Design and Control of Lightweight Bionic Arm Driven by Soft Twisted and Coiled Artificial Muscles

Abstract

Twisted and coiled actuators (TCAs), which are light but capable of producing significant power, were developed in recent times. After their introduction, there have been numerous improvements in performance, including development of techniques such as actuation strain and heating methods. However, the development of robots using TCA is still in its early stages. In this study, a bionic arm driven by TCAs was developed for light and flexible operation. The aim of this study was to gain a foothold in the future of robot development using TCA, which is considered as the appropriate artificial muscle. The main developments were with regard to the design (from actuator design to system design), system configuration for control, and control method. First, a process technology for repeatedly manufacturing TCA, which can be used practically and delivers sufficient performance, was developed. Based on the developed actuator, a joint was designed to move the elbow and hand. The final bionic arm was developed by integrating the TCA, pulley joint, and control system. It moved the elbow up to 100° and allowed the hand to move in three degrees of freedom. Using the control method for each joint, we were able to show the movement by using the hand and elbow.

Introduction

Natural muscles are outstanding complex actuators that have developed through long-term evolution.¹ People have made many attempts to mimic natural muscles through traditional actuators including motors and engines.^2–4 These actuators are similar to muscles in that they generate power, although their driving principles and methods are extremely different. Natural muscles are light and flexible, whereas motors and engines are heavy and solid.

To overcome these limitations, soft actuators such as pneumatic actuators, dielectric elastomer actuators (DEAs), and shape memory alloy actuators (SMAs) have been developed for designing artificial muscles.^5–7 Each type of actuator has been used to mimic nature based on specific characteristics. For example, pneumatic actuators have been used in octopus robots for flexibility, DEAs have been used in insect robots for fast driving, and SMAs have been used in worm robots based on their light weight.^8–11 However, they have limitations when used in developing robotic arms (bionic arms) for human applications.

A robot arm with a pneumatic actuator necessitates a noisy and heavy pump, making it difficult to apply directly to humans.¹² DEAs are difficult to fabricate in large-scale human muscle sizes, and they lack the stroke to mimic human arm movements, except for external prestrainers.^7–13 In addition, although SMAs are light and strong, they are difficult to control owing to their high hysteresis.¹⁴

Lightweight but high-power actuators, called twisted and coiled soft actuators (TCAs), were recently developed by applying twist insertion to a polymer fiber.¹⁵ Using polymer fibers that expand radially by heat, a contraction or expansion force is created.¹⁶ Similar to an SMA, a heat-driven method for a TCA can be easily implemented through electrical Joule heating.¹⁷ Although hysteresis still exists, it is significantly less than that of an SMA, and a control method has been developed to compensate for it.¹⁸ After the advent of TCAs, a number of studies have been conducted to improve their performance.

First, to generate heat for actuation through electricity, a conductive coating method and a fabrication method that uses a metal wire have been developed.^17,19 In addition, various materials and annealing methods have been applied to further increase the displacement and force of the TCA.²⁰ Recently, a double-helix twisted and coiled soft actuator (DTCA) was developed that satisfies the requirements of electrical drive, large output, and displacement, by twisting the elastic and conductive fibers together.²¹

The specific output of the TCA is ∼5.3 kW/kg, which is much higher than that of the motor (∼0.2 kW/kg), thus it is possible to make a lightweight robot.^15,22 Through the advantages of TCA, lightweight robotic applications were also developed but it is still insufficient.^23,24 Only hands or one joint was studied and the driver, cooling system, and sensor for control could not be integrated in the robot. Also in a previous article, Zhang et al. summarized major challenges of super-coiled actuator (same as TCA) to practically utilize for robotic applications.⁶ The challenges are bundle actuator to obtain large forces and cooling systems.

In this article, the following key contents were studied. Development of effective TCA in bundle, design of joint mechanism and robot, and configuration of control system with sensor, cooler, and controller will be described. Finally, a human-sized lightweight robotic arm called the bionic arm was developed by integrating all systems (Fig. 1). The resulting bionic arm could be controlled with the elbow and a three degree-of-freedom (DOF) hand by embedded control systems.

FIG. 1.

Configuration of lightweight bionic arm driven by TCA. TCA, twisted and coiled actuator.

Artificial Muscle Actuator section describes the actuator, manufacturing method, performance, and driving conditions. Design and Mechanisms section describes the human-sized design of the bionic arm. After introducing the joint mechanism, an elbow joint design method and a joint design method for the hand are proposed. The design of an actuator bundle for driving a joint was also described. In Control Systems section, a control system constructed with an embedded driving circuit for controlling the bionic arm and an algorithm for establishing control are detailed. Experiments and Results section presents the final results and a demonstration of the proposed system, and Conclusions section provides the concluding remarks.

Artificial Muscle Actuator

Twisted and coiled soft actuator

The TCA is shaped like a spring, so actuation is similar to stretching and contracting a spring (Fig. 2b). Therefore, a joint, which is where the actuator is applied, is composed of an agonist–antagonist couple, similar to a human muscle, as shown in Figure 2a. Bundle actuators and joints are designed based on this structure.

FIG. 2.

(a) Agonist–antagonist and actomyosin system of human, (b) TCA system requiring bidirectional system, (c) structure of DTCA, (d) DTCA fabrication device, (e) performance evaluation equipment. DTCA, double-helix twisted and coiled soft actuator.

In a previous article, Zhang et al. wrote that it is difficult to estimate the performance of a bundle actuator.⁶ Therefore, it is important to achieve repeatability in the manufacturing of an actuator for a similar performance. In this study, a DTCA that can be repeatedly produced was developed and applied. DTCA is made by twisting two fibers of spandex (Creora Power-fit 840D; Hyosung) and conductive thread (235/34 dtex 2-ply HC; Shieldex™) together, and the surface is insulated with spandex, which is effective for bundling. DTCA was implemented by developing a process technology using manufacturing equipment. Figure 2c shows the structure of a DTCA.

Fabrication and performance evaluation

For an actuator generating a high force, a number of unit actuators should be bundled together like natural muscles. Therefore, equipment was developed and used to manufacture DTCAs with similar performance. The configuration of the equipment is illustrated in Figure 2d. It is composed of a motor, a linear stage, and a relay switch. With fabrication equipment, the detailed fabrication process is provided in the Supplementary Data S1.

Also, for the desired size of a robot arm and joint, the characteristics of the actuator should be analyzed according to length and temperature. Isothermal and isometric tests of the fabricated DTCAs were conducted to analyze their characteristics and uniformities. Figure 2e shows the experimental device used to evaluate the performance of the actuator. A linear stage, push-pull gauge (RZ-1; Aikoh Engineering), IR thermal sensor (AMG8833; Adafruit), cooling fan (Fan 12025 red; Coolertec), and a PC were used to configure the equipment. The characteristics, performance, and differences over the unit's driving period were experimentally evaluated. Four identical samples were used for the experiments.

First, an isothermal test was conducted to determine the tensile properties of the actuator. These results are shown in Figure 3a. Six to 10 data points were used for performance evaluation. The maximum error between the samples was 0.561 N at 220 mm. After formulating the change in tension according to the increased length through linear fitting, the average value of a slope and a constant between the samples is as follows.

FIG. 3.

(a) Actuator extension test results. The actuator was subjected to tension through a linear stage, and this process was repeated 10 times at a speed of 2.5 mm/s from the initial 120–220 mm of actuator lengths. Inserted figure shows the resistance of the samples, and the resistances were 10.8, 10.9, 11.9, and 12.7 Ω. (b) Actuation strain by actuator length and load test results. The actuators were repeatedly driven 10 times from 25°C to 85°C for each load. (c) Actuator stress according to temperature and length, and (d) actuator stress according to length at 70°C. Color images are available online.

F_{k} = 0.0389 \cdot Δ x + 1.632,

(1)

where F_k is the tension of the actuator at room temperature and $Δ x$ is the elongated length. In addition, the resistance of the driver was measured during tensile testing. The inserted graph in Figure 3a shows the resistance of the samples during the test. Although a difference in resistance by the lengths was observed among the samples, it is assumed that the change in resistance is negligible and constant with an average value of 11.6 Ω.

Next, to determine the condition in which the actuator can be used most effectively, the contraction ratio (actuation strain) according to the load and length was evaluated (Fig. 3b). The actuation strain was calculated by dividing the length contracted by the actuation by the total length of the actuator stretched by the load. $A c t u a t i o n s t r a i n = \frac{c o n t r a c t e d l e n g t h}{a c t u a t o r l e n g t h} \times 100 (%) .$

Under a load of 300 g (2.94 N, 4.15 MPa), the actuator was not sufficiently tensioned, and thus, there was little displacement that could be driven. As the load increased, the displacement from the contraction increased, and the coils reached this load because the value exceeded the force that the actuator could produce. Finally, a maximum actuation strain of 38.1% at 400 g (3.92 N, 5.52 MPa) was shown with the length of ∼181 mm. However, at a high load of 450 g (4.41 N, 6.21 MPa), the actuator was unable to contract.

Figure 3c and d shows the results of an isometric test and the measured tension, which varies with temperature at a fixed length. As shown in Figure 3c, the strength of the actuator increased as the temperature increased. In addition, as with the previous isothermal test, the tension increased as the actuator length increased. Figure 3d shows the tension of each sample at 70°C. TCA is driven longitudinally by heat and can be modeled in terms of its mechanical properties (lengths) and thermal properties (temperature).²⁵ $F = F_{k} + F_{d} + F_{T},$ (2)

where F is the total tension of the actuator, F_d is the force due to the damping of actuation, and F_T is the force generated by heating. F_T can be expressed in various ways. This is because the force that increases with temperature varies with the stretched length. If the change according to the stretched length is set as the deviation from the mean, F_T is expressed as follows. The value of c was obtained from the slope of the trend line in Figure 3c. $F_{T} = c \cdot Δ T, c = 0.040 \pm 0.012 N ∕^{\circ} C$ (3)

The expression for the actuator tension at 70° is expressed as follows: $F_{70} = 0.0273 \cdot x + 0.6703 .$ (4)

If Equation (1) is subtracted from Equation (3), the force generated by the heat at 70° can be estimated. Δx is the length that extends from 120 mm. $F_{T, 70} = - 0.0116 \cdot Δ x + 1.644 .$ (5)

Design and Mechanisms

Overall design

The size of the bionic arm manufactured for this study was comparable with that in the human body (Supplementary Fig. S2a), and the design focuses on realizing the main movements of the arm. The movements to be implemented are at the elbow, thumb, index finger, and the remaining three fingers. The motion of the elbow has a range of 100° for mimicking a human motion, as shown in Figure 4a and d.²⁶

FIG. 4.

(a) Configuration of elbow joint, (b) design of elbow joint, and (c) characteristics of elbow joint; the blue line represents torque by the triceps, the red dotted line represents torque by the biceps, the yellow line represents torque by heating–actuation of the biceps, the green line represents torque by the forearm, and the red line represents torque by four strands of DTCA. (d) Elbow moves from 20° to 100° by biceps actuators. (e) Configuration of joint for hand, (f) tendon connection of three DOF hand, (g) design of DSP for hand, (h) three-pulley mechanism for adaptive grasping of three fingers, and (i) displacement and force required for hand actuation. DOF, degree-of-freedom. Color images are available online.

The fingers focus on the action used to hold the object, the thumb and index fingers move independently, and the other three fingers move together using a single actuator bundle. First, design variables were selected as shown in Supplementary Figure S4b and c to make a robot similar to a human arm. From Equations S1–S9 in the Supplementary Data, the core design goals were determined approximately, as shown in the following equation: $l_{u, a} < 191 m m, r < 30.9 m m, l_{f, a} < 179 m m,$ (6)

where $l_{u, a}$ is the length of the actuator in the upper arm, r is the radius of the elbow joint, and $l_{f, a}$ is the length of the actuator in the forearm.

Joint design

Similar to human muscles, TCAs can move by tensing an antagonist element. First, the forces that can be obtained in an actual joint are as follows: $\sum M_{e l b o w} = M_{a g o n i s t} - M_{a n t a g o n i s t} - M_{f} \pm M_{e} = 0,$ (7)

where M_f is the moment by the weight of the forearm, and M_e denotes the moment by external forces. $M_{a g o n i s t}$ represents the moment by the actuator contracting in the moving direction, and $M_{a n t a g o n i s t}$ is the moment when the actuator extends in the opposite direction. In the elbow, an agonist designates the biceps, and an antagonist designates the triceps. Using Equation (2), the detailed expression of Equation (7) is as follows: $\begin{matrix} (F_{B k} + F_{B d} + F_{B T}) r_{B} - (F_{T k} + F_{T d} + F_{T T}) r_{T} \\ - W_{f} s i n (θ) \cdot l_{g} \pm M_{e} = 0, \end{matrix}$ (8)

where $F_{B k}$ is the elastic force of the biceps, $F_{T k}$ is the elastic force of the triceps, $F_{B d}$ is the damping force of the biceps, and $F_{T d}$ is the damping force of the triceps. The heat forces are expressed as $F_{B T}$ and $F_{T T}$ . $θ$ is the joint angle, W_f is the weight of the forearm, and l_g is the length from the center of the joint to the center of mass of the forearm. In addition, r_B and r_T are the lengths of the moment arms acting on the joints. This equation is divided into each part of this study and summarized as follows: $\begin{matrix} \frac{(F_{B k} r_{B} - F_{T k} r_{T})}{M e c h a n i s m} + \frac{(F_{B d} + F_{B T}) r_{B} - (F_{T d} + F_{T T}) r_{T}}{C o n t r o l a l g o r i t h m} \\ = \frac{W_{f} s i n (θ) \cdot l_{g}}{D e s i g n} \pm \frac{M_{e}}{D i s t u r b a n c e} . \end{matrix}$ (9)

The joint mechanism attempts to address the force reduced by actuator tension. The problem is the decrease in the agonist's force by the antagonist as the pulley rotates. If TCAs are used for a joint, unlike in human muscles, high tension is continuously applied. Therefore, to properly generate force through the TCA, the joint must be designed to offset the tension. As a basic principle, the torque of both muscles is balanced regardless of the angle of the pulley. $F_{B k} \cdot r_{B} (θ) - F_{T k} \cdot r_{T} (θ) .$ (10)

Here the elastic force of a unit actuator can be obtained through an isothermal test. The actuator was used as a bundle, but it is assumed that each individual exhibits the same performance. The equation can be expressed as follows according to the number of actuators applied to each muscle. $F_{B k} = N_{B} \cdot (k_{u} (Δ x_{B} + x_{B 0} - x_{B i}) + F_{k i}),$ (11)

F_{T k} = N_{T} \cdot (k_{u} (Δ x_{T} + x_{T 0} - x_{T i}) + F_{k i}),

(12)

x = x_{0} \pm \int_{0}^{θ} r d_{θ},

(13)

where k_u is the spring constant of the unit muscle actuator (0.0389 N/m from the performance test result), $Δ x_{B}$ and $Δ x_{T}$ are the changed lengths from $x_{B 0}$ and $x_{T 0}$ , which are prestretched lengths in origin, $x_{B i}$ and $x_{T i}$ are the minimally required lengths for preventing entanglement (120 mm in this study), $F_{k i}$ is the required force to prevent entanglement of the unit actuator, and $N_{B}$ and $N_{T}$ indicate the number of muscle actuators. Using Equations (S10)–(S22) in the Supplementary Data, Equations (9)–(12) are summarized as follows.²⁷ $C_{r t}$ is a constant. $\begin{matrix} r_{B} = \frac{(C_{2} r_{T} + \frac{N_{T}}{N_{B}} R_{T} r_{T})}{\sqrt{{C_{1}}^{2} - 2 (C_{2} \int_{0}^{θ} r_{T} d θ + \frac{N_{T}}{N_{B}} \int_{0}^{θ} R_{T} r_{T} d θ)}} \\ w h e r e R_{T} = \int_{0}^{θ} r_{T} d θ, C_{1} = x_{B 0} - x_{B i} + \frac{F_{k i}}{k_{u}}, \\ C_{2} = \frac{N_{T}}{N_{B}} \cdot (x_{T 0} - x_{T i} + \frac{F_{k i}}{k_{u}}), r_{T} = r_{m a x} - C_{r t} \cdot θ \end{matrix} .$ (14)

Elbow joint

Based on the size determined in Equation (6), the maximum length of the actuator was set to 180 mm and the maximum radius of the pulley was selected to be 30 mm, considering the manufacturing process. By applying Equations (6) and (13), the Dual Spiral Pulley (DSP) is designed as shown in Figure 4b. The detailed parameters used in the design are listed in Table 1.

Table 1.

Design Parameters of Dual Spiral Pulley for Elbow Joint ( x_i is the Initial Length, N_B and N_T Are the Number of Double-Helix Twisted and Coiled Soft Actuators in the Bundle)

Parameters	Values	Parameters	Values
$x_{B 0}$	180 mm	$x_{T 0}$	120 mm
x_i	120 mm	r_T	30–6θ mm
k_u	0.0389 N/mm
$F_{k i}$	1.632 N
N_B	20	N_T	20

Figure 4c shows the torque for each element acting on the joint predicted by the designed DSP. The dotted red and blue lines indicate the torque exerted by the biceps and triceps on the joint, respectively. From these results, it can be seen that the actuators cancel each other by acting with the same force according to the angle.

Hand design and joint

The joint for driving the hand is used only to extend the actuator, fingers are connected to the actuator through the tendon (Fig. 4e). Therefore, the joint for the elbow was designed to create the desired angle with the actuation displacement, whereas the joint for the hand was designed to allow the size of the pulley to be fitted on the wrist. With Equation (S9) in supplementary and the wrist frame radius of 30 mm, the maximum outer radius of the designed pulley was set 20 mm. The elastic band (Spandex) used as an extensor was constructed with an elastic modulus of the actuator. The final designed pulley is shown in Figure 4g, and the variables and coefficients are listed in Table 2.

Table 2.

Design Parameters of Dual Spiral Pulley for Hand Joint

Parameters	Values	Parameters	Values
$x_{a g o, i}$	177 mm	$x_{a n t, i}$	120 mm
$x_{a g o, f}$	120 mm	r_T	20–6θ mm
k_u	0.0389 N/mm	$k_{a n t}$	0.0894 N/mm
$N_{a g o}$	2

$x_{a g o, i}$ is the initial length of the agonist, $x_{a n t, i}$ is the initial length of the antagonist, $x_{a g o, f}$ is the fully contracted length of the agonist, $k_{a n t}$ is spring constant of the antagonist, and $N_{a g o}$ is the number of double-helix twisted and coiled soft actuators in the agonist.

The configuration of the tendons connected to the hand is shown in Figure 4f. The tendons of the thumb and index finger are directly connected to the actuator. However, the tendons for the other three fingers are connected because it may not be efficient to implement all the DOF of the hand in a confined space. A part of the hand synergy for adaptive grasping is employed for efficient three-finger operation by using three pulleys (Fig. 4h).^28,29 Three-pulley mechanisms allow the three fingers to grasp the object adaptively. The displacement and force required to bend each finger are shown in Figure 4i.

Actuator bundle design

The number of actuators in the muscle bundle of the elbow can be determined from Figure 4c. First, the arm was assumed to be standing perpendicular to the ground. The green line represents the torque applied to the joint according to the target forearm weight, and the yellow line represents the torque caused by the force generated when heat is applied to the biceps. In Figure 4c, the yellow line represents the force owing to the F_T term. According to the results, the forearm could be lifted with 20 strands. However, there is little power left after the forearm is lifted to ∼80°.

If none of the actuators in the bundle shows sufficient performance, a situation may occur in which an object cannot be lifted. To compensate for this, if four actuators are added to the biceps, the red line shown in Figure 4c can be obtained. The red line represents the torque generated by the elastic force of the actuator added to the biceps. This torque can offset the torque caused by the forearm load to a certain extent, and the equilibrium is achieved at ∼25°, where the red and green lines intersect. For fingers actuation, two strands of TCA are used. The detailed hand and bundle design is shown in the Supplementary Data.

Control Systems

Prototype of bionic arm

As previously designed, the bionic arm is divided into an upper arm system for driving the elbow and a forearm system for driving the hand. The elbow is aimed at angle control, and the hand is aimed at grasping. The uppermost shoulder part was made of aluminum, and the remaining parts were produced through 3D printing with a carbon pipe as the main frame. As a result, it weighed only 940 g excluding the cover and 1190 g including the cover. The overall size reflects the length of the adult male arm; the length of the upper arm is 320 mm, and that of the forearm is 430 mm. The length of the hand is 180 mm (Fig. 5a). The final bionic arm with the cover is shown in Figure 5e.

FIG. 5.

(a) Inside view of the designed bionic arm, (b) detailed arrangement of upper arm circuit, sensor, actuator, and cooling fan, (c) arrangement of forearm sensor, actuator, and cooling fan, (d) simplified circuit diagram, (e) outside view of the covered bionic arm.

The detailed configuration of the upper arm system shown in Figure 5b is as follows. An encoder used to measure the angle of the elbow, two IR thermistors (MLX90614, CJMCU) to measure the temperature of the actuator, and two load cells to measure the tension of the actuator were installed. In addition, two pairs of actuator bundles (24 and 20) and two pairs (seven each) of cooling fans were installed. The detailed performances of cooling fans are given in the Supplementary Data.

The hand was designed for grasping with three DOF (Fig. 5c). Therefore, three DSP joints were attached to the wrist in a row, and it was difficult to mount the encoder. For the control, an IR temperature sensor and cooling fan were mounted, and temperature-based control was performed according to the model equation of TCA.

Driving circuits

To utilize the bionic arm in the future, embedded driving circuits (Supplementary Fig. S4b, c) were developed with two Micro Controller Unit (STM32F373, Arduino nano). Also for variable inputs, Metal Oxide Semiconductor Field Effect Transistors (CSD18510KTT—TI, AOD4185—AOS) are used. Both of the sensors and actuators of the Bionic Arm can be operated with these circuits. The configuration of the driving circuit is shown in Figure 5d and Supplementary Figure S4a.

Control algorithm

Figure 6a shows an isothermal test measured while moving the forearm of the prototype with an external force. The routine of the elbow movement is shown in Figure 6b. This test shows that the designed characteristics and actual measured characteristics are similar but different. Therefore, this article proposes a method to infer a control model equation based on a sensor mounted for elbow control. From Figure 6a, a mechanical model for control can be obtained. Subsequently, the thermoelectric model should be calculated. The following equation from Equations (S23)–(S25) is added for temperature change according to the electrical input.²⁵

FIG. 6.

(a) Designed muscle actuator force and measured force by load cell, (b) isothermal test routine for (a). Color images are available online.

C_{t h} \frac{d T (t)}{d t} = \frac{V^{2} (t)}{R} - λ (T - T_{a m b}),

(15)

where $C_{t h}$ denotes the thermal mass of the muscle actuator, $\frac{V (t)}{R}$ is the electrical input to the actuator, and $λ$ is the absolute thermal conductivity. $C_{t h}$ can be calculated as the product of the time constants τ and $λ$ .²⁵ T is the current temperature, and $T_{a m b}$ is the ambient temperature.

Response experiments with the step input were conducted to determine all the constants and control equations. The experiment was designed in the same way for the biceps and triceps. First, six-step voltages were used for the biceps experiment, and heating for 100 s and cooling for 100 s were repeated (Fig. 7a). Figure 7b shows the response of temperature (red line) and joint angle (blue line) according to time. Figure 7c indicates the response to temperature in each heating step. Over time, it reaches steady state, and if the temperature change is constant, the absolute thermal conductivity can be calculated. Equation (15) is summarized in the steady state as follows.³⁰

FIG. 7.

(a–e) Experiments of biceps muscle actuator: (a) input voltages (0.47, 0.97, 1.41, 1.87, 2.35, and 2.82 V), (b) results of temperature and elbow joint angle, (c) temperature change by input voltage, (d) relationship between force and temperature during a 2.82 V heating cycle, (e) range of joint angle change by input voltage. (f–j) Experiments of triceps muscle actuator: (f) input voltages (0.14, 0.28, 0.42, 0.56, 0.70, and 0.84 V), (g) results of temperature and elbow joint angle, (h) temperature change by input voltage, (i) relationship between force and temperature, (j) range of joint angle change by input voltage. Color images are available online.

λ = \frac{V^{2} (t)}{R \cdot (T - T_{a m b})} .

(16)

The characteristics of the actuator bundle $F_{k}, F_{T}$ can be obtained from Figure 7a and d. From the data in Figure 7a, the $F_{B k}, F_{T k}$ equations are calculated through linear fitting. As in Equation (3), the thermal constant can be calculated using a linear trend line.

In Figure 7e, the relationship between the angle and input can be obtained by considering that the angle has increased the most in the steady state. The gray line is the trend line drawn along each point in a steady state. $K_{B, P}$ can be obtained by mapping the gray line.

Models for the triceps were inferred using the same process for the biceps. Six-step voltages were used for the triceps experiment with same cycles (Fig. 7f). Figure 7g shows the response of temperature (red line) and joint angle (blue line) according to the heating cycle time. Figure 7h–j is derived in the same way as the biceps. The gray line shown in Figure 7j is a trend line that connects the elbow angle in the steady state. Table 3 gives the values of the finally obtained variables.

Table 3.

Model Parameters for Elbow Joint

Parameters	Definition	Value	Unit
$k_{B, θ}$	Biceps spring constant	−0.43	N°
c_B	Biceps thermal constant	0.57	N/°C
R_B	Biceps resistance	2.6	Ω
$λ_{B}$	Biceps absolute thermal conductivity	0.083	W/°C
$C_{B, t h}$	Biceps thermal mass	2.26	Ws/°C
$k_{T, θ}$	Triceps spring constant	0.46	N°
c_T	Triceps thermal constant	0.42	N/°C
R_T	Triceps resistance	2.6	Ω
$λ_{T}$	Triceps absolute thermal conductivity	0.10	W/°C
$C_{T, t h}$	Triceps thermal mass	0.45	Ws/°C

The values that change according to the power of the cooling fan are summarized in Supplementary Table S4.

Experiments and Results

Elbow control

Based on the obtained models and Table 3, a position control experiment was conducted. The overall thermomechanical control system is expressed as follows: $\begin{matrix} I_{f} \frac{d^{2} θ}{d t^{2}} + b_{θ} \frac{d θ}{d t} + k_{T, θ} \cdot θ + k_{B, θ} \cdot θ \\ = (c_{B} \cdot Δ T_{B}) r_{B} - (c_{T} \cdot Δ T_{T}) r_{T} - W_{f} s i n (θ) \cdot l_{g}, \end{matrix}$ (17)

where I_f is the moment of inertia of the forearm. The feedback controller according to Equations (17) and (S27)–(S30) was configured using the mentioned control system, as shown in Figure 8a.

FIG. 8.

(a) Closed-loop controller for elbow control, (b) the result of elbow position control for sine wave of 100 periods, (c) the result of elbow position control for sine wave of 20 periods, (d) angle error between real angle and desired angle in (b, c), (e) control response for step input of 35° with P-gain and D-gain, And (f) frequency response plot from sine wave control in Supplementary Figures S7 and S8. Color images are available online.

The test was conducted with sine waves and a 35° step input (Fig. 8e). In sine wave tests, the desired elbow angle was an amplitude of 20° and periods of 100 to 1 s (Supplementary Figs. S7 and S8). Figure 8b and c shows the experimental results at periods of 20 and 100 s. Figure 8d shows the error between the desired angle and the actual angle, and the maximum angle error was 2.1°. Figure 8f shows the frequency response plot from sine wave tests.

Hand grasping

Hand grasping was conducted based on the actuator models in Equations (1)–(5). Figure 9a shows the movement of the thumb continuously, Figure 9b shows the movement of the index finger continuously, and Figure 9c shows the movement of the remaining fingers continuously. Figure 9d shows a graph of the temperature change of the actuator during hand operation using the mounted IR temperature sensor. A colored circle is displayed near the steady state, indicating the moment when the finger is bent.

FIG. 9.

Continuous movement of (a) thumb actuation, (b) index finger actuation, and (c) the rest fingers actuation. (d) Temperature change of actuator and points at continuous figure time. The yellow timeline is for the thumb actuator. To match the pictures of the continuous motions in a–c, the corresponding points are marked using colored circles. A 4.5 V step input is applied to the actuator moving the thumb, and 10 V step input is applied to the actuator moving the index and the rest of the fingers. Color images are available online.

Combined motion

The developed elbow control method and hand grasping method were combined to hold the object, which was the target of the bionic arm, and raised to the desired angle. The soft hair was caught through the bionic arm, and after being held, it was controlled at 40°. Figure 10 shows the process of performing the action in a series of pictures.

FIG. 10.

Bionic arm combined motion using three DOF hand grasping and elbow position control.

Conclusions

Since the development of an excellent artificial muscle called TCA, many studies and improvements have been made. However, studies using TCA as a robot have been insufficient. Based on the process of actuator to joint design according to the characteristics, we developed a novel bionic arm including control system (cooling system, embedded controller, and control algorithm) summarized in Table 4. The bionic arm was manufactured to be 750 mm long, mimicking the size of a real human arm and weighing ∼1 kg. The bidirectional control algorithm was extracted, and angle control within 2.1° error was achieved only by p-gain control.

Table 4.

The Specifications of the Developed Bionic Arm

Specification		Value
Length	Upper arm	430 mm
	Forearm (hand)	320 mm (180 mm)
	Total	750 mm
Weight	Upper arm	0.73 kg
	Forearm	0.21 kg
	Total (with cover)	0.94 kg (1.19 kg)
Maximum speed	Bending motion	10.7°/s
Maximum speed	Stretching motion	−15.2°/s
Frequency	20° amplitude	0.05 Hz
Frequency	0.9° amplitude	0.5 Hz
Maximum load capacity	Bending motion	2.1 N
Maximum load capacity	Stretching motion	4.6 N

Maximum speed is extracted from Supplementary Figure S8f. And the maximum load capacity is obtained in Supplementary Figure S9.

In this article, the measured data of a newly developed actuator were used to design the robot, and methods may be difficult to use for general TCA. If the previously developed model of TCA is utilized, it will be possible to design and utilize the robot for various and general TCAs.^31–36 Also, we have focused on implementing movement. In the future, the number of actuators in the muscle bundle will be increased to exert greater power. The actuator, design, and control system of bionic arm in this study can be used for a variety of robots in the future. For example, it could be used as a manipulator with multiple bidirectional joints, or as a gripper or an assistive robot.³⁷

Footnotes

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This study was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government(MSIT) (No. 2021R1A2C3012387) and the convergence technology development program for bionic arm through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (No. 2014M3C1B2048175).

Supplementary Material

References

Hunter

, Lafontaine

. A comparison of muscle with artificial actuators. In: 5th Technical Digest Solid-State Sensor and Actuator Workshop, Hilton Head, SC: IEEE,, 1992:178–185.

Bar-Cohen

. Biomimetics—using nature to inspire human innovation. Bioinspir Biomim, 2006; 1:1–12.

Ananthanarayanan

, Azadi

, Kim

. Towards a bio-inspired leg design for high-speed running. Bioinspir Biomim, 2012; 7:046005.

Żbikowski

, Ansari

, Knowles

. On mathematical modelling of insect flight dynamics in the context of micro air vehicles. Bioinspir Biomim, 2006; 1:R26–R37.

Rodrigue

, Wang

, Han

M-W

, et al. An overview of shape memory alloy-coupled actuators and robots. Soft Robot, 2017; 4:3–15.

Zhang

, Sheng

, O'Neill

, et al. Robotic artificial muscles: current progress and future perspectives. IEEE Trans Robot, 2019; 35.3:761–781.

Jung

, Yang

, Cho

, et al. Design and fabrication of twisted monolithic dielectric elastomer actuator. Int J Control Autom Syst, 2017; 15:25–35.

Laschi

, Mazzolai

, Mattoli

, et al. Design of a biomimetic robotic octopus arm. Bioinspir Biomim, 2009; 4:015006.

Cao

, Burgess

, Conn

. Toward a dielectric elastomer resonator driven flapping wing micro air vehicle. Front Robot AI, 2019; 5:137.

10.

Yuk

, Kim

, Lee

, et al. Shape memory alloy-based small crawling robots inspired by C. elegans. Bioinspir Biomim, 2011; 6:046002.

11.

Kim

, Lee

, et al. An earthworm-like micro robot using shape memory alloy actuator. Sens Actuat A Phys, 2006; 125:429–437.

12.

Ohta

, Valle

, King

, et al. Design of a lightweight soft robotic arm using pneumatic artificial muscles and inflatable sleeves. Soft Robot, 2018; 5:204–215.

13.

Anderson

, Gisby

, McKay

, et al. Multi-functional dielectric elastomer artificial muscles for soft and smart machines. J Appl Phys, 2012; 112:041101.

14.

Webb

, Dimitris

, Andrew

. Hysteresis modeling of SMA actuators for control applications. J Intell Mater Syst Struct, 1998; 9:432–448.

15.

Haines

, Lima

, Li

, et al. Artificial muscles from fishing line and sewing thread. Science, 2014; 343:868–872.

16.

Haines

, Li

, Spinks

, et al. New twist on artificial muscles. Proc Natl Acad Sci U S A, 2016; 113:11709–11716.

17.

Park

, Yoo

, Seo

, et al. Electrically controllable twisted-coiled artificial muscle actuators using surface-modified polyester fibers. Smart Mater Struct, 2017; 26:035048.

18.

Masuya

, Ono

, Takagi

, et al. Modeling framework for macroscopic dynamics of twisted and coiled polymer actuator driven by joule heating focusing on energy and convective heat transfer. Sens Actuat A Phys, 2017; 267: 443–454.

19.

Semochkin

. A device for producing artificial muscles from nylon fishing line with a heater wire. In: 2016 IEEE International Symposium on Assembly and Manufacturing (ISAM), Fort Worth, TX: IEEE,, 2016:26–30.

20.

Yang

, Cho

, Kim

, et al. High performance twisted and coiled soft actuator with spandex fiber for artificial muscles. Smart Mater Struct, 2017; 26:105025.

21.

Kim

, Cho

, Jung

, et al. Double helix twisted and coiled soft actuator from spandex and nylon. Adv Eng Mater, 2018; 20:1800536.

22.

Wang

, Lu

. Building lightweight robots using single-motor drives—a survey and concept study. In 2016 IEEE International Conference on Advanced Intelligent Mechatronics (AIM), Banff, AB, Canada: IEEE,, 2016:676–682.

23.

Arjun

, Saharan

, Tadesse

. Design of a 3D printed hand prosthesis actuated by nylon 6-6 polymer based artificial muscles. In: 2016 IEEE International Conference on Automation Science and Engineering (CASE), Fort Worth, TX: IEEE,, 2016:910–915.

24.

, de Andrade

, Saharan

. Compact and low-cost humanoid hand powered by nylon artificial muscles. Bioinspir Biomim, 2017; 12:026004.

25.

Yip

, Niemeyer

. On the control and properties of supercoiled polymer artificial muscles. IEEE Trans Robot, 2017; 33:689–699.

26.

Malagelada

, Dalmau

, Vega

, et al. Elbow anatomy. Sports Inj Prev Diagn Treat Rehabil, 2014; 2:527–553.

27.

Cho

, Kim

, Yang

, et al. Artificial musculoskeletal actuation module driven by twisted and coiled soft actuators. Smart Mater Struct, 2019; 28:125010.

28.

Catalano

, Grioli

, Serio

, et al. Adaptive synergies for a humanoid robot hand. In: 2012 12th IEEE-RAS International Conference on Humanoid Robots (Humanoids 2012), Osaka, Japan: IEEE,, 2012:7–14.

29.

Santina

, Piazza

, Grioli

, et al. Toward dexterous manipulation with augmented adaptive synergies: the Pisa/IIT SoftHand 2. IEEE Trans Robot, 2018; 34:1141–1156.

30.

Simeonov

, Henderson

, Lan

, et al. Bundled super-coiled polymer artificial muscles: design, characterization, and modeling. IEEE Robot Autom Lett, 2018; 3:1671–1678.

31.

Sharafi

, Li

. A mutliscale approach for modeling actuation response of polymeric artificial muscle. Soft Matter, 2015; 12:1–18.

32.

Yang

, Li

. A top-down multi-scale modeling for actuation response of polymeric artificial muscles. J Mech Phys Solids, 2016; 92:237–259.

33.

, Zheng

. A modeling of twisted and coiled polymer artificial muscles based on elastic rod theory. Actuators, 2020; 9:25.

34.

Swartz

, Higueras-Ruiz

, Shafer

, et al. Experimental characterization and model predictions for twisted polymer actuators in free torsion. Smart Mater Struct, 2018; 27:1–12.

35.

Higueras-Ruiz

, Feigenbaum

, Shafer

. Moisture's significant impact on twisted polymer actuation. Smart Mater Struct, 2020; 29:125009.

36.

Higueras-Ruiz

, Center

, Feigenbaum

, et al. Finite element analysis of straight twisted polymer actuators using precursor properties. Smart Mater Struct, 2020; 30:025005.

37.

Yang

, Cho

, Kim

, et al. Soft fabric actuator for robotic applications. In: 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Madrid, Spain: IEEE, 2018:5451–5456.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

44.28 MB

0.00 MB