Soft robotics and soft wearable sensors are receiving increased attention due to their potential applications such as for rehabilitation or assistance purposes (Wehner et al., 2013; Polygerinos et al., 2015), human health monitoring systems (Ramasamy and Balan, 2018; He et al., 2021), human-machine interactions (Zoss et al., 2006; Banala et al., 2008; Roh et al., 2015), and human motion monitoring (Yang et al., 2018). Many soft actuator devices have already been developed. Examples of soft actuators include soft pneumatic actuators (Faudzi et al., 2018; Mohamed et al., 2020; Tschiersky et al., 2020; Ying et al., 2020; Luo et al., 2021), soft electroadhesion (Shintake et al., 2016; Guo et al., 2018; Okuno et al., 2019; Persson and Guo, 2019), stretchable pumps (Cacucciolo et al., 2019; Seki et al., 2020; Shigemune et al., 2021), dielectric elastomer actuators (DEAs) (Ji et al., 2019; Kajiwara et al., 2019; Minaminosono et al., 2019; Wang et al., 2019; Hiruta et al., 2021; Takagi et al., 2021; Wiranata et al., 2021; Zhao et al., 2021) a shape memory polymer (Besse et al., 2017; Li et al., 2020; Miao et al., 2020; Scalet, 2020; Zeng et al., 2020), and gel actuators (Maeda et al., 2007, 2010; Hwang et al., 2019; Mao et al., 2020a, 2020b; Zhang et al., 2020). To create an interface between soft actuators and human activity, the actuator must be equipped with an appropriate sensing method. One example is a pneumatic exoskeleton (Polygerinos et al., 2015), where the exoskeleton amplifies the normal muscle work precisely when the sensor and actuator are fully calibrated with the human motion.

In the context of human motion detection and soft robotics actuation monitoring, strain sensing is essential to detect deformation of the system. Although conventional strain sensors can convert the strain produced by the external stimulation into a signal response (Han et al., 2017), they cannot detect the full range of human motions. The conventional strain sensor structure usually consists of a high stiffness material conductor or semiconductor, limiting the strain range sensing (Barlian et al., 2009). Since the movement of human joints induces a large strain range (more than 30%), flexible strain sensors with a high stretchability are needed to detect the full range of human motions.

In the sensor-actuator integrated system, the flexible strain sensor provides feedback to the system by sending the resistance data change. The change in electrical resistance R of the flexible strain sensor is affected by the strain ε of the elastomer. The relative change of R to the ε is denoted as the gauge factor (GF) which also known as the sensitivity of the sensor. When the flexible sensor undergoes a uniaxial stress, the sensor should expand in one direction causing a change in the geometric shapes. The change in geometric shape is described as follows in Equation 1 (Shintake et al., 2018). l e = ε 1 l e 0 + l e 0 , w e = w e 0 ε 1 + 1 , t e = t e 0 ε 1 + 1 , (1) Where l _e is the electrode length, w _e electrode width, t _e electrode thickness and ε ₁ is the strain in the loading direction. l _e0 , w _e0 , t _e0 , are initial length, width, and thickness of the electrodes respectively. By assuming the cross section of the flexible sensor is uniform, the resistance change can be described as follow (Eq. 2) (Shintake et al., 2018) R = ρ l e 0 w e 0 t e 0 ( ε 1 + 1 ) 2 = ρ ρ 0 R 0 ( ε 1 + 1 ) 2 , (2) Where ρ ₀ is the reference resistivity, ρ is the resistivity of stretchable electrodes, and R ₀ is the reference resistance. Then the sensitivity of the sensor (GF) can be defined using Eq. 3 (Shintake et al., 2018) GF = Δ R R 0 ε 1 = 1 ε 1 ( ρ ρ 0 ( ε 1 + 1 ) 2 − 1 ) . (3)

Many researchers have focused on producing high-quality stretchable strain sensors. To improve the sensor quality (in terms of sensitivity, stretchability, and reliability), both material selection and the fabrication method are necessary to optimize devices containing stretchable electrodes. Different types of stretchable electrodes have been investigated, including carbon-based electrodes (Rosset and Shea, 2013), metal thin-film electrodes (Rosset and Shea, 2013), composite silicone-carbon electrodes (Shintake et al., 2018), and ionic gels (Keplinger et al., 2013). Each electrode has distinctive characteristics when it is used as a stretchable strain sensor.

Several fabrication methods have been reported for strain sensors, including a lamination technique for stretchable electrodes, mixing and creating electrodes-polymer composites, and direct electrode patterning on the elastomer. There are some examples of lamination techniques such as electrode pad printing (Poulin et al., 2015), Langmuir-Schaefer (LS) (Ji et al., 2018), and supersonic cluster beam implantation (SCBI) (Taccola et al., 2018). Among these methods, pad printing is the simplest to create stretchable electrodes because user can stamp the electrodes directly on the elastomer surface using a commercial pad-printing machine. Pad printing requires a liquid type of electrode, which requires additional treatment [e.g., mixing of carbon black and elastomer to form a carbon black-elastomer composite (Poulin et al., 2015)] to produce high-quality stretchable electrodes. Moreover, this methods also require a pad-printing machine (Poulin et al., 2015) which can add additional cost to the overall production of stretchable electrodes. In the case of SCBI (Taccola et al., 2018), it requires special equipment that can accommodate high pressure. The SCBI process also requires an additional inert gas which adding an additional cost to the fabrication process. In the case of LS (Ji et al., 2018), the method is simply by transferring the electrodes from the surface of the water to the target surface (e.g., PDMS surface), this method requires a good combination of conductive material before transferring to the target surface [e.g., a composite between hydrophobic poly (alkylthiophene) and hydrophilic multiwalled carbon nanotubes (MWCNT) (Ji et al., 2018)]. The material addition also can cause an additional cost. Moreover, a special cleanroom is also needed since the working properties is liquid and to avoid any contaminations.

Another fabrication technique is to mix and create electrode-polymer composites (Shintake et al., 2018; Teixeira et al., 2018; Jin et al., 2020; Dong et al., 2021; Song et al., 2021). This popular method can create stretchable electrodes for strain sensors but a long mixing process is required to realize a uniform mixing solution. Another fabrication method is electrode patterning by 3D printing (Muth et al., 2014). In this process, the key is to produce a good substrate quality before the electrodes are printed on the elastomer. Hence, substrate preparation requires additional complicated processes and special skills to achieve a good substrate composition.

Currently, the development of stretchable strain sensors via rapid, easy, reliable, and cost-effective fabrication methods remains challenging although the demand for stretchable strain sensor products is high (Song et al., 2021). In the soft robotics area, there are several fast fabrication techniques known as do-it-yourself (DIY) methods. In principle, DIY is an activity where researchers create a product or a process through individual or a collective production practice for a specific personal purpose, and usually the methods used are by researcher’s own innovation (Wolf and Mcquitty, 2011; Camburn and Wood, 2018). The DIY in soft robotics area can produce highly reliable actuators, including flexible electrohydrodynamic (EHD) pump technology (Seki et al., 2020) and basic dielectric elastomer actuator (DEA) technology (Wiranata et al., 2021). DIY methods are popular because they offer users the freedom to modify and define the parameters during device fabrication. Many open sources and communities exist to support the method. The DIY method also has been implemented in other areas such as biotechnology (Wong et al., 2018), biochemistry (Szymula et al., 2019), and other topics related to laboratory development (Kubínová and Šlégr, 2015; Grinias et al., 2016; Vallejo et al., 2020). Reliable DIY for rapid prototyping is essential to accelerate the development and innovation in soft robotics.

Previously, we have developed a method to reliably fabricate stretchable electrodes using the brushing method of CNT powder for DEAs (Minaminosono et al., 2021; Murakami et al., 2021; Wiranata et al., 2021; Wiranata and Maeda, 2021). Hand brushing CNT on the elastomer is a relatively easy and fast process. In order to eliminate the human influence in the brushing process, improve the consistency of the brushing and ease of brushing process, we automate the brushing process using brushing machine. An automatic brushing machine is a convenient technology for future mass production. For the brushing machine, we customized a low-cost commercial DIY-kit three-axis machine tools to realize an automatic brushing process. Then we improved the elastomer compatibility towards fabrication of stretchable strain sensor using automatic brushing method. The elastomer compatibility and quality was optimized by adding polyethyleneimine (PEIE) in the previously reported pre-polymer of polydimethylsiloxane (PDMS) 30-1 (Wiranata et al., 2021). Adding a small amount of PEIE (approximately 0.11 wt%) increased the stickiness of the elastomer, enhancing the compatibility for the brushing process of the CNT powder. To examine the effect of powder size on the stretchable strain sensor properties, we assessed three conductive powders: multiwalled carbon nanotubes (MWCNTs) with outer diameters of 6–9 nm (MWCNTs-1), MWCNTs with outer diameters of 10–20 nm (MWCNTs-2), and MWCNTs with outer diameters of 50–90 nm (MWCNTs-3).

Then to enhance the compatibility of the sensor for wearable devices, we introduced a lamination process using a thin Ecoflex membrane for the stretchable strain sensor after the brushing process. The Ecoflex membrane can cover the sensor since both the Ecoflex membrane and sensor display high stretchability mechanical characteristics. A high stretchability can ease material handling of micrometric layers of the Ecoflex membrane since the membrane does not rip easily during the lamination process. Finally, we integrated a stretchable strain sensor with a low-cost microcontroller (Arduino) as the signal processor to detect human hand movement. The hand movement was captured and displayed using virtual software.

Our study aims to provide a DIY approach for rapid, easy, reliable, and cost-effective fabrication methods for fully laminated stretchable strain sensors using automatic brushing machines to brush a carbon nanotube powder on the elastomer. The reasons for the cost and time reduction in our DIY approach include the use of low-cost commercial DIY-kit three-axis machine tools covered with a portable glovebox that can ease the brushing process in regular laboratory conditions. This can eliminate the requirement of a costly cleanroom. Furthermore, a direct brushing of CNT on the surface of the PDMS can speed up the fabrication process.

As a prototype of our powder-based stretchable sensor for wearable device, we integrate the sensor with a low-cost microcontroller (Arduino) to detect hand movements. Figure 1 depicts the overall fabrication sequence of the powder-based stretchable sensor and the wearable device prototype. We expect that our novel DIY strategy will contribute to the development of human-soft robotics interactions. Many researchers can easily access this proposed technology since our fabrication process is simple and reliable. We believe this research can attract more researchers and lead to the faster development of soft robotics, especially the human-soft robotics interface.

The electromechanical properties of the sensor (Figure 10) confirm that each MWCNTs powder has its own strengths and weaknesses characteristics when fabricated using a brushing method. We realize that the characteristic of our stretchable sensor is affected by the MWCNTs powder agglomerates. The smaller the agglomerates of the powder can produce less noise in the sensor reading. In our case the smaller agglomerates were found in MWCNTs-1. The dispersion of the MWCNTs particles in nanometric scale can strengthen the network connection between MWCNTs particle which also realize lower noise at higher strain value of stretchable sensor. This strong and uniform dispersion of the MWCNTs powder is found in MWCNTs-1 (Figure 9a1 and 9a2). This phenomenon makes PDMS-PEIE/MWCNTs-1 has a high potential for future development of stretchable strain sensor.

On the other hand, MWCNTs-3 shows a high nonlinearity and lower noise reading of the sensor. The high nonlinearity is highly affected by the structure of the MWCNTs-3 which looks stiffer than the other MWCNTs and the less density of the MWCNTs network structure. These two combinations of less density of network structure problem and stiff structure of MWCNTs particle can cause a degradation of sensor sensitivity when the sensor is at high strain condition.

In MWCNTs-2 case, the micro metric image (Figure 9B) shows that there are more agglomerates than MWCNTs-1 and MWCNTs-3. These agglomerates make the MWCNTs-2 has higher sensitivity and noise of the sensor reading at the same time. When the sensor is integrated to the system, a filtering algorithm (e.g., moving average filter or exponential filter) can remove the noise in the sensor reading of MWCNTs-2. However, in some cases, when a low-cost microcontroller is used, the filter algorithm application can cause a delay in the sensor reading. Since more terrible noise was found at a higher strain of PDMS-PEIE/MWCNTs-2 (Supplementary Figure S5B) than PDMS-PEIE/MWCNTs-1 and PDMS-PEIE/MWCNTs-3, PDMS-PEIE/MWCNTs-2 is more suitable for sensing a strain approximately between 0 and 50%.

By reviewing the strength and weakness of each type of powder when it is used as stretchable sensor, we found that PDMS-PEIE/MWCNTs-1 shows a good performance in terms of linearity and repeatability in lower and higher strain sensing areas. Hence, we recommend fabricating a combination of PDMS-PEIE and MWCNTs-1 via the brushing method for low-cost strain sensing projects (e.g., human movement detection) and wearable sensor such as smart gloves, which can detect finger movements.

In the future applications, this stretchable sensor can be used as a motion detection in soft actuator, soft bio inspired robot and human-soft robot interaction project. The flexibility of the sensor and the simple fabrication makes the stretchable sensor easily to be integrated to any device.

In this current research we integrate the stretchable sensor with a low-cost micro controller to detect finger movement. The principle is by detecting a higher or lower response of the stretchable strain sensor (due to the different movement angles shown in Figures 13C,D) and mapping the value of the sensor response through a simple transfer function to control a virtual hand. We prepared a virtual hand model using Blender (opensource software from blender.org). Blender receives the mapped signal from the strain sensor attached to the glove and mimics the human hand movement. On the instrumental side (wearable device), Supplementary Figure S8 shows the glove used to detect the hand movement and the electric instrument. In this stretchable strain sensor simulation, we used economical DIY equipment including Arduino Uno as the signal receiver and signal processor from the stretchable sensor. The electric circuit consisting of 1-MΩ resistors (Supplementary Figure S8) works as a simple voltage divider so that the analog signal can be read using Arduino Uno. Moving the glove (Supplementary Figure S8) induces a difference in the resistance of the system. This resistance change due to the hand-glove finger movement influences the voltage signal read by the analog pin in Arduino. Then the signal from the analog pin is further mapped and sent to the personal computer USB port. Finally, Blender software captures the hand movement simulation. Supplementary Figure S9 shows the detail of the electric connection circuit and peripheral of the wearable device.

To improve the movement accuracy of the wearable device (Supplementary Figure S8), a simple calibration algorithm was used to detects the maximum and minimum reading values of the sensor. This simple calibration algorithm is necessary since the movement range of every finger differs by person. A person’s hand size affects the value range of the sensor readings. Then the maximum and minimum range are further mapped to vector values that the Blender software can understand.

Figure 14 overviews the demonstration of this stretchable strain sensor attached to a cotton glove to control a virtual hand. (Supplementary Movie S2 shows a full demonstration of the control simulation, and Supplementary Figure S10 depict the strain possibility experienced by the stretchable sensor due to the finger movement). All of the materials, software, and peripherals are open-source and available in the marketplace. For instance, Sylgard 184 is commercially available with the price of $120 for 500 g, PEIE is around $100 for 100 g, MWCNTs ranges from $63 to $144 for one package. The price depends on the diameter of the MWNNTs, and one package can fabricate at least 100 samples sensors. The DIY-Kit x-y-z machine tools (SainSmart Genmitsu CNC Router 3018-PRO) is around $250, the portable enclosure glovebox (As One 3-116-01 SM-1) is around $390, while Ecoflex is around $40 per kilogram. Six nylon paintbrushes cost around $5, and the DIY-Kit Arduino is around $11. Hence, these materials and technologies are accessible and affordable for researchers studying soft robotics and stretchable sensors. Consequently, we expect that our DIY approach, which can simply fabricate an elastomer strain sensor using MWCNTs powder for a wearable sensor, can accelerate the growth of soft robotics research innovation.

Inconclusion, herein we fabricated the first carbon powder-based stretchable strain sensor produced via a DIY-approach and an automatic brushing machine. We highly recommend PDMS-PEIE with the composition of 3.2 wt% curing agent and 0.11 wt% PEIE for the stretchable elastomer. PDMS-PEIE is 50% more stretchable than that previously reported PDMS 30-1 (Wiranata et al., 2021) and realizes a 63% stickier elastomer. Stretchability is essential to cover a wide sensing range. Furthermore, a highly sticky surface improves the physical bonding quality between the elastomer surface and the MWCNTs powder. Strong physical binding between MWCNTs and the elastomer surface enhances the electrode conductivity (Wiranata et al., 2021) and reduces the risk of self-delamination of MWCNTs powder from the elastomer surface.

We also found that material selection for the electrode powder affects the sensor quality. Each carbon powder has its own strengths and weaknesses. MWCNTs-1 has a relatively better performance in terms of linearity and repeatability. The combination of MWCNTs-1 and PDMS-PEIE has a potential for higher strain sensing. We recommend using PDMS-PEIE/MWCNTs-1 strain sensor fabricated via a brushing machine for wearable devices because our demonstration shows that this material combination has a high reliability and repeatability.