays. Sensors made with nanomaterials can provide high performance while reducing power consumption and extending battery life; (3) Malleability: nanomaterials are very flexible and can be produced in a variety of shapes and sizes for a wide range of application scenarios. They can be easily integrated into the sensor design, increasing the plasticity of the system; (4) Reliability: the properties of nanomaterials can make sensors more stable and reliable. For example, some nanomaterials have better resistance to oxidation and corrosion, which can make sensors more durable and long-lasting; (5) Large-scale fabrication: more and more nanomaterial fabrication techniques are being developed, and the fabrication of large-area flexible sensors is another major advantage of nanomaterials over traditional flexible sensors. These can help VR/metaverse systems to better integrate with different parts of the human body. Thus, providing a smoother, more realistic immersive experience.
This review summarizes the different approaches to the development of nanomaterials and flexible sensors based on these materials for VR systems. To ensure secure attachment to the skin, the flexible functional materials must have mechanical properties similar to those of the skin to maximize the perception of various feedback signals, reduce the mechanical and thermal loads experienced by the human body, and prevent any skin irritation or allergic reactions. Additionally, such materials must be breathable, wearable, and biocompatible to maximize user comfort. In this section, we have described the limitations of traditional VR devices and discussed the advantages of NMFSs as potential VR devices. Section 2 describes the classification of flexible nanomaterials based on their dimensionality. Section 3 describes the application interface of nanomaterials and the trigger mechanism of VR interaction. Section 4 discusses the applications of machine learning (ML) for VR. Section 5 presents several remarks on the future of VR systems in their development.

Fabrication
Methods of manufacturing nanomaterials can be classified as either top-down or bottom-up [42]. The top-down approach involves starting with larger structures and then using increasingly finer tools to create correspondingly smaller structures. Such as laser interference [ . This method produces fine structures of nanomaterials and allows precise control of material size and structure, but is generally more expensive and requires specialized instrumentation (e.g. lasers, lithographs, inkjet printers, etc). The bottom-up approach refers to the self-assembly of smaller particles based Copyright 2021 American Chemical Society. An electronic skin that can rely on the mechanical force of the skin for interaction and provide physiological signal monitoring (iii). From [45]. Reprinted with permission from AAAS. And a sensor that provides feedback on body surface temperature and ambient temperature (iv). [ [57], and layer-by-layer (LBL) assembly [58]. This method is generally inexpensive, requires no special equipment, and is widely used in scientific research, but it is not possible to precisely control the size and orientation of the material assembly.
Nanomaterials of different dimensions will be discussed in this review. However nanowires (NWs) and nanotubes are more effective than nanoparticles in forming conductive pathways. Metal NWs and carbon nanotubes (CNTs) are widely used as conductive fillers because of their high conductivity, good compatibility, and ease of device fabrication.
NMFSs are formed by mixing or depositing nanomaterials with polymers or fabrics on their surfaces. The distribution of nano conductive fillers has a critical effect on the mechanical and electrical properties of NMFSs [59]. Meanwhile, nanomaterial composites are designed to make better use of the properties of different nanomaterials, mainly in terms of sensitivity, conductivity, responsiveness, stretchability, and stability to significantly improve the electrical and mechanical performance of sensor devices [60]. Lim et al reported an ultra-low percolation threshold (1.56 vol%) method for the synthesis of whisker gold nanosheets (WAuNS), in which the addition of platinum-plated W-AuNSs to the prepared nanocomposites significantly improved the electrical conductivity charge storage capacity. This results in devices  [91,92] based on W-AuNS stretchable nanocomposites with excellent electrical conductivity and high stretchability [61]. The stretchability of nanocomposites is one of the important properties of flexible sensors and is also a major factor affecting the electrical and mechanical aspects of the sensors. Jung et al prepared highly conductive and stretchable nanocomposites using zero-dimensional (0D)/one-dimensional (1D)/2D silver nanomaterials that exhibited resistance variations in the range of <1% over 1000 strain cycles at <50% uniaxial and biaxial strain, while the electrical conductivity was as high as ≈31 000 S cm −1 [62].

Classification
Nanostructured materials may refer to materials with nanoscale internal or external structures or materials containing internal or surface nanostructures [63]. Multiphase materials with the phase being nanomaterials or the composite phase having nanoscale distances between them are called nanocomposites [64]. Many innovative nanomaterials, differing in structure, shape, size, and chemical synthesis method, have been applied to prepare NMFSs. Based on the theory proposed by Pokropivny and Skorokhod, nanomaterial can be classified according to its shape and size [65]. Nanomaterials are typically classified according to the paths along which electrons can move in the material. Electrons are generally stable in 0D nanomaterials. In 1D nanomaterials, which are of the order of 100 nm or smaller, electrons can move along uniaxial paths. In two-dimensional nanomaterials, electrons can move along biaxial paths [64]. NMFSs are widely used in VR systems owing to their large surface area, scalability, high biocompatibility, and compatibility by applying appropriate manufacturing processes (table 1). These materials can be evaluated based on their mechanical, chemical, and electrical properties and application, and provide the following advantages for NMFSs in VR devices [1,66]: (1) better flexibility and adaptability: NMFSs can adapt to the surface of objects with different shapes and curvatures, thus better matching the curves and contours of the human surface and then improve the accuracy and precision of the sensor. (2) Lighter: NMFSs are very light and can reduce the weight of a VR device which helps users wear the device for long periods without feeling overweight or fatigued.
(3) Lower power consumption: NMFSs can operate under low power conditions, reducing energy consumption and extending the life of energy storage devices. (4) Better biocompatibility: the good biocompatibility of NMFSs can improve the safety and comfort of VR devices. (5) Higher sensitivity: NMFSs can detect small movements and changes, improving the interactivity and accuracy of VR devices, and giving users a more realistic experience. (6) Better durability: NMFSs can usually withstand prolonged use and a lot of bending and twisting, and have a longer lifespan, reducing maintenance and replacement costs. These features can provide a more accurate, realistic, and comfortable experience for VR devices. In this section, recently reported nanomaterial-based sensors are With recent advancements, flexible sensors, and electronic skin have been widely used in healthcare and human-computer interaction applications. In terms of horizontal intermolecular forces, effective resistive or capacitive sensing must be realized to transform sensor deformations into electrical signals. Unlike traditional metal strain sensors, research on sensors developed using 0D nanomaterials is in the nascent stage. However, such nanomaterials can promote the development of flexible sensors. Nanoparticles, as typical 0D nanomaterials, have been prepared by various methods, such as co-precipitation, hydrothermal technique, inert gas condensation, sputtering, microemulsion method, microwave-assisted, laser ablation, sol-gel, ultrasound, spark discharge, template synthesis, biological synthesis, etc [93]. Table 2 summarizes the sensing principles and performance metrics of several NP-based flexible sensors.
This section focuses on NPs as representative 0D nanomaterials and summarizes the relevant research on their use in NMFSs. NPs typically have diameters of no more than 100 nm and can be divided into metal NPs (e.g. Au [104], Ag [85]) [64]. Metal oxide NPs [105] and non-metal oxide NPs [99] and organic NPs [106]. Metal NPs are the most widely used 0D nanomaterials in sensors. High sensitivity can be obtained owing to the excellent electrical conductivity, spatial properties, and unique electron transfer properties of metal NPs. Moreover, NPs have higher plasticity than other nanomaterials. A flexible substrate is typically used to construct a conductive network structure. According to the penetration abilities, the charge conduction between NPs can be divided into contact type, tunneling effect, and variable-range jump. NPs are typically used to fabricate highly sensitive devices based on resistive sensing [107]. In particular, NPs are widely used to prepare flexible sensors, for the following four reasons: (1) NPs can be easily prepared in large quantities; (2) they can bind with a variety of molecular ligands; (3) the size, shape, and composition of NPs can be controlled [108]. (4) The particle spacing can be controlled when preparing thinfilm materials [6].
The spatial characteristics of NPs can be exploited to fabricate high-performance flexible strain sensors using various strategies, for example, binding to conductive hydrogels [109,110], cross-linking with molecules [111], ligand exchange [112], self-assembly [113], atomic layer deposition [114], drop-casting [115], dip coating, and spin-coating [116]. Zhao et al demonstrated that lignin-coated silica NPs can enhance the elasticity of conductive hydrogels to prepare flexible sensors with high stability and repeatability [99]. The electrically insulating nature of the linker leads to tunneling charge transfer in the AuNP network (figures 2(a) and (d)) [6]. NPs form hydrogels in characteristic solutions through dynamic electrostatic interactions and hydrogen bonding, and the resulting sensors exhibit high scalability (>1000%), rapid self-healing ability (93.3% after 10 min), low detection limits (0.5% strain and 25 Pa), and excellent fatigue resistance (>1000 cycles) [106]. Zhang et al fabricated a strain sensor based on gold NPs interconnected by flexible and responsive molecular joints. The sensor exhibited a high specification factor, high sensitivity, and high robustness against tensile and compressive strains [117]. Strain sensors can also be prepared by the cross-linking of organic molecules with NPs. The connection between molecules relies mainly on electrostatic adsorption and chemical bonding. The strain sensitivity in NP-based sensors is attributable to the disturbance of the charge carriers between adjacent AuNPs by the tunneling electrons (figures 2(b) and (e)) [111]. NP-driven sensors can be prepared using Langmuir-Blodgett assembly and LBL assembly techniques in a low-cost and facile manner while precisely controlling the membrane growth and fabrication process [113].
In particular, the LBL assembly process involving immersion and spin-coating is a simple and effective method for assisting NP adsorption (figure 2(c)) [116]. The tunneling (h) Schematic diagram of the tunneling current of the conducting network of graphene nanoparticles excited by ultrasound (Reprinted from [121], Copyright 2019, with permission from Elsevier). (i) Schematic of electron transfer in a healable AgNS network (Reproduced from [8]. CC BY 4.0). (j) Resistance variation of strain sensor based on CB/AgNPs during the tensile process (Reproduced from [98]. CC BY 4.0). (k) The change of the connection resistance of antibodies-Au@ magnetic nanoparticles thin films magnetically assembled between two electrodes [124]. John Wiley & Sons. effect and percolation theory are the main sensing mechanisms of NP-based sensors. Conductive thin-film sensors composed of NPs with polymers can be applied to fabricate high-performance flexible sensors. In general, the electron tunneling model is highly sensitive to the polymer configurations and gaps between NPs. The variations in these gaps and polymers under deformation depend on the relative position of adjacent NPs, which in turn determines the electron transport direction and overall sensor response [118]. In flexible nanonetworks, the interfacial interaction between the conductive filler and polymer matrix involves electrostatic interactions or hydrogen bonding (figure 2(f)) [106]. Mechanical loading enables the tracing of the variable electrical transport properties. Specifically, in rotating electron transport, nanocomposites are connected by thousands of single tunnels, and these serial interconnections can be determined by mechanical loading (figure 2(g)) [119]. Based on the tunneling effect and distribution of NPs, the sensor performance can be evaluated to guide future sensor design [120]. Liao et al performed molecular dynamics simulations to develop a thin-film type piezoelectric sensor formed by graphene NPs. When an ultrasonic signal traverses the sensor, the acoustic pressure triggers a quantum tunneling effect in the permeable NP network, which affects the piezoelectric properties of the sensor (figure 2(h)) [121]. Suh et al developed an electron-tunnelingassisted permeation network consisting of silver nanosatellite (AgNS) particles with high conductivity and full reversibility for oily, healable, and malleable nanocomposites. An exceptionally high conductivity (1.02 × 10 3 S cm −1 ) was achieved in oily nanocomposites. The AgNS particles were uniformly and densely distributed with a particle gap of 3.1 nm, which enabled a fully reversible reconstruction of the permeation network. Moreover, an electrical healing efficiency of ∼100% was achieved in up to 1000 cycles of fracture/healing tests (figure 2(i)) [8].
The tunneling effect is also widely exploited in resistive sensors [122]. Zhang et al considered the connection resistance between adjacent NPs (red circles) as a variable resistance composed of R con (constant during deformation) and R tun and R disc that varies with strain. The resistance of carbon black/AgNP-based strain sensors changes with variations in the distance between neighboring NPs as the applied strain increases during stretching (figure 2(j)) [98]. In LBL techniques with selective deposition and aligned placement of NPs and magnetically assembled NP films, inter-electron transport and hopping between nanofibers serve as the main pathways for electrical signaling (figure 2(k)) [123,124]. Notably, due to the weak interaction forces between adjacent NPs, the number of dispersed NPs gradually increases as the number of repetitive strain/release cycles increases, resulting in an irreversible increase in the resistance of the conductive layer.

1D nanomaterials.
The 1D nanomaterials for fabricating flexible sensors are typically selected considering the conductive component and elasticity coefficient. Highly conductive materials can enhance the conductivity of the sensors. Typical 1D nanomaterial preparation methods such as chemical vapor deposition (CVD), electric-arc discharge, laser ablation, laser torching, and liquid electrolysis are commonly used to prepare CNTs [125]. While anodic oxidation techniques, electron beam lithography, nano-imprint lithography, hydro/solvothermal synthesis, template-assisted, electrochemical deposition, and physical vapor deposition are commonly utilized to prepare NWs [126]. This section focuses on the 1D nanomaterials used for fabricating flexible sensors, such as CNTs [127], metal and metal oxide NWs (gold [128], silver [129], copper [130], zinc oxide [131]) (table 3).

CNTs.
The main methods for preparing CNTs have been mentioned above [142], and among all the major synthesis methods, CVD is the most widely used method for synthesizing CNTs because it has some distinct advantages, such as the simplicity of the method, low cost, large scale production, high purity CNTs can be obtained, and it is suitable for standardization in the industry [143]. According to their shapes, carbon-based nanomaterials can be divided into CNTs, carbon black, graphene, and fullerenes. This section explores the application of 1D CNTs in NMFSs. CNTs have high intrinsic carrier mobility (10 6 cm 2 VS −1 ), high electrical conductivity (10 4 S cm −1 ), and high chemical stability. Consequently, the resulting sensors exhibit high flexibility and excellent mechanical properties [144]. Polymeric conductive nanocomposites mixed with CNTs are promising materials for developing flexible sensors. In particular, liquid polymers can easily adsorb onto CNT foam, thereby filling the gaps of the CNT network in the nanocomposite and endowing the nanocomposite with excellent electrical conductivity [127].   [133]. Copyright 2020 American Chemical Society). (d) 3D diagram and sensitivity of capacitive pressure sensor based on CuNWs/PDMS electrode (Reprinted from [165], Copyright 2022, with permission from Elsevier). (e) Adsorption mechanism of water molecules and schematic diagram of surface acoustic wave sensor based on ZnO NWs Reprinted with permission from [174]. Copyright 2020 American Chemical Society.
researchers applied CNT-based nano-compounds to prepare flexible sensors and synthetic muscles (figure 3(a)) [136]. In general, CNTs are typically applied as electrodes or flexible sensor components [145]. Multi-walled CNTs (MWCNTs) can be combined with different types of polymers and conductive nanomaterials to form sensing materials or electrodes for sensors. This configuration allows MWCNT-based sensors to infer the macroscopic and microscopic stress variations from structural changes, rendering them advantageous for human body detection [146]. For example, Suo et al developed an MWCNT-filled PDMS-based sensor array for the detection of muscle mechanical activities [147].

AuNWs.
Although the mechanical flexibility of metals is typically low, metal NWs with excellent electrical conductivities and high aspect ratios can act as stretchable electrodes and sensing materials to obtain flexible and electrically conductive NMFSs. Permeation or network formation are typical strategies for preparing NMFSs based on conductive nanomaterials. High aspect ratios ensure electrical contact between the NWs under high tensile strain, and the resulting NMFSs exhibit high conductivity and stretchability. AuNWs have attracted considerable research interest owing to their high electrical conductivity, biocompatibility, stretchability, optical and mechanical properties, unique morphology, and chemical inertness [148,149] and have been widely applied to develop NMFSs [150,151]. The synthesis of AuNWs in solution has been extensively studied in the literature, among which hydrothermal synthesis is the most widely used, and polyvinylpyrrolidone and hexadecyl-trimethylammonium bromide (CTAB) are the most commonly employed crystal seeding agents and surfactants for the preparation of Au NWs by hydrothermal synthesis [152].
AuNWs can also be synthesized through chemical reduction growth and electrodeposition [149,153]. An et al developed a frictional, self-powered, electronic tattoo-like standing AuNW (V-AuNW) based flexible nanomaterial sensor that can generate signals even when mechanically deformed by up to 500%. Furthermore, V-AuNWs can be embedded in a variety of ultrathin elastic structures that can be attached to human skin and combined with AI to create interactive pressure-sensitive tattoos. This self-powered tattoo-like flexible nanomaterial sensor can be integrated with a circuit board to wirelessly control electronic devices to enable human detection [154]. Additionally, V-AuNWs can be combined with pyramidal microstructures to design highperformance NMFSs. This sensor exhibits high sensitivity, low pressure, high durability, fast response (<10 ms), and signal stability (10 000 cycles). When this sensor is combined with a Bluetooth low-energy module, high-quality wrist artery signals can be detected wirelessly (figure 3(b)) [151]. Furthermore, by impregnating textiles with AuNW solution, they can be converted into conductive nanomaterials. Such AuNWs/textile conductive nanocomposites are excellent flexible sensors for monitoring human movements [132]. Based on a three-dimensional AuNW sponge antenna, Wang et al demonstrated that a battery-free wireless pressure sensor could detect a wide range of pressures. The wide detection range (0-248 kPa) of the sensor can be adjusted by controlling the elastomer hardness or sponge thickness [128].

AgNWs.
AgNWs exhibit stable physicochemical properties, excellent mechanical flexibility, electrical conductivity, and thermal transparency and can thus be used to fabricate NMFSs [155,156]. Various methods for synthesizing AgNWs were originally derived from those for preparing metallic NPs. Initially, AgNWs were mainly prepared by electrochemical methods. However, these methods resulted in low yields and size heterogeneity. Later, other methods such as the polyol method [157], hydrothermal methods [158], template techniques [159] (hard and soft template methods), and photochemical reduction methods were developed. The template technique and multivariate method are the most commonly used. Specifically, the template technique results in high yield and uniform size and morphology of the AgNWs, and the multivariate method is characterized by facile production, low cost, and simplicity [129,160]. Zhu et al used a glass template approach to fabricate high-performance NMFSs with a hybrid structure involving a gradient-like distribution of AgNWs and polyurethane (PU). The PU/AgNW-based flexible nanomaterial sensor exhibited a fast response (20 ms), high sensitivity, wide detection range, high stability (2300 cycles), and excellent flexibility (figure 3(c)) [133]. AgNWs are commonly prepared through chemical synthesis techniques [158]. Additionally, polyol and one-pot processes have been used to prepare transparent and flexible localized AgNWs for 3D touch flexible sensors with 3D touch systems that can map complex 3D structures using pressure as a third coordinate axis. Moreover, 3D data can be successfully collected in wireless and wearable conditions [161]. Yang et al blended AgNWs into PDMS/polyvinylidene difluoride (PVDF) electrospun films to generate a robust structure involving a conductive network. Sandwich-structured sensors were prepared with a GF of up to 654.5. The high sensitivity of the sensor was attributable to the microcrack structure in the sensing layer. A sensor GF of up to 3058 was obtained in the presence of a wrinkled pattern [159]. The orientation of the AgNWs affects the sensitivity of the resulting flexible strain sensors: the strain sensor with vertically aligned AgNWs exhibits the highest sensitivity (GF = 89.99) at 25% tension, which is 7.08fold higher than that shown by horizontally aligned AgNWs (GF = 12.71) [162]. The stability and uniformity of AgNW networks must be enhanced to realize the large-scale commercialization of AgNW-based flexible sensors.

CuNWs.
CNTs, AuNWs, and AgNWs represent the most promising candidates for soft electronics. However, CNTs exhibit low conductivity, and gold and silver NWs are characterized by high costs or low yields. CuNWs, which have a high conductivity (16.78 nΩ m) and yields, are compatible with solution phase synthesis [163] and thus represent promising frameworks for the commercialization of flexible sensors and soft electrodes [164]. The main methods for synthesizing CuNWs include catalytic synthesis, the hydrothermal method [165], and the hydrazine method. Catalytic methods allow the size and unitarity of CuNWs to be controlled; hydrothermal methods yield CuNWs with large aspect ratios; and hydrazine methods are characterized by high yields and fast processing. Therefore, the synthesis strategy must be selected considering the application requirements [163]. Highquality CuNWs can be synthesized on a large scale through a facial aqueous reduction method [166]. Wu et al used this method to prepare high-quality CuNWs that were mixed with graphene to form flexible core-shell aerogels with excellent electrical and mechanical stability [167]. Yu et al developed a highly sensitive and flexible capacitive pressure sensor consisting of a CuNW network and a wrinkled PDMS elastic membrane. The proposed sensor exhibited a pressure sensitivity of 0.162 kPa −1 and high performance in the stretched state (<200%). Furthermore, this sensor was stable over strains from 0% to 80%, with the variation in the sensitivity factor being lower than 0.8%. Therefore, the proposed sensor could be used as an elastic skin strain sensor to monitor large human limb motion (figure 3(d)) [165]. By exploiting the permeable network structure of the CuNW electrode and composite film of the polymer, a sensor with a high light transmission of 81.1% was developed, which was seven times more sensitive than an elastomeric piezoelectric sensor with the same design. Furthermore, the sensor was stable (1000 cycles) under the effect of the strong affinity between the resin and CuNW network [130]. Notably, the tendency of CuNWs to aggregate remains a technical challenge in building electronic devices. Therefore, different printing techniques are gradually being developed to prepare high-quality CuNPs for the emerging applications of CuNW flexible sensors [168].

ZnO NWs.
1D ZnO NWs have been widely used in NMFSs because of their unique piezoelectric properties and high electron mobility. The preparation methods include hydrothermal [169], thermal oxidation [170], and vapor-liquid-solid [171]. Hydrothermally grown ZnO NWs Thermoresistive
are typically used to prepare flexible self-powered pressure sensors [172]. Chakraborty et al demonstrated the ultraefficient, convection-assisted, seed-free synthesis of highdensity, vertically aligned ZnO NWs over large areas by heating substrates with adjacent temperature-controlled surfaces [131]. Yin et al developed a transparent, super-sensitive, and flexible humidity sensor using a mixture of ZnO NWs and graphene as a composite sensing layer. Owing to the large specific surface area of ZnO NWs [173], the sensor exhibited high humidity sensitivity, stability, and repeatability and could detect human breath (figure 3(e)) [174]. The same hybrid nanomaterials were also used to construct a flexible bifunctional sensor that could detect vertical pressure and ultraviolet (UV) light. The pressure sensitivity at low pressures was 1.7 nA kPa −1 [175]. In general, ZnO NW blends are inexpensive and can be easily processed, rendering them promising for flexible electronic device applications. The electrical conductivity of flexible electronic devices can be enhanced by increasing the number of charge movement paths within the material. Because electrons can move in a biaxial direction in 2D nanomaterials, these materials' conductivity is higher than that of 1D nanomaterials. Therefore, 2D nanomaterials have attracted significant research interest. Choi and Kim critically assessed the advancements and limitations of 2D nanomaterial strategies in biomedical engineering considering their unique physical, chemical [178], electronic [179], and optical properties conferred by the planar topography of 2D nanomaterials [180]. This section focuses on the hierarchical structural advantages and applications of 2D nanomaterials (i.e. black phosphorus (BP) [181], Ti 3 C 2 Tx titanium alloy (MXene) [182], graphene and its derivatives [183], titanium carbide [184], transition metal dichalcogenides [185] and transition metal oxides [186]) in wearable and flexible electronic devices [187]. The structure of 2D nanomaterials can be optimized to overcome the existing limitations of 1D nanomaterials (e.g. increased surface area, improved elasticity, and enhanced electrical properties) and prepare high-performance flexible electronic devices (table 4).

Graphene and its derivatives.
Since its discovery in 2004 [200], graphene has been widely incorporated into flexible sensors and electronic devices [201,202] owing to its excellent mechanical properties, flexibility, transparency, electron mobility, electrical conductivity, and high Young's modulus. Derivatives of graphene include graphene oxide (GO) and reduced GO (rGO). The synthesis of graphene can also be performed by two main methods: top-down and bottom-up methods. Top-down methods include unzipping of CNT, electric arc discharge, liquid phase exfoliation, mechanical exfoliation, and oxidation-reduction reactions. The bottom-up approach is to synthesize graphene and its derivatives using other carbon materials, including template preparation, epitaxial growth, CVD, total organic synthesis, and substrate-free gas-phase synthesis [142]. In the past decades, laser-assisted technologies, such as laser-treatment-induced flexible patterning of graphene, layered structures, etching, and shock [203], have emerged as promising tools for graphene synthesis. Yan et al developed a flexible 16 × 16 pixelated triboelectric sensing array (TSA) with a resolution of 8 dpi for self-powered real-time haptics by patterning laserinduced graphene electrodes (7 Ω sq −1 ) through complementary cross-over between the top and bottom electrode arrays. Due to its unique pattern structure, the TSA is optimized in terms of the complexity and number of data channels, enabling real-time visualization of multi-touch and motion tracking functions at zero power consumption. In addition, an intelligent wirelessly controlled human-machine interaction (HMI) device was built, which consisted of a 9-bit digital array touchpad that can wirelessly control personal electronic devices (figure 4(a)) [204].
A prerequisite for stretchable energy storage devices is the autonomous resilience of electronics. However, most materials exhibit limited stretchability or complex patterning during strain adaptation. These problems can be resolved using graphene. Park et al developed a stretchable microsupercapacitor based on an rGO/gold heterostructure by direct laser patterning. Upon transfer to PDMS, the capacitor achieved a high conductivity (∼10 5 Sm −1 ), and it maintained a conductivity (∼10 4 Sm −1 ) even at 50% strain (figure 4(b)) [205]. Despite the success of graphene-based NMFSs, most existing flexible sensors can only detect strain amplitude and cannot distinguish strain direction, thereby limiting the information completeness of strain vectors. Huang et al successfully developed wearable strain vector sensors by preparing large-area parallel-aligned vertical graphene. This sensor can sensitively detect both the direction and amplitude of the strain vector with excellent accuracy. Specifically, the strain vector sensor could recognize the direction and amplitude of human finger rotation (figure 4(c)) [206]. To develop electronic skin and robotics, it is necessary to use flexible devices that can be affixed to any surface to be tested. Temperature is a fundamental physical quantity and a key indicator in many detection scenarios. Liu et al reported a flexible temperature sensor with high performance and mechanical properties based on rGO, which is a sensitive material. Moreover, the rGO-based flexible temperature sensor, which is stable under different types of pressures and not affected by the external environment, can be applied for object detection or robot skin development ( figure 4(d)) [195]. Notably, NMFSs can not only monitor a wide range of human microscopic activities [170] but also be used for temperature [188], humidity [207], vibration, and sound detection (figure 4(e)) [194].

Titanium carbide (Ti 3 C 2 Tx
). In 2011, researchers at Drexel University reported a novel 2D material, i.e. a transition metal carbide/nitride termed MXene [208]. In general, selective etching of Al layers in Ti 3 AlCl 2 powder using hydrofluoric acid is the most commonly used method for the preparation of 2D Ti 3 C 2 Tx MXene layers [209]. 2D Ti 3 C 2 Tx MXene films have unique microstructural and physicochemical properties such as high tunable electrical conductivity (up to 20 000 S cm −1 ), high bulk capacitance (1500 F cm −3 ), high planar mechanical strength (570 MPa), high flexibility, easy processing. Adjustable porosity, and easy processability, rendering them suitable for use in flexible electronic filmbased devices [210]. Ti 3 C 2 Tx spontaneously forms accordionlike sheet structures on the microscopic scale with a large specific surface area (figure 5(a)) [211]. The flexible substrate in NMFSs typically consists of a hydrophobic elastomer. The interaction between substrates can be tweaked by adding nano-filling materials while maintaining the excellent conductivity of MXene [193]. MXene can also be combined with single-walled CNTs (SWNT) to create flexible sensors with 3D microstructures that are highly sensitive to speech vibrations and hold great promise for speech acquisition and recognition ( figure 5(b)) [212]. The stretchability of conductors under large deformation and bending is essential for flexible electronic devices, and the device design must be optimized [189]. The metallic conductivity of MXene exceeds that of other solution-treated nanomaterials, and it can be prepared as a conductive and stretchable fiber. Seyedin et al used the wetspinning technique to fabricate MXene/PU composite fiber strain sensors with a strain factor of approximately 12 900 (≈238 at 50% strain) and a maximum stretch range of ≈152%. MXene/PU fibers can be woven into a one-piece elbow sleeve with textile threads to enable load-free detection of elbow movements (figure 5(c)) [213].
Notably, for flexible electronic applications, it is necessary to ensure large deformation and the ability to achieve HMI with minimal external stimuli. Inspired by the sensing mechanism of the human skin, Wang et al developed thin films based on Ti 3 C 2 /natural microencapsulated bio-composites by mimicking the hierarchical, interlocking, and patterned structures of human skin. The pressure sensor with layered and interlocking structures exhibited a 9.4-fold increase in pressure sensitivity (24.63 kPa −1 ) relative to that of the Ti 3 C 2 flexible sensor with a planar structure (2.61 kPa −1 ). Additionally, the sensor exhibited a fast response time (14 ms) and high cycling repeatability (5000 cycles) (figure 5(d)) [184]. Recently, a demand for heating functionality has emerged for wearable electronic devices and medical sensing devices. Park et al used an electrostatic assembly to fabricate a metal-like 2D  [205]. CC BY 4.0). (c) Principle diagram of PAVG strain vector sensor for detecting strain direction and amplitude (Reprinted with permission from [206]. Copyright 2019 American Chemical Society). (d) Application of skin temperature sensor based on reduced graphene oxide (Reproduced from [195]. CC BY 4.0). (e) Schematic of E-GWF manufacturing process and sensitivity (GF) (Reprinted with permission from [192]. Copyright 2019 American Chemical Society).
MXene heater on the substrate surface of a negatively charged MXene sheet, which exhibited excellent optical properties (>65%), fast electrothermal feedback, low sheet resistance, flexibility, and intrinsically high electrical conductivity. The heater was noted to be suitable for not only defrosting devices and local heating but also thermal therapy and health monitoring (figure 5(e)) [214]. Flexible smart textile devices are being increasingly applied for human monitoring and healthcare. However, it remains challenging to prepare multifunctional wearable smart textiles based on MXene. Luo et al fabricated waterproof and breathable smart textiles with stable photothermal and electrothermal conversion properties, strain sensing properties, and temperature sensing capabilities (figure 5(f)) [196].

BP.
Since the 2D BP first discovery in 2014 [215], it has attracted widespread attention because of its unique folded monolayer structure, high biocompatibility and biodegradability, anisotropic electrical conductivity, extraordinary surface activity, and high hole mobility, and applied in various domains such as catalysis, energy storage devices, and sensors [181]. Synthesis processes for bulk BP include catalysis, high pressure, chemical vapor transfer, and high-energy mechanical milling. Layered BP manufacturing processes include electrochemical exfoliation, liquid phase exfoliation, mechanical exfoliation, and laser-assisted exfoliation [216]. This subsection focuses on the research progress of 2D BP in NMFSs. NMFSs for next-generation medical devices must be able to accurately and continuously detect (f) PDA/Mxene/PDMS textile preparation process and structure schematic diagram, as well as human motion detection applications (Reprinted from [196], Copyright 2021, with permission from Elsevier).
physiological signals under self-powered conditions. Zhang et al designed a self-powered smart sensor system based on MXene/BP using an LBL self-assembly process with direct laser writing technology. In addition, using MXene/BP as the sensing layer of the pressure sensor, the pressure sensitivity of the device could be increased to 77.61 kPa −1 , a response time of 10.9 ms could be attained, and the smart sensor system could detect human pulses in real time under physiological conditions (figure 6(a)) [217]. A hybrid framework of BP and LEG (BP@LEG) was used to develop a highly sensitive dual-modal temperature-strain hybrid sensor to modulate eskin sensing functions. The hybrid sensor exhibited excellent properties such as an ultra-low strain resolution of 0.023%, high strain sensitivity (GF) of up to 2765 (>19.2%), high thermal index of 8106 K (25 • C-50 • C), and long-term durability (>18 400 cycles) and thus represented a promising device for future robotics and human health monitoring technologies (figure 6(b)) [197].
Textiles capable of harvesting energy from various body activities and physiological signals through frictional electrical effects are also in demand for developing flexible electronics. Xiong et al blended BP with cellulose oleate NPs as a synergistic electron capture coating to prepare textile nanogenerators with long-term reliability and frictional electrical properties. A skin-triggered textile-based frictional electric nanogenerator was developed for harvesting mechanical energy from deliberate and non-deliberate body movements. Large outputs (∼250-880 V, ∼0.48-1.1 µA cm −2 ) were obtained when the sensor was manually touched with a certain force (5 N) and frequency (4 Hz). Moreover, a low output of 60 V was also achieved during accidental contact with the skin when mounted on clothing or skin (figure 6(c)) [218].  electrode based on a hybrid nanocomposite (polyaniline (PANI)@BP) that could provide energy for biometric devices. An integrated device powered by this capacitor was used to track the human heartbeat, and it was noted to be promising for different real-time health monitoring applications (figure 6(d)) [219]. Owing to its unique structural and material properties, BP has been used to prepare various nanomaterials and improve the properties of nanofilms through innovative structural design. BP can potentially be applied in flexible sensing, energy storage, electrical device, and medical detection applications [220].

2D transition metal/MXene.
Owing to their unique nanostructures and excellent natural properties, 2D transition metals are promising candidate materials for storage devices and sensing applications. MXenes, as representative nanomaterials in the family of 2D transition metal nitrides, carbides, and carbon tetrads, have large surface areas and inherent semiconducting properties. Consequently, they have been widely applied to develop flexible electronic devices [221]. The synthesis of MXenes relies on several etchants, namely hydrofluoric acid-based etching, molten salt etching, UV-assisted etching, electrochemically assisted etching, and alkali-based etchants [222]. Large-area, flexible tactile sensors with high toughness and adhesion have attracted significant research attention. The existing passive tactile sensors are susceptible to multi-line crosstalk, which limits the accuracy of the output signal. Active-matrix-driven tactile sensors can overcome these obstacles, but the transistor array film must be adequately stable to effectively control the integrated electronic components. Shinde et al presented a waterassisted transfer process for the growth of metal-organic chemical vapor-deposited MoS 2 films using hydrofluoric acid for the surface treatment of silica. This treatment enables the direct transfer of large, sequential, and defect-free MoS 2 films, enabling the direct fabrication of flexible electronic devices ( figure 7(a)) [223].
Similarly, Park et al fabricated a large-area active-matrix array of tactile sensors (8 × 8 arrays) by exploiting the excellent mechanical properties of MoS 2 . This tactile sensor has a wide sensing range for object shapes (1-120 kPa), high  1-20 kHz), the device could be used as a microphone, speaker, or body temperature detector. When used as a microphone, the device exhibited a speech recognition accuracy of 98% and could measure an audio signal with a performance equivalent to that of a commercial microphone. When used as a loudspeaker, it exhibited a high sound pressure level (∼90 dB), low total harmonic distortion (∼1.41%), and uniform directivity (standard deviation ∼4 dB). As a flexible transducer, the device exhibited high breath-monitoring performance and a high-frequency temperature coefficient of −289 ppm K −1 (figure 7(c)) [225]. With the development of 2D transition metals, materials with various properties have been assembled to prepare nextgeneration flexible sensing devices for modern medicine and AI. Zhang et al hybridized vanadium nitride (VN) nanosheets with vertically aligned CNT arrays (VN/CNTs) to fabricate metal aerogel composites for flexible strain sensors. The 2D VN nanosheets provided the main skeleton structure, and the sandwiched aerogel formed a highly conductive spatial network, allowing the flexible sensor to demonstrate excellent sensitivity, high strain factor (GF = 386), fast response, high stability, and structural compatibility over a small strain range of 10% (figure 7(d)) [191].

Hybrid nanomaterials.
Many researchers have explored the sensing properties of NMFSs, focusing on nanomaterials with different dimensions. Notably, individual nanomaterials cannot simultaneously satisfy the characteristics of flexible electronic devices such as large area, high conductivity, flexibility, and biocompatibility. Therefore, NMFSs have been developed using hybrid nanomaterials to exploit the advantages of nanomaterials of different dimensions. Sandwich structures [226] and core-shell structures [227] have been typically applied to develop flexible sensor devices. For instance, Xu et al developed a sandwich structure with two encapsulation layers and 3D conductive networks to prepare a green skin sensor with high sensitivity (2.54 kPa −1 ) and a wide detection range [228]. Notably, sandwich structures can be incorporated into not only contact but also non-contact flexible sensors. Shu et al developed a non-contact flexible magnetic thin-film sensor with a sandwich structure that exhibited ultra-high flexibility, fast response, and reliable stability (10 000 cycles). This sensor could identify the magnetic field direction and density based on non-contact magnetic bonding. Moreover, core-shell nanomaterials have been used to develop flexible sensor devices owing to their high biocompatibility, large area, excellent conductivity, and ion transport mechanisms. Kim et al prepared a low-resistance, flexible, and highly sensitive humidity sensor based on a chitosan and poly(amidoamine) core-shell structure (CNT@CPM) [229]. The optimized sensor exhibited an average response/recovery time of fewer than 20 s, high sensitivity, consistent responsiveness, stability (>15 000 cycles), high linearity, and low hysteresis (-0.29 to 0.30%RH). Self-powered smart sensors are typically prepared using a core-shell structure. Zhao et al prepared a core-shell composite using conductive metal-organic backbone (Cu-CAT)/carbon nanofiber networks (CNFNs) through electrospinning and hydrothermal methods. The Cu-CAT/CNFNs were used to prepare the sensitive layer of the flexible pressure sensor, which exhibited a wide sensing range (0.5-60 kPa), high sensitivity (30.40 kPa −1 ), excellent response/recovery time (0.24/0.31 s), and high repeatability [230]. Nanomaterials of different dimensions have different properties and show different advantages and disadvantages in the preparation of flexible sensors. 0D NPs have the advantages of excellent sensing properties, better biocompatibility, and tunable physicochemical properties; while they have the disadvantages of complex preparation process, high cost, and stability problems such as agglomeration and aggregation. Compared with 0D NPs, 1D nanomaterials, and 2D nanomaterials are highly flexible and stretchable, and can be tightly fitted to the skin to achieve better sensing and control effects. However, 1D and 2D nanomaterials are also subject to stability problems such as agglomeration and oxidation. In addition, because 2D nanomaterials are limited in some aspects by their 2D structure, their thermoelectric properties may not be as good as 0D and 1D nanomaterials, which would affect the performance of NMFSs. Therefore, when high sensitivity and biocompatibility are required in VR devices, 0D nanomaterial sensors are generally chosen. If the device requires high electrical conductivity, 1D or 2D nanomaterial sensors are chosen. And in VR devices where thermoelectric properties are being explored, 0D or 1D nanomaterial sensors are preferred.

Skin-mechanics-triggered interface
The metaverse, as a template for the development of the nextgeneration Internet, is aimed at establishing a hyperspace, a fully immersive shared virtual space to provide a convenient place to live, play, work, and socialize. Skin electronic sensing is currently the main interaction method for developing VR devices [231]. VR technologies have emerged as the main technological tool for people to access the metaverse. NMFSs with remote haptic interfaces of nanomaterials for bilateral interactions represent the future paradigm to facilitate access to the metaverse. Through wireless transmission and remote haptic reception, haptic-based NMFS technologies can transcend the realm of traditional VR interactions to enable virtual human-to-human interactions [232]. Sun et al prepared a ring with multimodal sensing that could enhance tactile perception and haptic feedback ( figure 8(a)). A customized low-power wireless platform directly drove all the ring functionalities. The haptic sensor enabled the detection of finger movements with high resolution and enhanced the performance of gesture/object recognition through ML analysis. Feedback functions and modal sensing were combined to implement an interactive metaverse platform that could travel through space to trigger haptics, enabling immersive face-to-face virtual social experiences [233]. Therefore, haptic interfaces with feedback functions and accurate sensing must be developed. Soft materials are one of the most prominent nanomaterials for realizing scalable electronic devices and haptic interaction. Biswas et al proposed a new fabrication method for highly integrated wearable health monitoring devices that achieves the integration of complex stretchable circuit primitives on a thin polymer substrate. The promise of soft electronics for wearable health monitoring applications is demonstrated ( figure 8(b)) [234]. Inspired by the formation and hierarchical structure of human fingertips, Qiao et al prepared frictional electrodynamic pressure sensing and vibrotactile texture recognition. The artificial haptic sensory system was successfully applied to soft robotic skin, mimicking the simultaneous fast and slow adaptive multimodal perception of the finger during grasping movements. (figure 8(c)) [235].
In general, the human perceptual system can realize sensing with spatial and temporal resolutions on the millimeter and sub-millisecond scales, respectively. Consequently, touch interaction interfaces are considered key to human-computer interaction and the metaverse. Xu et al designed a wearable textile touchpad with high-resolution touch sensing. This textile-based touchpad is biocompatible with human skin. It is able to achieve handwriting interaction with good mechanical capability (114 MPa), which is nearly 4145 times higher  [236]. Copyright 2023 American Chemical Society). (e) Communication between speech-impaired and non-signing people in VR using triboelectric smart gloves (Reproduced from [238]. CC BY 4.0). than pure hydrogel. Furthermore, the concept of using textilebased ionic touch panels for handwriting interaction has been demonstrated (figure 8(d)) [236]. Tactile perception can convey semantic and emotional information in social interactions or HCI [237]. Haptics plays a crucial role in communication between people affected by a true absence of speech, with tactile sign language being the touch communication method used by many people. Wen et al demonstrated an AI-based sign language recognition and communication system consisting of a sensing glove, virtual interface, and deep learning module [238]. The proposed system could effectively recognize words and sentences. Through new word reorganization, unfamiliar sentences and words could be recognized with an average correct rate of 86.67%. The sign language recognition results were mapped to a virtual space and translated into audio and text, enabling bidirectional communication between signers and non-signers ( figure 8(e)). Although the skin-mechanics-triggered interface of VR devices can simulate realistic tactile feedback, so that users feel the real sense of touch, thereby improving the sense of immersion and realism.
However, it also has some disadvantages. For example, reliability problems: skin-mechanical sensors can be affected by various disturbing factors such as temperature, humidity, etc, thus affecting their reliability and accuracy; usage limitations: since skin-mechanical sensors need to be in contact with the user's body, there are many limitations in terms of cleaning and maintenance.

Temperature-triggered interface
Thermal sensation occupies an equally important role in the human perceptual system as visual and tactile sensations. The temperature sensory system on the skin provides a wealth of physical information regarding objects and the environment. Thermal perception can often achieve results that cannot be obtained using auditory and visual perception. Specifically, objects with different thermal properties can be distinguished without any visual or auditory cues. Therefore, thermal sensation is a research hotspot for building realistic virtual environments. The degree of perception of flexible sensors can be improved through thermal or temperature sensing to enhance the perception and recognition of objects [20]. The existing research on thermotactic sensors and wearable e-skins has promoted the development of thermally inspired flexible sensors for use in the metaverse [19]. Oh et al developed a multimodal functional smart feedback glove with a heater sheet [46] that provided accurate and rapid thermal haptics even when stretched. Specifically, the smart glove with a heater sheet enhanced the contact state felt by the user and confirmed that materials with different temperatures can be differentiated in VR environments ( figure 9(a)). Lee et al proposed a highly flexible, stretchable, skin-like thermal haptic sensor that could be used for virtual world interactions [239]. The sensor could actively heat or cool a deformed skin surface, and when combined with ML algorithms, it could adjust the temperature in a timely and accurate manner to mimic the ideal heat sensation of human skin. The thermal haptic sensor was integrated with smart gloves to provide human thermal sensory information to the skin in various situations and demonstrated that hot coffee cups and cold beer bottles could be sensed in virtual spaces ( figure 9(b)).
Thermally inspired sensor interfaces can be used to perceive temperature information in the metaverse, allowing users to feel the temperature in the virtual world while enriching the senses and increasing the realism of experiences in the virtual environment. Shin et al developed an integrated laser-induced reduction sintering technique and a unique monolithic structure to achieve an artificial skin based on a negative temperature coefficient, with a highly sensitive thermistor (figure 9(c)) [240]. Choi et al regulated finger friction by changing the local temperature. The friction force on the finger surface was increased by approximately 50% when the surface temperature of the object increased within a certain range [241]. The virtual features were rendered by surface temperature modulation in the absence of thermal perception ( figure 9(d)). The method of modulating finger friction by temperature can be applied in virtual spaces and touchscreen HCI interfaces. The thermal haptic feedback from a nichrome wire heater allowed users to grab hot (to varying degrees) coffee in the virtual space (figure 9(e)) [233]. The temperature-triggered interface of the VR system can simulate real-world thermal feedback, allowing users to feel real heat in the virtual world and improve awareness of ambient temperature. However, it can also be perturbed by the ambient temperature, thus affecting the accurate temperature feedback, and is generally suitable for temperature sensing and monitoring in small areas.

Magnetism-triggered interface
Magnetism-based NFMS can be used to construct a new sensory dimension to allow humans to use the surrounding magnetic field as a stimulus for non-contact perceptual experiences. This non-contact sensing is similar to wireless devices but does not require complex hardware and internal programs, relying only on magnetic fields to achieve noncontact sensing. Magnetism-triggered interfaces and skinmechanical interfaces are two different types of sensors with different operating principles and measurement parameters. Magnetism-triggered interfaces are sensors that can detect changes in magnetic fields, typically using magneto resistors or the Hall effect to detect changes in magnetic fields [242]. They are typically used to measure the strength, direction, and rate of change of the magnetic field [198]. Skin-mechanical interfaces are sensors that measure external forces, detecting stress and pressures from humans, robots, and other objects [232]. Skin-mechanical interfaces are sensors that typically use measurement principles such as resistance, capacitance, piezoelectricity, or piezoresistive effects [243]. As a result, the magnetism-triggered interface and skin-mechanical interface differ significantly in terms of measurement parameters, operating principles, and applications. Magnetic sensors are mainly used to detect parameters such as the direction, position, velocity, and acceleration of an object, while skinforce sensors are mainly used to detect forces and pressures of objects in contact with the human body. In VR devices, both sensors are typically used in different application scenarios to provide a more accurate, realistic, and natural interaction experience.
These NMFSs based on magnetic induction provide novel opportunities for human perception and may function as alternative channels for acquiring information beyond the five senses [244]. The contact and non-contact perception capabilities of e-skin can be used to experience virtual worlds to manipulate virtual objects. Ge et al built a flexible magnetic microelectromechanical system of a bifunctional electronic skin capable of transmitting both tactile and non-tactile sensations through pressure and magnetic field stimulation [245]. Combination with virtual data (virtual knobs) can facilitate interactions with a physical object ( figure 10(a)). Cañón et al developed a flexible and mechanically stable e-skin compass system [198]. The recognition, remembering, and learning of stimuli that humans cannot perceive is a novel research area in interactive smart electronics. Kim et al proposed an AI-assisted magnetoreception synaptic system based on the magnetic cognitive ability of birds for navigation and localization, combined with an array of ferroelectric field-effect transistors [44]. This system can be used as a navigation compass for ultra-sensory synaptic mimicry to facilitate automatic navigation and mapping of moving objects ( figure 10(b)). Shu et al demonstrated a flexible magnetic sensor based on a sandwich structured film (SSF) that can detect tensile and bending deformation through feedback from resistance changes. In addition, the ultra-sensitivity and stability of the SSF sensor allow the direction and density of the magnetic field to be clearly identified, enabling the development of a magnetic keyboard for contactless interaction (figure 10(c)) [246]. Makushko et al demonstrated a non-contact interactive magnetic-sensing interface for contactless interactive electronics on the skin based on flexible all-metal Co/Pd-based spin valves that were sensitive to out-of-plane magnetic fields [29]. The flexible switch was integrated with the interactive electronics on the skin. The switch functioned as a contactless HCI and it could turn on and off the navigation software on the virtual display ( figure 10(d)). The most important feature of VR devices based on a magnetism-triggered interface is their low energy consumption, as they rely on magnetic fields to enable low-power measurements. It can also be easily integrated into VR systems without the need for additional hardware or equipment. However, magnetic sensors require a certain amount of space to achieve measurement and tracking, so space constraints must be considered in the design of VR devices, and it is also susceptible to metal interference.

Neuro-triggered interface
The human body has many tactile receptors (mechanoreceptors and thermoreceptors) located in the human skin. Under the control of the nervous system, we rely on these receptors to perceive the real world [247]. Mimicking neural structures, Lee et al have developed an asynchronously encoded flexible sensor that maintained an exceptionally low readout latency while transmitting thermal tactile information, even though it involved an array of over 10 000 sensors [32]. The archetypal array of this sensor involved up to 240 artificial receptors and transmitted events with a 1 ms delay while exhibiting an ultra-high temporal accuracy (<60 ns). Consequently, the proposed sensor could satisfy the fine timing requirements needed for tactile sensing ( figure 11(a)). Additionally, the sensor required only one conductor for propagating the signal, enabling dynamic reconfigurability of the sensor array. Human neurons exist in a 3D form. The construction of a 3D networked electronic system that mimics the structure of neuronal synaptic networks can usher in the next era in neuronal computing. Wu et al developed flexible 3D artificial synaptic networks based on neuronal prominence network structures, in which mnemonic devices at both ends functioned as electronic synapses (e-synapses) [248]. The main features of e-synapses were similar to those of biological synapses and 3D artificial e-synapses could simulate learning and trainable memory. Capabilities with a large tolerance to input errors and alterations. Such synapses could provide a tractable physical platform for implementing the next step in layered neural networks for complex computations ( figure 11(b)).
In general, the substantial growth in information has necessitated high data processing speeds. Such speeds can potentially be realized by extending the neuromorphic paradigm to process information data. Wang et al developed a neuromorphic sensory system with gas sensing, data storage, and processing capabilities. A Pt/Ag/TaOx/Pt-based amnestic device transmitted signals from sensory neurons to relay neurons. Subsequently, the relay neurons processed the signals from the synapses and performed gas classification [31]. The proposed system could simulate biosensing when integrated with different sensors ( figure 11(c)). Kim et al developed an organic e-skin to mimic human sensory nerve functions. The artificial input neural system captured pressure signals (1-80 kPa) from a cluster of pressure sensors, converted the pressure signals to action potentials (0-100 Hz) using ring oscillators, and integrated the action potentials from multiple ring oscillators via transistors. The biomimetic hierarchical structure could detect the motion of objects and distinguish braille characters [249]. The input-based nervous system was connected with motor nerves to construct a hybrid bio-organic e-skin reflex arc to drive muscles ( figure 11(d)). The interface of sensory neurons in the intradermal somatosensory system helps tissues process complex perceptual information. Wan et al demonstrated an artificial perceptual neuron organization structure that could recognize the spatiotemporal features of processing tactile patterns, including sensing, transmission, and processing components [250]. When ML algorithms were incorporated into the system, the perceptual recognition error rate could be decreased from 44% to 0.4%. The development of e-skin with AI neuromorphic forms can facilitate the design of the neuronal structure of human tiredness ( figure 11(e)). VR devices with the neuro-triggered interface can achieve a more natural human-computer interaction experience, allowing users to control VR devices without moving their bodies, reducing motion fatigue and making it easier for users to immerse themselves in the VR world. However, sensors based on neural-triggered interfaces are complex and costly to prepare and are also susceptible to human mental states.

ML-assist metaverse/VR application
ML and NMFSs are the two fundamental elements in the construction of the metaverse. In recent years, ML has attracted considerable attention in the VR domain because of its powerful data analysis abilities [251]. To enhance the usefulness of data and effectively integrate data with the metaverse, ML is being widely used to clarify the associations and differences between multiple datasets. The development of appropriate ML algorithms [252], and big data analysis tools can facilitate real-time interactions in VR worlds [253]. Zhu et al developed an ML-based, sustainable, self-powered HCI that can be applied to enhance human finger movements and virtual activities [254]. Using the PCA/t-SNE + GMM/Kmeans algorithm and output data, the activity trajectory of the target object could be accurately tracked and reproduced with high visualization performance. The decoupling algorithms can not only classify the velocities of finger movements but also enhance motion patterns and virtual activities through frictional electric and self-powered HCI flexible sensors ( figure 12(a)). Using the PyTorch model and scikitlearn algorithms, Yang et al developed a multimodal wetelectric self-powered sensor composed of GO to monitor pressure, light, temperature, and humidity in real-time with unique self-powered functions [255]. The proposed sensor was used to construct an HCI wristband to monitor the human body temperature, pulse, and sweating conditions in multiple dimensions and facilitate communication in sign language. The proposed sensor device combined with the ML model helped develops a continuously self-powered multimodal e-skin for VR applications ( figure 12(b)).
At present, the metaverse must be accessed through HCI devices. Data gloves, as primary HCI devices, are often used in conjunction with ML to enable the manipulation of objects in virtual spaces [235,256]. Data gloves developed based on the principle of grasping objects with human fingers are expected to compensate for vision-based VR systems [257]. Meier et al incorporated neural networks to enable fast touch interactions in VR interfaces. A wrist-based sensing system, TapID, was proposed, which could complement the hand pose of helmet VR devices to acquire inputs. This neural network classifier included five convolutional layer blocks and two linear layers. User studies demonstrated the high detection accuracy (99.7%) and gesture recognition finger accuracy (93%) of TapID [258]. In addition, surface gesture applications of soft modular gloves in VR are demonstrated ( figure 12(c)). Deeplearned soft electronics can be combined to capture dynamic motion at a distance to obtain comprehensive measurements of the human body [259]. Yang et al developed a flexible Ti 3 C 2 Tx MXene with a built-in ML model to obtain a high sensitivity (GF: >1000) sensor [260]. With edge data processing  [266]. Copyright 2022 American Chemical Society). (d) The edge-modularized MXene sensors enable whole-body motion reconstruction of the virtual avatar (Reproduced from [260]. CC BY 4.0). (e) Demonstration of deep learning-driven frictional electric smart sensor socks in virtual space-physical space interaction applications (Reproduced from [262]. CC BY 4.0). (f) s-SNE algorithm-enabled triboelectric nano-generator sensor visualization process of the structure, results, and automatic grasping object recognition accuracy (Reproduced from [267]. CC BY 4.0). (g) Deep learning-driven smart mats combined with virtual technology as a floor monitoring system (Reproduced from [265]. CC BY 4.0). and built-in ML, the sensor-enabled full-body motion detection, data information classification, and character reconstruction. Specifically, the sensor-enabled high-precision virtual character animation reconstruction could successfully imitate continuous whole-body motion ( figure 12(d)). Notably, advancements in ML and the Internet of Things have raised concerns regarding data information security at both the individual and country levels. Traditional security measures typically provide only limited protection [261]. Zhang et al developed textile-based friction and electricity-driven smart socks with integrated deep learning that could harvest energy from body movements to transmit data [262]. This wearable sensor could transmit information regarding a user's identity, health status, and activity, thereby protecting the user's information. Using an optimized 1D CNN-based deep learning model, the recognition accuracy was 93.54% for 13 participants and 96.67% for five human activities.
In practical applications, the physical signals collected by the socks can be fed back to the virtual space, enabling a VR system for game engines, human activity monitoring, and healthcare to be built ( figure 12(e)). The algorithmic fusion of multi-sensor datasets is another computational approach for the application of a hierarchical support vector ML algorithm to intelligent systems [263]; however, this approach is limited by the low sensor-data quality and dataset incompatibility. Wang et al reported a bio-inspired data fusion architecture for gesture recognition by fusing somatosensory data from SWNT-based strain sensors with visual data [264]. The architecture used convolutional neural networks for visual data information processing and then implemented sparse neural networks for data fusion and feature recognition from different sensors. Additionally, this architecture achieved a 100% recognition rate even under image interference and underexposure or overexposure conditions ( figure 12(f)). Shi et al demonstrated a self-powered floor mat based on frictional electricity and an intelligent monitoring system integrated with ML [265]. The location of people on the floor mat, their behavior, and their identity information could be determined based on real-time monitoring data ( figure 12(g)). This intelligent monitoring system based on floor mats could lay the foundation for the next generation of automated furniture and security applications. ML-assisted VR systems can achieve more natural and intelligent human-computer interaction through natural language processing, emotion recognition, and other technologies to improve the user's feeling of the virtual experience. However, AI requires a large amount of computational resources and algorithm support, which increases the cost and difficulty of using the device. At the same time, AI still has many challenges in privacy and security and algorithm reliability.

Conclusions
This review summarizes the recent advancements and innovations in NMFSs for metaverse or VR applications. First, we describe the flexible nanomaterials with different dimensions and highlight the application of NMFSs with varying interfaces of triggers in VR. Notably, 0D-2D nanomaterials can be used for flexible sensor development. Different human sensory feedback modalities can be applied to virtual technologies for novel nanomaterial development and the realization of real-time and long-term interactions with the metaverse environment. The properties of nanomaterials with different dimensions are summarized to provide a reference for the preparation of novel nanomaterials with varying application requirements (figures 13(a)-(c)). Nanotechnologies have been widely used to develop nanomaterials with excellent compatibility to mimic human skin. These bionic, highperformance flexible sensors can sense various environmental factors to optimize human comfort and other requirements. Additionally, timely and accurate sensory feedback must be provided to close the loop of sensor systems. Multimodal and multifunctional NMFSs have been developed to improve the accuracy of sensors. NMFSs can overcome the dependence of traditional VR on heavy equipment. Metaverse technology requires people to be able to interact with physical spaces in the virtual world in real time. We believe that natural and realistic perceptual feedback interfaces and tactile reproduction are necessary to promote human interaction with the metaverse.
Subsequently, we reviewed four perception-triggered nanomaterial-based interfaces for VR applications, each with unique advantages and disadvantages [268,269]. Skinmechanics-triggered interfaces are one of the most widely used haptic technologies in VR and are the most rapidly developing HCI platforms ( figure 13(d)). Their immersive interactive experience, multimodal perceptual feedback, and high biocompatibility render them useful for various medical detection, AI, and metaverse space applications [232]. For the development of the metaverse, it is necessary to establish more convenient energy supply devices and non-contact triggering methods. Temperature-triggered interfaces (figure 13(e)) can sense the temperature of the material and environment and enhance the realistic experience of the metaverse and are compatible with contact and non-contact modes [19]. For example, unique environments (i.e. extreme heat or cold) can be simulated in virtual environments for military or special personnel training [20]. However, these interfaces are characterized by high perceptual lag time, insufficient cooling capacity, and low performance of thermoelectric materials. Magnetic sensor devices as an electronic sixth sense, represent a novel perception channel to obtain information (figure 13(f)). Because of the Earth's magnetic field distribution, magnetically sensitive sensors have natural advantages in recognizing direction, determining geographic location, 3D motion tracking, and VR interaction [244,246]. Moreover, the presence of the geomagnetic field is a key factor that makes magnetically sensitive sensing devices susceptible to interference. This problem must be addressed before developing magnetosensitive sensors, for example, by constructing independent magnetic field sensing systems. Neuro-triggered interfaces are the most complex among emerging perception technologies. These interfaces require not only proficiency in electrical transmission principles but also a profound understanding of the network structure of human neurosynapses [43]. Nevertheless, the neural-triggered interfaces provide the most realistic and natural haptic perceptual interaction as well as the most secure privacy information protection and thus have significant potential for metaverse exploration ( figure 13(g)). For the application of neuro-triggered interfaces in bionic neural systems, it is necessary to improve the stability, digital analysis computational capability, and security [247], The resulting systems can facilitate medical treatment, human virtual activity interaction, and full integration of the physical space and metaverse. Figure 13 summarizes the advantages and limitations of different types of nanomaterials and the four triggered interface technologies. Advanced nanomaterials are indispensable for developing sensory technologies. The development of highly moldable and biocompatible nanomaterials has shifted human perception technology from the traditional reliance on actuators to bionic electronic skin interfaces. The emergence of novel nanomaterials and flexible sensors is expected to provide innovative channels and vehicles for human perception acquisition. In terms of virtual health [13,270], NMFSs represent portable and comfortable health monitoring and diagnostic tools that can be worn over the long term to detect human health conditions and measure human functional qualities in human interaction with the metaverse to prevent diseases. NMFSs are expected to contribute to the breakthrough of perception technologies in the metaverse.  [271]. Copyright 221 American Chemical Society). (c) Summary of 2D nanomaterial characteristics (Reproduced from [272] with permission from the Royal Society of Chemistry). (d) Skin mechanics-triggered Interface (Reprinted from [273], Copyright 2019, with permission from Elsevier). (e) Temperature-triggered Interface (Reproduced from [274]. CC BY 4.0). (f) Magnetism-triggered Interface (Reprinted from [275], Copyright 2022, with permission from Elsevier). (g) Neuro-Triggered Interface (Reprinted with permission from [276]. Copyright 2020 American Chemical Society).

Outlook
NMFSs are an important part of VR applications and can capture information such as the user's movement, position, pose, and gestures to provide a more immersive experience. In the future, the development of VR sensing can lead to changes and advances in higher accuracy and lower latency [270]: more advanced technologies such as ML and deep learning for higher accuracy and lower latency. This will improve the user experience and naturalness of interaction. Smaller size and weight: smaller and lighter for users to wear. This will improve the wearability of VR and make it easier for users to enjoy immersive experiences. More sensor types: integrate more types of sensors, such as brain waves, emotions, smell, and breathing sensors, to enable more sophisticated physiological health monitoring and HMI. More application scenarios: broader use in healthcare, industry, and education to improve efficiency and reduce risk. Enhanced security and privacy: strengthen data security and privacy protection to prevent misuse or leakage of user data. In a nutshell, the future development of VR sensing will make the VR experience more realistic, immersive, and natural, providing users with a genuine experience of the living environment and working atmosphere.