Highly Stretchable and Sensitive Strain Sensor Based on Porous Materials and Rhombic‐Mesh Structures for Robot Teleoperation

Wearable sensors for human motion capture offer a promising human–robot interface to control robots in the teleoperation scenario, where robots could function as the second body of human operators to fulfill tasks remotely and accurately. In this paper, a novel strain sensor based on a soft polyurethane (PU) sponge and carbon nanotubes (CNT) is designed for motion capture of human joints. The unique 3D porous microstructure of the PU‐CNT sponge provides the sensor with high sensitivity. To bridge the gap between the high sensitivity and high stretchability of the strain sensor, a rhombic‐mesh structure with optimized geometric parameters, in conjunction with a pre‐compression design, is proposed for strain sensor prototyping, which endows the sensor with an extra elongation rate during the stretching process. The proposed PU‐CNT strain sensor manifests promising sensing performance with a stretchability of up to 300% and a maximum gauge factor of 3893, together with long‐term durability, low detection limit, and fast response capacity. Finally, the validation of the strain sensor is carried out by deploying the sensor on a human elbow to realize the teleoperation of a robot arm, which could be monitored through the digital twin model of the robot in a real‐time manner.


Introduction
[3][4] For DOI: 10.1002/adsr.202300044example, areas, such as Industry 5.0, [5] Healthcare 4.0, [6] homecare, [7] and exploration endeavors in space as well as underwater [8] are awaiting to profit from the robotics technologies currently in development.However, unpredictable conditions are prone to occur in unstructured environments, which bring challenges to robots to fulfill tasks since current machine intelligence is limited.Hence, it necessitates the intervention of human intelligence in task implementation, where professionals could control robots remotely in challenging tasks. [7]n such a context, a wearable strain sensor is a promising human-robot interface to enable human motion capture.It could be fixed on the human body, especially joints, thereby transferring human motion to robots intuitively. [9,10]One of the main challenges is to enhance the stretchability of the wearable strain sensor while maintaining high sensitivity, [11] since the rotary freedom of some joints is considerable, such as elbow and knee which give rise to large strain to sensors attached to it. [12]o date, researchers have proposed various strain sensor prototypes based on soft materials.[15] Among them, strain sensors based on soft porous materials, such as polyurethane (PU) sponge, are of great application potential. [16]The sensing capability of the strain sensors is attributed to the addition of conductive materials, such as carbon nanotubes (CNT) and various metallic nanowires, [17] which are deposited on the scaffold of the porous material. [18]The deformation of the scaffolds under external stimuli contributes to the connection or disconnection of the conductive fillers, [19] thus achieving the changes in sensing signals, as demonstrated in Figure S1, Supporting Information.The unique 3D porous microstructure provides sensors with high sensitivity, [20] as the open cell of the porous structure deforms even under a tiny force. [21]In addition, interferences toward human motion could be minimized due to the lightweight feature of soft porous materials. [22]o derive the motion signals of the joints in a full range, especially for joints with large rotary freedom, a wide detection range is a crucial property for wearable strain sensors.A common strategy is pre-stretching toward the substrate before the coating of the conductive fillers; [23,24] thus, the conductive paths of the sensor would keep connected when the strain sensor is stretched. [25]This method is effective in extending the detection range of the sensor when the inherent stretchability of the substrate is enough for the application scenario.For the substrate that has a limited stretchability, its stretchable structure design is a promising strategy, such as kirigami pattern, [26,27] wavy structure, [28] intertwined-coil configuration, [29] serpentine structure, [30] and double helical design. [31]In essence, the structure design transforms the local deformation of the strain sensors into local displacement when stretching, thus presenting a larger elongation rate along the stretching orientation.However, the decrease of the local strain attenuates the variation of the sensing signal, since the sensing function of most strain sensors is attributed to the deformation of the sensors.As a result, the stretchability of the strain sensor based on structure design is enhanced at the expense of sensitivity.
Herein, a wearable strain sensor prototype, which is composed of soft porous PU and CNT, is designed based on a rhombicmesh structure.Thus pre-compression along the stretching direction could be conducted toward the strain sensor with the aid of encapsulation of silicone rubber.Both high sensitivity and stretchability are achieved in this work.On one hand, the proposed sensor with the pre-compression design is endowed with an extra elongation rate, compared with that stretched from its initial length, as depicted in Figure 1a.On the other hand, the scaffold of soft porous PU covered with CNT deformed during the pre-compression process, and more conductive paths were generated thereby, which further enhanced the resistance change of the sensor during stretching.The stretchability of the strain sensor is further optimized by adjusting the geometric parameters of the rhombic-mesh structure based on the control variate method.Thus, the signal of joints in a full range of motion could be captured by strain sensors for robot teleoperation and the robot states could be monitored remotely through the robot digital twin, as depicted in Figure 1b.The contributions of this work are listed as follows: 1) The strategy of pre-compression is proposed based on rhombic-mesh structure design to enhance the stretchability of the strain sensor, while maintaining high sensitivity; 2) The stretchability of the strain sensor is further enhanced based on the optimization of the geometric parameters of the rhombic-mesh structure; and 3) A proof of concept for the proposed wearable strain sensor is demonstrated by attaching the sensor to human joints for robot teleoperation, where the robot states could be visualized by the robot digital twin, enabling remote monitoring of robot activities.

Prototyping of Soft Porous Strain Sensor
The proposed strain sensor mainly consists of a soft porous substrate, conductive fillers, elastomer encapsulation layers, and wires for electrical interconnection.In our study, a soft PU sponge is used as the porous substrate for its great elasticity and lightweight.In addition, the large specific surface area and high porosity of PU sponge are also crucial properties.Thus, more conductive fillers could be accommodated on the inner surface of the sensor, which is essential for sensing performance.To enhance the stretchability of the strain sensor, the middle region of the PU substrate is designed to be a rhombic-mesh structure with the parameter of 45°.Other detailed dimensions of the structure design are exhibited in Figure 2a.The deformability of the rhom-bic geometry endows the sensor with space for pre-compression.The rhombic-mesh structure of the PU sponge is patterned based on laser cutting technology.CNT with outstanding conductivity is selected to be coated on the scaffold of the PU sponge, forming conductive backbones.The PU-CNT sponge is pre-compressed to further improve the stretchability of the strain sensor.In order to maintain the pre-compression state of the strain sensor, each surface of the sensor is covered with a silicone rubber layer of 1 mm.The encapsulation of silicone rubber could also provide the sensor with better resilience and a longer life cycle considering its properties of water and heatproof, as well as high elasticity.The concrete fabrication procedure is exhibited in Figure 2a.Notably, the stretchability of the adopted silicone rubber is much higher than the pre-compressed PU-CNT sponge, circumventing the influence on the stretchability of the proposed strain sensor.To achieve electrical interconnection, copper wires with a diameter of 0.2 mm are inserted into the top and bottom of the sensor, both of which are perpendicular to the stretching direction of the sensor.
Figure 2b,c demonstrates the micromorphology of the pristine PU sponge and the PU-CNT sponge under a scanning electron microscope (SEM).The pristine PU sponge exhibited a porous and open-cell structure with a smooth surface.After the deposition of the CNT, the inner surfaces of the pristine PU sponge are covered with bunches of CNT fillers, which form conductive paths in the PU-CNT sponge.

Optimization of the Geometric Parameters
In order to achieve higher stretchability of the strain sensor, two more rhombic-mesh structures with different geometric parameters are designed for comparison.The detailed dimensions of the rhombic-mesh units are exhibited in Figure S2, Supporting Information.In order for variate-controlling, other geometric parameters of the two structures are designed to be the same as that of the design depicted in Figure 2a.The overall stretchability of the strain sensor depends on the shape of the rhombic-mesh unit of the substrate since the interior angle (45°, 90°, and 135°) determines the deformability of the rhombic mesh during the stretching process.The rhombic-mesh regions of the three samples with an initial length of 60 mm are pre-compressed to 30 mm, 30 mm, and 35 mm, respectively.The pre-compression degree of each sample is defined when the force needed for further compression is larger than 0.05 N, since excessive pre-compression will lead to buckling of the sponge sensor in practical application.To manifest the superiority of the rhombic design on stretchability enhancement, a PU-CNT conductive sponge without pattern design is also tested, functioning as the control group.All four samples are encapsulated with silicone rubber.
The stretchability of the four samples is compared under a uniaxial stretching load.Figure 3 presents the elongation rate of the four samples before fracture.It is depicted that the strain sensors with rhombic-mesh structure and pre-compression design are significantly better than that without structure design.In addition, the rhombic mesh with a parameter of 45°provides the strain sensor with maximum stretchability with an elongation rate of 335% when the fracture occurs, which is attributed to the larger deformability along the stretching direction compared with the rhombic geometry with parameters of 90°and 135°.Thus, the rhombic-mesh design with a parameter of 45°i s adopted for the proposed strain sensor in this work.To avoid damage caused by overstretching toward the strain sensor, the sensing range of the PU-CNT sensor is set as a strain of 300%.It should be noted that a smaller angle of rhombic mesh may not mean a wider sensing range, although the rhombic-mesh structure with the angle of 45°has the widest sensing range among the three samples.In this paper, the sensing range consists of a pre-compression stage and a tensile stage.A smaller grid angle intuitively improves the tensile stage by accommodating larger deformation along the stretching direction.But, it may shrink the whole sensing range due to the less space provided for precompression.Systematic mathematical modeling and more experiments are required to derive the explicit relationship between the grid angle and the resulting detection range.

Calibration of the Strain Sensor
The sensing performance of the PU-CNT strain sensor is investigated by monitoring the relative resistance variation ((R-R 0 )/R 0 ) at different strains (), where R and R 0 denote the resistance of the PU-CNT sponge after and before stretching, as shown in Figure 4.The sensitivity of the strain sensor is assessed using the gauge factor (GF) value, which is defined as ((R-R 0 )/R 0 )/∆.In accordance with the deformation form of the PU-CNT sponge, the resistance-strain curve could be divided into two stages.At the first stage of 0-100% strain when the PU-CNT sponge is in the pre-compression state, the resistance of the strain sensor increases monotonically during stretching.The GF value of the strain sensor is 1.8 at the original state, and increases continuously to 78.4 at the strain of 100% when the PU-CNT sponge recovers to the initial length.At the second stage of 100-300% strain, the GF value increases dramatically and reaches 3893 at the strain of 300%.The high stretchability and sensitivity are comparable to existing works on related strain sensors, which will be discussed in detail in Section 2.5.
The sensing mechanism of the proposed strain sensor is analyzed as follows.PU-CNT conductive backbones on different sides of each rhombic mesh contact with each other during the pre-compression process, with the formation of large amounts of conductive paths.During the first stage of 0-100% strain, the deformed rhombic geometry recovers gradually and the sides on each rhombic mesh separate from each other, leading to extensive disconnection of conductive paths.As a result, the impedance of the PU-CNT strain sensor increases.The second stage starts once the PU-CNT sponge is stretched to its initial length.Further stretching leads to the elongation of PU-CNT conductive backbones, and thus the CNT on each backbone loses electrical connection gradually, and the impedance increases continuously. [19]

Dynamic Performances of the Strain Sensor
To evaluate the dynamic stability of the PU-CNT strain sensor at various strains, 10 cycles of stretching-release loads are exerted on the sensor for each strain value, including 25%, 50%, 75%, 100%, 125%, 150%, 175%, and 200%.The dynamic stability performances of the sensor are demonstrated in Figure 5a,b.In each cycle, the relative resistance variation reaches a peak of a specific value after stretching and returns to the original value when released.It could be concluded that the PU-CNT strain has great repeatability and dynamic stability over a wide working range, including both the pre-compression state (0-100% strain) and the tensile state (100-300%) described in Figure 4. Furthermore, the sensing capability toward tiny strains is also a crucial indicator of the strain sensor.It is depicted in Figure 5c that the sensing response of the strain sensor remains almost unchanged at 0.5% strain under the dynamic load of 1 Hz.The relative resistance variation reaches 1.4%, which is attributed to the high sensitivity of the strain sensor at the pre-compression state.The low limit of the detection range endows the strain sensor with the potential for tiny movement detection of human joints, which is essential to the motion capture of the operator in teleoperation scenarios.
In order to explore the durability of the PU-CNT strain sensor, 10 000 stretching cycles are implemented to the sensor at both the pre-compression state (0-40% strain) and the tensile state (100-140% strain).It is shown in Figure 5d that the sensing signal remains stable at the pre-compression state, and no obvious draft occurs.As to the durability test at the strain of 100-140%, a slight decrease of about 2% occurs to the sensing signal, which is calculated by comparing the relative resistance variation of the first and last 20 cycles, as depicted in Figure 5e.The small draft of the sensing signal at the tensile state mainly results from the slight movement of CNT conductive fillers during the cyclic stretching process.Figure 5f demonstrates the transient response performance of the strain sensor.The response time is less than 82 ms under the strain of 40%, while the recovery time is 64 ms.The fast response capability of the wearable strain sensor is of great potential when it comes to the capture of quick actions of the human operator in teleoperation scenarios.

Discussion and Analysis
[34][35][36][37] Several performance indicators are selected for a more comprehensive comparison.It is shown that the developed strain sensor is comparable to the sensors listed in the table, especially in terms of stretchability and sensitivity (GF value).The balance of high stretchability and sensitivity is attributed to the combination of rhombicmesh structure and pre-compression design, as well as the high porosity of the PU sponge which enhances the specific surface area of the sponge to accommodate more conductive fillers.Thus, the PU-CNT strain sensor in this work is promising to be adopted for wearable devices to achieve sensitive detection of human joint motion in a full range.

Application in the Teleoperation Scenario
In the scenario of robot teleoperation, the wearable sensor mediates the joint motion signals into electrical signals, which function as the control input for the robot to move in a specific manner.Based on the robot digital twin model, the state of robot motion could be monitored remotely by the human operator in real time.Thus, the teleoperation of the robot as the second body of the human operator could be realized.
To verify the application prospects of the proposed PU-CNT strain sensor as a wearable device on joints, the sensor is attached to three typical joints, including the knee, elbow, and wrist, while the sensing signals of the PU-CNT strain sensor are recorded in real time during the joint motion. [38,39]As shown in Figure 6a-c, when the bending degree of each of the three joints increases, the signal derived from the sensor increases accordingly.The response of the strain sensor is stable and sensitive.Thus, the movement of joints could be measured based on the distinguishable resistance variation of the PU-CNT strain sensor.
To further validate the practical performance of the PU-CNT strain sensor in a teleoperation scenario, the sensor was fixed on the elbow of an operator to control a robot arm Jaco2 remotely.Based on the voltage divider rule, a readout circuit was established to enable real-time measurement of the resistance of the strain sensor.The resistance variation at certain joint angles was recorded in advance to derive the mapping between joint motion and the sensing data.
As shown in Figure 6d and Movie S1, Supporting Information, teleoperation toward the third joint of the Jaco2 robot arm was realized when the operator bends the elbow with the joint angle of 30°, 60°, 90°, and 120°in sequence.It is exhibited that the resistance of the strain sensor increases dramatically with the motion of the elbow.The sensing signal maintains a relatively stable value when the operator keeps his elbow at a certain bending degree.The small burr on the curve is caused by the slight shaking of the human body during the movement.More apparent burrs could be observed with the increase of the bending degree due to the higher sensitivity of the strain sensor at larger strains.Based on the real-time signals, together with the mapping between joint motion and sensing data that derived in advance, the robot arm could be controlled to follow the joint motion of the operator.Meanwhile, the posture of the robot arm could be monitored through the robot digital twin in real time, illustrating promising application prospects for robots to implement challenging tasks remotely with the aid of human intelligence.

Conclusion
In this study, a novel strain sensor prototype is proposed for motion capture of human joints.On one hand, the combination of rhombic-mesh structure and pre-compression design provides the strain sensor with an extra elongation rate, thus realizing stretchability improvement.On the other hand, the porous microstructure of the PU sponge endows the sensor with high sensitivity.The proposed PU-CNT strain sensor has a stretchability of up to 300% and a maximum gauge factor of 3893.Notably, the long-term durability of 10 000 cycles, as well as a fast response time of 82 ms, are also exhibited, thus ensuring stable and rapid sensing performance of the strain sensor.Finally, the developed sensor is deployed on the human joint for locomotion detection, illustrating appealing application prospects in teleoperation scenarios.
In future work, deeper analysis of the mapping relationship between the stretchability and the geometric parameter of the rhombic mesh would be carried out to achieve the optimization of the stretchability.

Characterization of the Strain Sensor:
As to the experiment of a geometric structure optimization, three strain sensors based on rhombic-mesh structures with different geometric parameters were fabricated as the aforementioned process.Then, a uniaxial stretching loading test was carried out toward the rhombicmesh part of the three sensors at the loading rate of 500 mm min −1 , using a universal test machine (34SC-05, Instron, USA).
As to the calibration test, the dynamic stability tests, and the durability tests, the strain sensor with the rhombic-mesh structure of 45°was stretched using the universal test machine, while the impedance values of the sensor were collected by a digital multimeter (Truevolt 34461A, Keysight Technologies, USA) at a sampling rate of 20 Hz (Figure S3, Supporting Information).
As to the transient response test of the strain sensor, a weight of 1 kg was tied to one side of the strain sensor, while another side was fixed (Figure S4, Supporting Information).The free fall of the weight at a specific height provided the sensor with a transient stretching load.The whole loading process ended when the weight falls on a cushion, which was utilized to control the elongation rate of the strain sensor.The resistance change of the strain sensor was recorded in real time by a digital acquisition (NI USB 6002, National Instruments, Texas, USA) at a sampling rate of 25 kHz.
Statistical Analysis: The original data derived from the transient response test were smoothed and denoised with the adjacent-average method using the software of Origin 2017.The average data was calculated for every 100 adjacent points.
Robot Teleoperation and Digital Twin System: The 6-degreeof-freedom robot arm Jaco2, provided by Kinova Robotics Inc., was controlled by the collected data from the sensor using a Mega2560 and a voltage divider circuit.The Mega2560 was connected to the control computer via wired serial communication.The control computer was connected to the robot's control port via Ethernet.The robot motion control was achieved using the robot operation system architecture with the joint position control interface of the kinova_ros open-source stack.The voltage signals collected by the Mega2560 were preprocessed through interpolation to obtain five levels of angles: 0°, 30°, 60°, 90°, and 120°, which were used to control the joint position of the third joint of the Jaco2 robot arm.Meanwhile, the robot digital twin was constructed using the Rviz visualization widget to monitor and reflect the real-time state of the robot.During the control process, all real-time data on the joint angles of the robot and sensor signal data were recorded using the rosbag plugin for subsequent analysis.Human involved in the robot teleoperation validation has provided informed consent, and the study protocol (IIT20220328B-R1) was approved by Clinical Research Ethics Committee of the First Affiliated Hospital, College of Medicine, Zhejiang.The operator in the robot teleoperation experiment was Z. Ye, one of the co-author, who had given his approval to publish these images and movies.

Figure 1 .
Figure 1.Design and application illustration of the PU-CNT strain sensor based on rhombic-mesh structure.a) Pre-compression design of the strain sensor, which endows the sensor with higher stretchability.b) Application of the strain sensor in robot teleoperation, which could be visualized in real time based on robot digital twin.

Figure 2 .
Figure 2. Prototyping of PU-CNT strain sensor.a) Structure design and fabrication process of the strain sensor.b) SEM image of a pristine PU sponge.c) SEM image of the PU-CNT sponge.

Figure 3 .
Figure 3. Stretchability comparison of PU-CNT sponge with different geometric parameters.

Figure 4 .
Figure 4. Calibration of the PU-CNT strain sensor.

Figure 5 .
Figure 5. Comprehensive sensing performances of the PU-CNT strain sensor.a) Dynamic stability of the strain sensor at strain values of 25%, 50%, 75%, and 100%.b) Dynamic stability of the strain sensor at strain values of 125%, 150%, 175%, and 200%.c) Detection limit test of the strain sensor.d) Durability performance of the strain sensor at the strain of 0-40%, together with the comparison between the first 20 cycles and the last 20 cycles.e) Durability performance of the strain sensor at the strain of 100-140%, together with the comparison between the first 20 cycles and the last 20 cycles.f) Transient response performance of the strain sensor.

Figure 6 .
Figure 6.The proposed strain sensor was adopted as a wearable device on the joint of a) elbow, b) knee, and c) wrist.d) Validation of the strain sensor was implemented by attaching the sensor onto the elbow of an operator to achieve robot teleoperation and real-time monitoring through a digital twin model.

Table 1 .
Comparison of strain sensors based on porous materials.