Accessorizing Quadrupedal Robots with Wearable Electronics

Robots epitomize one of the most popular human imaginations that are evolving to become reality in the most pervasive way. Especially with the rise of artificial intelligence, it is more hopeful than ever that robots one day will be the interactive companions with some degree of emotion. A variety of robots in different forms with considerable mobility are available today. Among them, legged robots are the most attractive due to their dexterity, traversal versatility, functional modularity, and locomotive stability. Legged robots like robotic dogs can survey unchartered territories in uncertain environments. In this comprehensive review article, the status quo of this emerging and exciting field and how electronics of various form factors can functionalize them in an unbiased manner are surveyed and discussed. Different accessories currently available for quadrupedal robots, wearable devices potentially integratable with these robots, data management/integration strategies for reliable interfacing, and future outlook are discussed.


Introduction
Since their first appearance in 1954, robots have become truly ubiquitous.From everyday products (e.g., autonomous cleaning robots and Furby robot toys) to application-specific robots (e.g., unmanned aerial vehicles, self-driving cars, and legged robots), robots are still on the rise in various sectors of human society.Among them, legged robots such as bipedal and quadrupedal robots are the most promising candidates.In addition to their commercial availability, [1][2][3][4] legged robots like quadrupedal robots can survey unchartered territories in uncertain environments like mountainous terrains, sub-Saharan areas, Arctic zones, underwater, the Moon, and Mars (see Table 1).This implies they need to have a robust but flexible mechanical structure.It is also critically important to functionalize these robots by interfacing them with additional electronics, preferably wearable electronics for facile integration.To take advantage of the capabilities of robots with wearables, it is crucial to have equally advanced hardware and software to collect, process, and ultimately respond to the input data.Moreover, as the number of sensors tends toward hundreds if not thousands and expands their breadth, it is crucial to have access to middleware capable of generating digital data that can be stored and processed by increasingly intelligent control algorithms.In this regard, there are practical aspects to be considered: physical conformity to curvilinear surfaces on robots, integration strategies, functionalities, and metrics to assess the devices.To answer these, this comprehensive review article surveys and discusses the status quo of this emerging and exciting field in an unbiased manner.
The article is structured as follows.First, we discuss different accessories with conventional form factors (i.e., payload and implanted devices).Wearable technologies currently available for the robots are then introduced, followed by a discussion of potential wearable technologies that can be integrated with the robots.Then, different data management strategies for wearable technologies are reviewed, and a future outlook is presented.

Recent Progress in Conventional Accessories for Quadrupedal Robots
With the recent progress in locomotion control of quadrupedal robots, there have been efforts to accessorize the robots with sensors to improve their locomotion by expanding their environment perception [5][6][7][8] and to acquire mission-specific data. [5,9,10]ctuators can also be integrated with robots to handle objects and provide sensory output.Furthermore, it is possible to incorporate energy harvesters on the surface of a quadrupedal robot to scavenge the ambient energy around it.Such devices-sensors, actuators, and energy harvesters-can be categorized into two groups: conventional accessories and wearables.Conventional accessories refer to devices, which are physically rigid and require to be firmly mounted or permanently integrated with the robot due to the large size and weight and the lack of physical conformity.They can be subcategorized into two subgroups based on the way they are equipped by quadrupedal robots: payloads and implants.Payloads represent the group of devices that are mounted on top of a quadrupedal robot's torso.[3][4] On the other hand, implants are designed to replace certain parts of a robot (e.g., foot) for functionalization, which makes them permanent modifications to the robot.As opposed to the conventional accessories, wearables are lightweight and physically flexible, so that they can be attached to any surface of the robot, without the need for a mounting rail or becoming a permanent modification.This makes the integration of wearable devices far easier than the conventional counterparts and simultaneously functionalizes the attached surface.In this section, we survey and discuss the recent developments in the conventional accessories-payloads and implants-and their applications.

Payload
Payload devices are well-established accessories that are widely used in commercial [1][2][3][4] and research settings [11][12][13] to enhance the environmental perception of quadrupedal robots.[16] A notable example of a payload sensor is the light detection and ranging (LiDAR) sensor, which has a transmitter and a receiver. [17]e transmitter modulates a light wave in its intensity, phase, or frequency and emits it to the surrounding environment, and the receiver detects the reflected wave and measures the time-of-flight delay to calculate the distance between the LiDAR and the surrounding objects. [18]ue to the shorter wavelength used, LiDAR possesses superior spatial resolution, which makes them favorable sensory platforms for simultaneous localization and mapping (SLAM). [5,9,10,15,17,18]In the DARPA subterranean challenge, different groups implemented SLAM with the aid of LiDARs (see Figure 1a). [10,19]The robots could accurately map the surroundings (e.g., walls, pillars, rocks, ground profile) and hence traverse safely in the dark environment.In addition to the detection of rigid objects, LiDARs can be used to map nonrigid obstacles such as grass, snow, and other overhanging obstacles. [20]Detection of deformable obstacles is especially useful for traversing in mountains, forests, and cryospheres, which can facilitate and hence allow the robots to help humans survey such environments.Detection of deformable obstacles can improve a robot's ability to plan its path and thus facilitate its traversals in mountains, forests, and arctic environments.This allows the robots to assist humans with surveying such uneven terrains.
Furthermore, quadrupedal robots can be equipped with payload actuators such as robotic arms.Such actuators can help quadrupedal robots handle objects or obstacles that might obstruct their way.It is also possible to mount other sensors such as microphones, cameras, and gas sensors on a robotic arm to enable the manipulation of these sensors and thus obtain data of higher quality (see Figure 1b). [5]Gehring et al. in addition to a LiDAR and depth camera, utilized a sensor platform, consisting of an actuated gimbal, visual, and thermal camera and a flashlight, for autonomous inspection of an offshore high voltage direct current platform as seen. [9]s evident from the examples, a payload device's strength comes from its modularity and robust sensing/actuating capabilities.However, their heavy and bulky nature is the limiting factor.3][4] While high-end models can afford to load multiple payload devices, consumer-level robots can only accommodate one or two light payload devices weighing about 2 kg each.Heavily loading a robot also increases its power consumption and therefore reduces its runtime.In addition, payload devices require a mounting rail to be attached to a quadrupedal robot.Incompatible designs of the mounting rail of the robot and the payload attachment point can hinder the universal integration of payloads with commercially available quadrupedal robots.Furthermore, mounting a payload device on top of a robot increases its total height and thus may prohibit it from entering small gaps or openings, which are common in rescue missions.

Implants
As described earlier, implants refer to devices that are tightly integrated as a part of a quadrupedal robot.Study shows that animals heavily rely on the tactile sensing of the ground during locomotion. [21]24] This can be achieved by modifying the feet of a quadrupedal robot with implantable tactile sensors for tight integration. [16,20,25,26]urthermore, implantable tactile sensors can inform quadrupedal robots of the ground condition, allowing them to calculate important quantities such as the center of mass and zero-moment point. [7]s illustrated in Figure 1c, Aoyagi et al. developed an arrayed capacitive tactile sensor, which was then integrated with the sole of a legged robot for real-time 3D force detection. [27]The sensor consists of two rubber hemispheres, between which there is an array of capacitive sensing elements, whose capacitance value changes with applied pressure.The capacitance change can be sensed by the peripheral circuit.Data acquired by the sensing elements are decoded by a neural network to retrieve the 3D force vectors.Such tactile sensing can inform the robot of the terrain, which acts as an additional set of proprioceptive input to improve its gait.
Similarly, Käslin et al. fabricated an adaptive tactile sensor with an ankle joint and integrated it with the feet of a quadrupedal (see Figure 1d). [25]The sensor consists of a six-axis force/ torque sensor and two inertial measurement units (IMUs).The force/torque sensor measures the contact force, and the IMUs determine the ground inclination by measuring the ankle orientation.The acquired contact force and ground information can be interpreted by a quadrupedal robot to improve locomotion.In another study, Kolvenbach et al. applied the sensors to a robot to support unmanned concrete inspection in a sewer. [28]Not only do the sensors enhance locomotion on slippery grounds by measuring the floor texture, but also they can detect cracks and measure surface roughness, enabling the evaluation of the concrete condition in the sewer.To utilize the sensors for concrete inspection, the authors developed a scratching motion, similar to how human inspectors would touch concrete floors and walls to gauge their conditions and textures.
In addition to sensory implants, it is possible to accessorize the robot with implants to augment its mechanical stability.Huang et al. explored the importance of having 3D flexibility using a relatively simple and practical leg structure. [29]nspired by biomimetics, the authors proposed a novel approach that maximizes both adaptability and speed by combining flexibility and rigidity (see Figure 1e).The structure's design uses flexible airbags and rigid bones.In Figure 1c, the design of the leg structure is shown.Implementing compressible airbags and carbon fiber rods significantly reduces the weight of the entire system.The structure can modify the leg length and stiffness by changing the position of the thrust ring.Aside from its practical leg design, the robot assumes the inner knee elbow structure for both sets of legs as illustrated in Figure 1d.This approach minimizes sliding between the ground and the robot's foot, enhancing stability.
Implants can provide several benefits ranging from ground sensing to stabilization.Their tight integration with the robot ameliorates their long-term reliability as well.However, the compact structure often demands a complex integration process and custom-made parts, because of which implants are typically left as permanent modifications to a robot.In the event of device failure, dissembling the integrated parts is necessary, which complicates the repair process.In addition, the model-specific design of an implant compromises its universal integration with different models of quadrupedal robots.

Wearable Electronics
Wearable electronics, in the context of quadrupedal robots, are defined as flexible devices that are lightweight and physically flexible, allowing them to be attached to any surface of a robot for functionalization, whether it is for sensing, actuating, or energy harvesting.Traditionally, wearable electronics are perceived as an augmentation to humans.Especially, in the field of healthcare, wearable electronics are indispensable due to their physical conformity and valuable sensing capabilities that inform us of various biomarkers from biofluid (e.g., blood and sweat), [30][31][32][33][34] brain, [35,36] skin, [37][38][39][40] and heart . [41,42]Beyond human applications, wearable electronics can enable facile functionalization of various robotic systems such as quadrupedal robots.While conventional add-ons would require disassembly or sometimes a complete redesigning of the host system to install, wearable electronics can simply be applied to the surface of the host system with relative ease.For instance, rather than searching for a manufacturer who has designed a robot specifically for underwater search and rescue, one can leverage their ingenuity to combine waterproof wearable sensors and a commercially available robot with underwater capabilities.This brings us to the discussion of different types of wearable electronics.
Physically flexible sensors play a pivotal role in wearable electronics, enabling the integration of sensing capabilities into everyday objects, garments, and accessories.These sensors provide real-time monitoring of various physiological and environmental parameters. [41,43,44]They typically utilize conductive polymers or polymeric composites, carbon-based materials, nanomaterials, or organic materials that exhibit excellent mechanical flexibility and electrical conductivity. [45]owever, several practical considerations need to be addressed for effective wearable technologies with high performance.These include the reduction in size, weight, power consumption, and manufacturing cost.Thus, wearable electronics must have the following components: an efficient power management system, data management (e.g., on-system memory and/or communication capabilities), sensors, and actuators.Additionally, reliability is crucial; wearable electronics must exhibit long-term robustness, physical conformity, mechanical stability, and environmental adaptability.Failure to achieve these goals will compromise their functionalities in the long term and necessitate frequent repair or replacement.Therefore, in this regard, we compile the following device metrics for wearable electronics: performance, minimum bending radius, maximum number of bending cycles, stretchability, device dimensions, and integration complexity.Performance refers to the sensitivity and output power density of sensors and energy harvesters, respectively.The minimum bending radius indicates a wearable device's flexibility, and the maximum number of bending cycles provides insights into its long-term mechanical reliability.Stretchability shows how stretchable a wearable device is, which is a beneficial metric to look at if one decides to attach a wearable device to, for example, a robot's leg joint.In the following subsections, we discuss several categories of wearable technologies for quadrupedal robots while highlighting these practical aspects.The first subsection focuses on the current wearable devices for quadrupedal robots, and in the second subsection, we survey different wearable technologies that can be potentially applied to the robots to augment their abilities.

Tactile Sensors for Quadrupedal Robots
Tactile sensing films are based on various mechanisms-including piezoresistivity, [7,[46][47][48][49] triboelectricity, [7] piezocapacitance, [50][51][52] and triboelectricity [7,53,54] -to accurately map contact force and hence provide insights into force distribution and surface texture.Piezoresistivity refers to the change in the electrical resistivity of a material in response to a mechanical strain.Piezoelectricity is the ability of a material to generate electrical charge when deformed.Piezocapacitance is defined as the change in capacitance in response to applied pressure.Triboelectricity refers to the generation of electrical charge in the event of friction between two dissimilar materials.Tactile sensing has applications in robotics, human-computer interaction, and material testing. [55]In the field of robotics, this provides robots with a sense of touch, which can enable collision detection and improve stabilization.
Wu et al. (see Figure 2a) developed a flexible piezocapacitive tactile sensor that can be attached to the soles of a small-legged robot to improve its gait control. [56]The sensor consists of an array of piezocapacitive tactile sensing elements, each of which measures the normal and shear force on the sole of the robot.When pressure is applied on each capacitive sensing element, the silicone island layer, which acts as the dielectric, gets compressed.This changes the effective gap between the two electrodes of the sensing element and hence induces a change in capacitance.The measured capacitance data were then fed to a support vector machine to classify the terrain type with 82.5% accuracy.Based on the classified terrain type, the legged robot adapted different gait configurations, which improved its speed by 17.1%.Xu et al. (see Figure 2b) developed a flexible multimodal sole sensor (FMSS) to enhance locomotion on challenging terrains. [7]The FMSS incorporates a combination of piezoelectric, piezoresistive, and triboelectric sensors.The piezoresistive element consists of a conductive porous layer and interdigitated electrodes.The conductive porous layer was prepared by mixing sodium chloride (NaCl), thermoplastic polyurethane, and chlorobenzene in N,N-Dimethylformamide (DMF) solvent, thermally curing the mixture and dissolving all NaCl particles in water to obtain the porous structure.The interdigitated electrodes were patterned on a flexible printed circuit board (FPCB), and the two structures were bonded together.The piezoelectric sensor was fabricated by spin-coating and thermally curing a P(VDF-TrEE)/DMF solution and then a conductive silver slurry on a prepatterned FPCB.The sensor was then encapsulated with polydimethylsiloxane (PDMS).Finally, the triboelectric sensor consists of a conductive cloth electrode layer and an ethylene-vinyl acetate foam acting as the triboelectric layer.The final structure is shown in Figure 2c.
In the piezoresistive sensor, the conductive porous layer's effective resistance can be modulated with applied pressure as the pores of the layer will be compressed and the total internal contact surface area will increase.The interdigitated electrode design enables the sensing of pressure distribution of a robot's feet and hence terrain recognition.The piezoelectric sensor produces an electrical charge in response to a given pressure, which decays exponentially with time.Terrain recognition can be enhanced by the dynamic piezoelectric input, and the sensor can be used to gauge the hardness of the ground, which relies on dynamic force sensing.The triboelectric sensor generates an electrical charge upon a contact-release motion, which also decays quickly with time, making it a dynamic sensor.Along with the piezoelectric sensor, it can improve the hardness recognition function of the sensor.Furthermore, because the amount of charge generated by a contact-release motion depends on the contact area, it can also be used for ground roughness measurement.The parallel piezoelectric, piezoresistive and triboelectric inputs were decoded to classify the terrain type with a high sensitivity of 0.66 kPa À1 and a wide pressure detection range from 20 Pa to 800 kPa.Through experiments conducted on both human and quadrupedal robot feet, the researchers successfully mapped intricate characteristics of the external surface, including terrain features, texture, hardness, and slippage.
Due to the thin and flexible form factor, such devices can be easily integrated with the feet of a quadrupedal robot to collect valuable information about the terrain (e.g., texture, roughness, inclination, and hardness).As opposed to implantable sensors that are tailored to specific quadrupedal robot legs, fabrication and application of the flexible tactile sensor are simpler, reinforcing its wide applicability.However, as larger quadrupedal robots generally weigh more, an even wider pressure detection range and more reliable encapsulation are necessary to employ the device in larger models.These studies highlight the ability of wearable tactile sensors to be easily integrated with other sensors to enhance the traversal of complex or obscured terrain allowing systems to reach human levels of sophistication in traversal.

Motion Sensors for Quadrupedal Robots
In the realm of robotics, motion sensors are indispensable components of a robotic system for mechanical stabilization.However, traditional accelerometers are often rigid and pose challenges when it comes to integration within preexisting robotic systems.Furthermore, conventional rigid accelerometers suffer from material fatigue and aging caused by long-term cycling, which leads to irreversible mechanical failure. [57]urthermore, motion sensors are usually embedded in the torso or joints of a quadrupedal robot to allow monitoring of the orientation of a robot.This poses a challenge in integrating additional motion sensors with the robot, especially without access to the structural specifications of the robot.On the other hand, wearable motion sensors can be easily implemented in quadrupedal robots and can mitigate the long-term reliability issue by utilizing inherently soft and flexible materials.
Exploiting the well-established microelectromechanical systems (MEMS) fabrication techniques, Mahmood et al. developed a flexible MEMS-based capacitive accelerometer with a minimum bending radius of 1 cm. [58]The accelerometer is a capacitive resonant sensor, which, when driven at its resonant Wearable devices for quadrupedal robots.a) Flexible piezocapacitive sole sensor for small-legged robots.Reproduced with permission. [56]opyright 2020, Institute of Electrical and Electronics Engineers.b) Structure of the FMSS with annotation to highlight the materials responsible for different sensing mechanisms.c) Different sensing mechanisms-piezoresistivity, piezoelectricity, and triboelectricity-are incorporated in the FMSS and their use cases.Reproduced with permission. [7]Copyright 2021, Multidisciplinary Digital Publishing Institute.d) Left: illustration and applications of the flexible inertial sensor with a soft graphene-coated eutectic gallium-indium (EGaIn) droplet proof mass.Right: scanning electron microscope (SEM) images of the soft proof mass and its surface.e) Working principle of the soft inertial sensor.Left: operational states without and with external acceleration.Right: variable resistance calculation depending on the proof mass position.Reproduced with permission. [8]Copyright 2022, American Chemical Society Publications.
frequency, changes its capacitance in response to an applied acceleration.The resonant frequency depends on the design and dimensions of the sensing element.The fabrication process with a relatively low thermal budget allowed the incorporation of polyimide as the encapsulation material.The double UV-LIGA process, which consists of lithography, electroplating, and molding, [59] boosts the device's peak sensitivity, ranging from 156 to 194 fFg À1 .Nevertheless, although nickel, whose Young's modulus is between 93 and 205 GPa depending on the electroplating condition, [60] was used as the proof mass because it is an inherently rigid material, the device will likely suffer from long-term usage.
This necessitates the use of a soft proof mass.Babatain et al. (see Figure 2d) developed a soft inertial sensor using laserinduced graphene (LIG) and graphene-coated liquid metal droplet (LMD), integrated into a wearable platform for healthcare monitoring and human-machine interfaces. [8]The wearable platform was fabricated using a scalable and rapid laser writing technique and incorporated a programmable system-on-a-chip to function as a standalone system for real-time wireless monitoring of movement patterns.The sensor has two tubes lined with LIG electrodes acting as flexible interconnects. [61]The electrodes, together with the LIG-coated LMD proof mass, act as a variable resistor.The proof mass position modulates the effective LIG interconnect length and hence varies the interconnect resistance, which can be used to back-calculate the proof mass position and hence the acceleration.The sensor exhibited a high sensitivity of 6.52% m À1 s 2 and excellent repeatability of more than 12 500 operational cycles and mechanical stability shown by 1000 bending cycles.It was attached to a quadrupedal robot to demonstrate its ability to properly operate and communicate with an external system (e.g., smartphone) via Bluetooth.The developed wearable inertial platform holds great promise as a next-generation wearable motion-tracking platform and a soft human-machine interface.

Potential Wearable Technologies
Although there is a great deal of wearable devices implemented on robots, there still lies a gap between state-of-the-art wearable technology and what is currently implemented in robotics.In this section, we explore prospective wearable technologies that may be integrated with robots to enhance their functionalities and power management.

Sensors
Flexible sensors can collect valuable sensory information from the surroundings.They possess the ability to be deployed on curvilinear and dynamic surfaces, enabled by their mechanical conformity, flexibility, and stretchability. [62][80][81] Thermal and tactile sensing are often combined in the form of electronic skin (e-skin), which mimics the sensing function of the human skin. [82]However, because electronic skins for robots are well described elsewhere, [55] we turn our attention to individual flexible sensors that can be potentially integrated with quadrupedal robots and the benefits stemming from the applications.
Motion Sensors: As mentioned, motion sensors can assist the stabilization and locomotion of a quadrupedal robot by expanding its proprioceptive perception.While traditional motion sensors rely on rigid structures, wearable versions feature thin materials that can be flexed.Improving upon the previous design, [8] Babatain et al. developed a flexible biaxial accelerometer featuring a graphene-coated LMD (see Figure 3a). [77]The graphene-coated LMD is encased in a reverse-pyramid cavity with LIG electrodes to sense its position and hence infer the acceleration.The working principle is similar to that of the previous work.The device exhibits a sensitivity of 978 Ωm À1 s 2 , cross-axis sensitivity of 3%, and remarkable repeatability of over 120 000 cycles.As demonstrated in ref. [8], it is possible to attach the flexible sensor, for example, to a robot's torso to monitor its acceleration and orientation.This will serve as additional proprioceptive input to the robot, and with sophisticated algorithms, it is possible to infer some information about the terrain as well.
In addition to flexible MEMS motion sensors fabricated with conventional techniques, [58,83,84] paper-based accelerometers are gaining traction as well.Zhang et al. fabricated a flexible capacitive accelerometer consisting of a paper substrate, silver nanoink, and an insulating polymer. [85]The working principle is similar to that introduced in ref. [84] The maximum sensitivity of the device in the range between 1 and 10 g is 20 fFg À1 .Figure 3c shows the paper-based uniaxial piezoelectric accelerometer developed by Wang et al. which was fabricated by incorporating hydrothermal growth of zinc oxide nanowires on a cellulose paper substrate. [86]igure 3b shows the structure and optical image of the device.The device showed an analog voltage sensitivity of up to 16.3 mVg À1 in the range between À10 and 10 g.
These devices highlight the prospect of flexible motion sensors superseding their rigid counterparts in the field of robotics, where flexing, bending, and vibration occur frequently.Moreover, paper-based accelerometers are cost-effective solutions to wearable accelerometers for quadrupedal robots traversing in hostile environments because they can be easily replaced in the event of damage or complete failure due to their low cost.
Imaging Sensors: Imaging sensors are essential in health monitoring, imaging, optical communication, and environmental sensing. [87]Transforming them into wearable devices creates smart systems capable of detecting light stimuli in real time, which can be equipped by quadrupedal robots to improve their environmental awareness.Torres Sevilla et al. demonstrated a flexible imaging sensor based on monocrystalline complementary metal-oxide-semiconductor (CMOS) technology (see Figure 3c). [75]The device is composed of many active device islands with stretchable spiral interconnects, which give rise to 700% stretchability.Each island contains ten p-n junction photodiodes whose anode current varies in response to the incident photoenergy.The isolated island design on a monocrystalline silicon substrate simultaneously achieves high photoresponsivity and physical stretchability.Moreover, its CMOS-compatible fabrication process can lead to further miniaturization, which improves the sensor's resolution, as well as improved responsivity and reliability.
90][91][92] Wu et al. presented a highly responsive organic imaging sensor based on monolithic, vertically stacked two-terminal pixels as illustrated in Figure 3d. [93]By utilizing a diode-type organic photodetector with photomultiplication and optimizing the injection electrode and rectifying layers, the device exhibits high responsivity, low dark current, and high rectification under illumination.The organic image sensor demonstrates a strong pixel photoresponsivity above 40 AW À1 , allowing for weak-light imaging capabilities even at low intensities of 1 μW cm À2 .
Flexible imaging sensors as such can be deployed on the side of the torso of a quadrupedal dog to enable 360 degree omnidirectional light detection and imaging.This eliminates any blind spots that the robot may experience and thus assists it with obstacle detection and path planning.
Environmental Sensors: Quadrupedal robots equipped with multisensory platforms are especially effective in rescue missions, hazardous environments, and unmanned explorations.For instance, quadrupedal robots integrated with gas sensing arrays can be highly effective as first responder assistants to assess the presence of potentially toxic industrial chemicals, combustible gases, and/or chemical warfare agents, especially in confined/hard-to-maneuver places.Thus, reliably quantifying the sensitivity and selectivity of specific gases is necessary.However, there is currently no singular gas-sensing technology  [77] Copyright 2023, Wiley-VCH.b) Left: 3D structural model of the sensing element of the piezoelectric accelerometer on a cellulose paper substrate.Right: Optical image of the fabricated accelerometer.Reproduced with permission. [86]Copyright 2018, Multidisciplinary Digital Publishing Institute.c) Optical image of the omnidirectional flexible and stretchable CMOS imaging sensor.Reproduced with permission. [75]Copyright 2018, American Institute of Physics Publishing.d) 3D model of the organic imaging sensor based on monolithic, vertically stacked two-terminal pixels.Reproduced with permission. [93]Copyright 2019, Wiley-VCH.e) Left: Cross-sectional view of the structure of a single toxic gas sensor based on field-effect transistor structure.Right: 3D model depicting the toxic gas sensor array.Reproduced with permission. [78]Copyright 2020, Wiley-VCH.f ) Optical image of the fully standalone flexible underwater multisensory platform capable of sensing temperature, pressure, and pH level.Reproduced with permission. [68]Copyright 2019, Wiley-VCH.
that can sense a broad spectrum of gases on such a robotics platform.Among the various gas sensing technologies, the field-effect transistor (FET) technology has significant advantages including low power consumption and small dimensions.A representative example of a multigas sensor array based on FET technology is the chemical-sensitive FET (CSFET). [78]igure 3e shows the CSFET's structure.The sensing array consists of multiple CSFETs with different sensing materials, including ruthenium (Ru), silver (Ag), and silicon oxide (SiO x ), acting as the chemical gate to enable the detection of ammonia (NH 3 ) and hydrogen sulfide (H 2 S) and humidity at room temperature, respectively.For instance, in the presence of high humidity, the SiO x chemical gate will form an n-channel and the SiO x CSFET will begin conducting current.Such a silicon technology can be transformed into a flexible form by utilizing a back-flexing method based on deep reactive ion etching (DRIE) [94][95][96] or releasing the top functional layer with xenon difluoride (XeF 2 ) vapor phase etching. [97,98]n addition to ground robots, legged robots can be useful in remote underwater surveying due to their ability to traverse complex and unstructured terrain.Picardi et al. reported a bioinspired underwater legged robot capable of traversing uneven terrains, precise positioning, and suppressing noise, which collectively achieves nondisruptive seabed inspection. [99]Although such robots usually come with sensors necessary to complete the survey mission, it is still advantageous to functionalize them with wearable sensors for added sensing capabilities.The lightweight and free form factor of wearable sensors enable facile integration with underwater legged robots.Furthermore, wearable sensors can be employed on a robot's nonfunctioning surfaces (e.g., torso and legs) to functionalize them.One example of such a marine wearable sensor is Bluefin developed by Shaikh et al. (see Figure 3f ). [68]The sensor can detect the temperature, pressure, and salinity level of the surrounding water.The temperature is measured by a resistive temperature detector, and the salinity sensor is made of an interdigitated electrode, whose resistance changes with salinity.The pressure sensor is a conventional piezocapacitive sensor, whose capacitance changes due to the dielectric deformation caused by the applied pressure.It is integrated with a Bluetooth module and a customized flexible antenna to accomplish wireless communications. [100]Mass deployment of underwater legged robots equipped with wearable sensory platforms can lead to the next generation of marine exploration.

Energy Harvesters
The wearable sensors introduced so far expand the perception and sensing capabilities of quadrupedal robots.Ideally, such sensory platforms should be standalone; in other words, they should have their dedicated power management systems.Not only does it make wearable devices independent from the host system (i.e., quadrupedal robot) in terms of power management, but also it can be integrated with a quadrupedal robot to extend its run time.
Several mechanisms can be exploited to realize a wearable energy-harvesting system-solar power, piezoelectricity, thermoelectricity, and triboelectricity.Solar cells generate power by absorbing photoenergy through the photoactive layer and thus generating electron-hole pairs, which are then separated and collected by the anode and cathode of the cell.Thermoelectricity is the generation of electrical charge in response to a temperature gradient across a material.The definitions of piezoelectricity and triboelectricity are given in Section 3.1.1.Table 2 summarizes the achievable power density of each energy-harvesting method.Solar cells exhibit the highest power density, surpassing those of other energy harvesters by two orders of magnitude, due to the relatively large amount of solar power available on Earth's surface and the high energy conversion efficiency of solar cells.Commercially available monocrystalline silicon solar cells, which are the most widely available type of solar cell, possess power conversion efficiency (PCE) between 10.5% and 26.7% with a nominal PCE of around 20%. [101] Realizing fully flexible solar cells enables seamless integration into various wearable devices, opening new possibilities for supplying power and hence enhancing their adaptability.By leveraging lightweight and bendable materials, flexible solar cells provide a convenient and unobtrusive solution for capturing solar energy to power wearable devices and quadrupedal robots.Furthermore, the ability to conform to curvilinear surfaces on quadrupedal robots allows for efficient utilization of available space and maximizes exposure to sunlight. [102,103]ahabry et al. introduced a cost-effective, ultraflexible solar cell module using a corrugation architecture. [104]The monocrystalline silicon solar cells, with 17% PCE, incorporate a periodic corrugated array and interdigitated back contacts.These solar cells exhibit remarkable mechanical resilience, enduring high stress, and reversible deformation to form zigzag and bifacial modules.They maintain stability over 1000 bending cycles, including convex and concave bending, with back contacts supporting a bending radius below 140 μm.This research presents a significant advancement in ultraflexible solar cells, promising widespread applications.El-Atab et al. fabricated ultrastretchable monocrystalline silicon solar cells using laser patterning and corrugation techniques. [105]The solar cells achieve 95% stretchability (see Figure 4a) and maintain a PCE of 19%.In another study, various corrugation patterns are explored, with different levels of flexibility and silicon area loss (see Figure 4b).It is possible to control the bending radius and directions by utilizing different corrugation patterns. [106]hermoelectric 480 [112]   0.051 [128]  478 [113]   Power densities of solar cells are estimated assuming the average annual solar radiation incident on the Earth's surface of 342 W m À2 . [129]Thermoelectric energy harvesters' power densities are calculated given the temperature difference is 30 K.
Additionally, perovskite is a promising material for flexible solar cell applications due to its high PCE. [102]Li et al. demonstrated a highly flexible perovskite solar cells on a 57 μm thick polyethylene terephthalate (PET)-based substrate featuring a flexible hybrid anode. [107]The hybrid anode consists of a conductive polymeric layer and a silver-mesh layer and showed an optical transmission of 82%-86% in the visible spectrum, which is needed for light transmission to the underlying photosensitive layer.For the photosensitive layer and the cathode, a stack consisting of poly (3,4-ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS), CH 3 NH 3 PbI 3 (MAPbI 3 ), and phenyl-C 61 -butyric acid methyl ester (PCBM) and a thin aluminum layer were used, respectively (see Figure 4c).Although the fabricated device exhibited a PCE of 14%, which is lower than those of commercially available monocrystalline silicon solar cells, it could achieve a minimum bending radius of 2 mm, at which .Wearable energy harvesters that can be implemented in quadrupedal robots.a) Optical image of the corrugation-enabled ultrastretchable monocrystalline silicon solar cell in its relaxed and stretched states.Reproduced with permission. [105]Copyright 2019, Wiley-VCH.b) Optical image of different corrugation patterns that can be used to obtain various bending radii and directions.Reproduced with permission. [106]Copyright 2020, American Chemical Society Publications.c) Left: Material stack of the flexible perovskite solar cell on a PET-based substrate.Right: Cross-sectional SEM image of the solar cell.Reproduced with permission. [107]Copyright 2022, Springer Nature.d) 3D model illustrating the structure of the TENG with enhanced PTFE/PDMS nanocomposite.Reproduced with permission. [110]Copyright 2019, American Chemical Society Publications.e) Top: 3D model of the thin-film micrograted TENG.Bottom left: Optical image of the device under a bending condition.Bottom right: SEM image of the PTFE nanoparticles coated on the complementary metal gratings.Reproduced with permission. [111]Copyright 2014, Wiley-VCH.
the device retained 98.1% of its PCE.Furthermore, after 5000 bending cycles at a bending radius of 5 mm, the device was able to maintain 95.4% of its PCE because the hybrid anode is inherently flexible and thus robust against mechanical deformation.To improve the interface and PCE of perovskite cells, Li et al. proposed a solution using a self-assembled monolayer and a NiO nanocrystal film. [108]The reported solar cells demonstrate a remarkable 24.7% efficiency and retain their performance even after 10 000 bending cycles at a bending radius of 15 mm, both of which are favorable in wearable applications.
In addition to solar cells, triboelectric nanogenerators (TENGs) can be promising secondary energy sources because they can scavenge the ample amount of mechanical energy made during a quadrupedal robot's locomotion.A TENG comprises two triboelectric materials of opposite polarities, each of which has an electrical contact on the opposite ends.Electron transfer takes place from one contact to another when either of the following motions occurs: horizontal sliding motion or vertical contact-release motion. [109]i et al. reported a highly efficient flexible TENG capable of reaching the maximum output power density of 802.31 mW m À2 . [110]The TENG contains two layers as shown in Figure 4d; the upper layer is composed of a conductive textile, polytetrafluoroethylene (PTFE) membrane, and corona-charged PTFE/PDMS nanocomposite.The lower layer consists of a PDMS substrate with an array of pyramids and an electrode layer including nanofibers and silver nanowires.The two layers are brought together to realize the full device stack.By exploiting the contact-separation mechanism, the TENG supports two mechanical modes-direct pressing and bending.During both motions, the PDMS pyramids become compressed, and the effective distance between the corona-charged nanocomposite and the bottom PDMS substrate decreases, inducing electron transfer; the reverse applies during the release motion.Such a flexible TENG can be attached to the soles or joints of a quadrupedal robot to scavenge the mechanical energy made to power wearable devices and the robot itself.Zhu et al. also developed a shaped-adaptive thin-film micrograted TENG (MG-TENG) (see Figure 4e). [111]The MG-TENG is composed of a PTFE film with a metal micrograting array on each side, coated with PTFE nanoparticles.The metal gratings are misaligned by half pitch to induce electrification by sliding motions.The device exhibited an average output power density of 500 mW m À2 .
There are other mechanisms (e.g., piezoelectricity and thermoelectricity) available for energy harvesting, but there are practical considerations to be addressed.Because piezoelectric nanogenerators (PENGs) and TENGs both harness mechanical energy, they are in direct competition when it comes to real estate on a quadrupedal robot.As TENGs' power densities are two orders of magnitude larger than those of PENGs, it is advantageous to utilize TENG to harvest mechanical energy induced by the robot.
Thermoelectric generators (TEGs) possess power densities comparable to TENGs. [112,113]It is possible to attach TEGs to a robot's battery, where a considerable amount of heat is generated, to harvest the thermal energy.However, TEGs require a large temperature difference (e.g., 30 K), which is not easily obtainable in ambient conditions.For instance, in a hightemperature environment, the temperature gradient will be smaller, which significantly reduces the output power.Moreover, for reliable power generation, it is crucial to maintain the temperature difference.If a wearable TEG is attached to harness the heat generated by a quadrupedal robot's battery, there will be a delay until the battery temperature saturates and provides a constant amount of heat.Furthermore, the amount of heat generated is directly related to the power consumed by the robot.In the absence of active locomotion, the robot will draw less power from the battery, and hence, the battery temperature will decrease, reducing the TEG's power generation.
The examples of sensors and energy harvesters discussed thus far bespeak the transformation and functionalization capabilities of wearable devices.With the appropriate set of wearable sensors, not only can quadrupedal robots perceive more of their surroundings-terrain, obstacles, and inclination-but also it can remotely sense critical environmental parameters in hazardous environments instead of humans.Furthermore, wearable energy harvesters can be used to power these wearable devices and prolong the operation time of quadrupedal robots.As discussed, wearable devices can accessorize unutilized surfaces of a quadrupedal robot and provide additional functionalities.
To give an overview of the wearable technologies highlighted in our discussion, Table 3 summarizes their key parameters.

Data Management of Wearable Devices
So far, we discussed the sensory interfacing and power management of wearable devices.The last puzzle piece is data management.In this section, we examine how data collected by wearable sensors can be converted, interpreted, and utilized to achieve functional enhancement of quadrupedal robots.

Overview of Data Integration in Wearable Electronics
Efficient integration of wearable electronics is key to enhancing locomotive control and general sensing capabilities of a quadrupedal robot.In the context of sensing, there are largely two scenarios that a quadrupedal robot can face in a deployment.In the first scenario, the robot is merely the carrier of wearable sensors attached to its surface to collect mission-specific data from the surroundings.In this case, the collected data can be either stored in an on-system memory of the wearable devices for retrieval after the data collection is complete or directly transmitted the data to an external server or hub, for instance, via Bluetooth.In the latter case, the robot requires the environmental data to fine-tune its gait for locomotive stability.For example, the robot will need data about the terrain (e.g., ground roughness, inclination, and coefficient of friction) if the ground is unstructured or slippery to avoid slippage.In addition, because the robot can infer potential hazards or obstacles in its surroundings from the environmental data, it is also useful for path planning.While wires can establish the connection between the sensors and the robot for data communication in both scenarios, the wired connection will complicate the integration of the sensors with the robot.Because the two main driving factors behind utilizing wearable devices for quadrupedal robots are facile integration and wide applicability regardless of the robot's model, wireless communication is still a necessity.Furthermore, in both cases, it is beneficial to preprocess the collected data before transmission.Therefore, the analog environmental data collected by wearable sensors must be converted into digital data, so that it can be processed by a microcontroller and wirelessly transmitted by a Bluetooth module.
Figure 5 shows a high-level illustration of the data transfer flow from wearable sensors to a quadrupedal robot and an external client/server.As discussed, wireless data communication is indispensable in the facile deployment of wearable electronics in quadrupedal robots.This can be broken down into data acquisition, preprocessing, and transmission, which the sensing element, the microcontroller unit (MCU), and the Bluetooth low energy (BLE) module are responsible for, respectively, as shown in Figure 5a.Depending on the specification of the wearable sensors, it is possible to integrate more than one sensing element on one device (see Figure 5b). [7,68]In this case, multiple sensing elements can be interfaced with an MCU, which usually comes with multiple internal analog-to-digital converter (ADC) modules and a BLE module.This way, the number of modules used and the total peripheral circuit area can be reduced.

Digital Data Aggregation
Bubeck et al. utilized a field programmable gate array (FPGA) as a hardware accelerator, capable of specific parallel computation of high-resolution capacitance values of up to 100 sensors whose values are placed in a shared queue and available in digital output for use by the rest of the robot. [114]Although the authors' approach addresses the need for parallel data conversion, FPGAs are often bulky and rigid, which is not suitable for wearable applications.Another approach is to simply attach an ADC to each sensor.By embedding a flexible CMOS stress sensor in a foil package, Mahsereci et al. developed a flexible hybrid systemin-foil with a thickness of 20 μm. [115]The device contains a series of pressure sensors connected to integrators, which accumulate the difference in current between a control sample and each pressure sensor.Then, the analog data are converted to digital output via an ADC.As there are commercially available MCUs with ADC channels that come in bare-die forms, [116,117] it is favorable to utilize these ADCs in terms of device dimension minimization and integration simplicity.The same integration strategy can be applied to BLE modules.

Classification and Interpretation of Sensed Input
With the digitally converted data in hand, it is essential to classify the data for use by the appropriate control algorithms.
Interfacing software with the broad availability yet specificity  [68] of data accumulation from wearable electronics creates an extremely powerful tool for enhancing quadrupedal robots.Jenelten et al. developed a slippage-estimating algorithm based on a hidden Markov model, which utilizes the input data from the kinematic sensors embedded in the legs of a quadrupedal robot and the IMUs. [118]Data acquired by the sensors are converted and interpreted by the control algorithm to estimate slippage events.If slippage is detected, the algorithm polls the kinematic sensors to determine which leg has the greatest deviation in movement, allowing for corrective action.Moving beyond proprioceptive sensing, exteroceptive wearables have extremely powerful implications, especially when complemented with additional external data.Manzi et al. utilized the input data of human subjects obtained from a depth camera with a deep randomized decision forest classifier-where decision trees for each class are run in parallel to determine the most probable classification, thus allowing for faster training and reduced overfitting-to map specific pixels to joint movements based on an intermediate body parts representation. [119,120]In this study, a skeleton model of 15 joints, expressed as 3D Cartesian coordinates, at a sample rate of 10 Hz is produced.However, the use cases for such exteroceptive sensors do not end there.When skeletal data are combined with customized IMUs worn by the human subject to detect precise hand motions, the robot can recognize ten specific human activities and their relative positions.Complementing wearable data with additional sensing capabilities exponentially increases the use cases for such data and, in this case, provides functionality at home, in the warehouse, or even in emergencies.

Robotic Operating Systems
When managing the complexities of data and controls for a robot, an operating system that offers end users a level of abstraction allowing facile incorporation of sensed data is beneficial.As Mackenzcki et al. reported, robotic operating systems such as ROS2 use a topics model, which allows various control programs to receive updates as sensor data become available, enabling asynchronous execution. [121]Al-Yacoub et al. used ROS2 to create a robotic framework capable of tracking the current mental and physical state of a person via parameters including heart and respiration rate, eye contact, muscle tension, and body temperature to measure stress and act accordingly. [122]Many of these parameters are tracked via sensors mounted to the human's body, while information such as eye contact can be tracked directly by the robot.Externally mounted sensors publish data via Wi-Fi to the central controller for the robot, which then routes the messages to the appropriate control packages.When working with novel sensors such as heart rate or brainwave, custom packages have to be developed for the ROS to process the data and act based on input, while others such as motion by the human can be tracked using packages already available within the operating system.

Future Outlook and Conclusion
Legged robots are an amazing creation by humanity.They are no longer confined to science fiction.Increasingly they are getting closer to our homes to help us in our daily lives.While we wait for that future, we can clearly see their immediate applications in unchartered territory with uncertain environmental conditions.By accessorizing these legged robots and seamlessly interfacing with them, we can enhance their functionalities.Although there exist some critical modules of a quadrupedal robot that wearable electronics struggle to implement to date (e.g., LiDAR and timeof-flight sensors), due to the physical conformity, wearable accessories can be applied to any unused surface of the robot to provide added functionalities such as environmental sensing and energy-harvesting capabilities.To do so, there are multiple considerations to be addressed.The physical flexibility of a wearable device should not be the limiting factor of the robot's mechanical motion, and it must maintain its performance under different environmental conditions and mechanical deformations.Also, for facile integration with a quadrupedal robot, it must be fully standalone in terms of data and power management.This implies the device must ensure seamless data integration with the host system (e.g., a quadrupedal robot and external client/ server) and minimize its power consumption from the on-system battery or the wearable energy-harvesting module.Therefore, wearable electronics can be a great enabler and enhancer of legged robots' ability to be more assistive to our goals.

Figure 2 .
Figure 2.Wearable devices for quadrupedal robots.a) Flexible piezocapacitive sole sensor for small-legged robots.Reproduced with permission.[56]Copyright 2020, Institute of Electrical and Electronics Engineers.b) Structure of the FMSS with annotation to highlight the materials responsible for different sensing mechanisms.c) Different sensing mechanisms-piezoresistivity, piezoelectricity, and triboelectricity-are incorporated in the FMSS and their use cases.Reproduced with permission.[7]Copyright 2021, Multidisciplinary Digital Publishing Institute.d) Left: illustration and applications of the flexible inertial sensor with a soft graphene-coated eutectic gallium-indium (EGaIn) droplet proof mass.Right: scanning electron microscope (SEM) images of the soft proof mass and its surface.e) Working principle of the soft inertial sensor.Left: operational states without and with external acceleration.Right: variable resistance calculation depending on the proof mass position.Reproduced with permission.[8]Copyright 2022, American Chemical Society Publications.

Figure 3 .
Figure 3. Wearable devices that can be potentially implemented in quadrupedal robots.a) Left: Zoomed-in optical image of the meander LIG interconnect design of the soft biaxial accelerometer.Middle: Optical images of the device's flexible LIG interconnect.Right: Zoomed-in optical image of the interdigitated LIG interconnect.Reproduced with permission.[77]Copyright 2023, Wiley-VCH.b) Left: 3D structural model of the sensing element of the piezoelectric accelerometer on a cellulose paper substrate.Right: Optical image of the fabricated accelerometer.Reproduced with permission.[86]Copyright 2018, Multidisciplinary Digital Publishing Institute.c) Optical image of the omnidirectional flexible and stretchable CMOS imaging sensor.Reproduced with permission.[75]Copyright 2018, American Institute of Physics Publishing.d) 3D model of the organic imaging sensor based on monolithic, vertically stacked two-terminal pixels.Reproduced with permission.[93]Copyright 2019, Wiley-VCH.e) Left: Cross-sectional view of the structure of a single toxic gas sensor based on field-effect transistor structure.Right: 3D model depicting the toxic gas sensor array.Reproduced with permission.[78]Copyright 2020, Wiley-VCH.f ) Optical image of the fully standalone flexible underwater multisensory platform capable of sensing temperature, pressure, and pH level.Reproduced with permission.[68]Copyright 2019, Wiley-VCH.

Figure 4
Figure 4. Wearable energy harvesters that can be implemented in quadrupedal robots.a) Optical image of the corrugation-enabled ultrastretchable monocrystalline silicon solar cell in its relaxed and stretched states.Reproduced with permission.[105]Copyright 2019, Wiley-VCH.b) Optical image of different corrugation patterns that can be used to obtain various bending radii and directions.Reproduced with permission.[106]Copyright 2020, American Chemical Society Publications.c) Left: Material stack of the flexible perovskite solar cell on a PET-based substrate.Right: Cross-sectional SEM image of the solar cell.Reproduced with permission.[107]Copyright 2022, Springer Nature.d) 3D model illustrating the structure of the TENG with enhanced PTFE/PDMS nanocomposite.Reproduced with permission.[110]Copyright 2019, American Chemical Society Publications.e) Top: 3D model of the thin-film micrograted TENG.Bottom left: Optical image of the device under a bending condition.Bottom right: SEM image of the PTFE nanoparticles coated on the complementary metal gratings.Reproduced with permission.[111]Copyright 2014, Wiley-VCH.

Figure 5 .
Figure 5.A high-level overview of data transfer among a quadrupedal robot, external client/server, and wearable sensors: a) data transfer diagram for multiple individual wearable sensors and b) Data transfer diagram for a wearable multisensory platform similar to Bluefin.[68]

Table 1 .
Comparative analysis of extreme environments.Diverse extreme environments are listed to highlight the challenging conditions that robots may encounter.

Table 2 .
Power Densities of different flexible energy harvesters.

Table 3 .
Key parameters of the wearable technologies discussed herein.