Mechanical Efficiency Investigation of an Ankle-Assisted Robot for Human Walking With a Backpack-Load

Abstract

The purpose of this work is to investigate the efficiency of wearable assistive devices under different load-carriage walking. We designed an experimental platform with a lightweight ankle-assisted robot. Eight subjects were tested in three experimental conditions: free walk with load (FWL), power-off with load (POFL), and power-on with load for different levels of force at a walking speed of 3.6 km/h. We recorded the metabolic expenditure and kinematics of the subjects under three levels of load-carried (10%, 20%, and 30% of body mass). We define the critical force, where at a certain load, the robot inputs a certain force to the human body, and with the assistance of this force, the positive effect of the robot on the human body exactly compensates for the negative effect. The critical forces from the fit of the assistive force and metabolic cost curves were 130 N, 160 N, and 215 N at three different load levels. The intrinsic weight of our device increases mechanical work at the ankle as the load weight rises with 2.08 J, 2.43 J, and 2.73 J for one leg during a gait cycle. With weight bearing increasing, the ratio of the mechanical work input by the robot to the mechanical work output by the weight of the device decreases (from 0.904 to 0.717 and 0.513), verifying that the walking assistance efficiency of such devices decreases as the weight rises.

1 Introduction

Walking with heavy loads plays an important role in the military [1,2] and our daily life [3,4]. Compared with walking without any load, walking with some load has a greater impact on the metabolic expenditure of human body. When humans walk at a speed of 1.25 m/s, mass added at the foot has an increasing metabolic cost of 7.5–8.5% per kg (kilogram) during walking, and it is 1–2% for mass added at the waist [5]. In addition, weight-bearing walking usually causes muscle fatigue, and metabolism will rise as a result [6,7]. In fact, the increase in metabolism over a short time will increase the risk of human falls and muscle strains [8], and the efficiency of work will decrease. Beyond that, long-term weight-bearing walking could cause damage to normal body movement functions [9].

In recent years, much works have focused on robots assisting human walking with load carriage. The exoskeleton is the earliest and most acceptable type of assist robot. The Berkeley lower extreme exoskeleton [10] developed by Berkeley University and factory walking assisting devices [11] developed by Honda are typical military and civil exoskeletons, respectively. Such devices could reduce muscle fatigue during load-carriage walking indeed. However, the exoskeleton, due to the hard structure itself, would greatly interfere with the users' natural gait, thus, resulting in some limitations on the research of lowering metabolic cost at a much smaller energy cost [12–15]. Subsequently, several excellent studies began to focus on assisting single or double lower limb joints with a flexible and wearable device.

Considerable research efforts have been devoted to the improvement of metabolic consumption with the assistance of these wearable devices during walking. An interesting approach in the use of wearable robots for walking assisting was described by Walson et al. in 2007. The robot had no motor, only ankle, hip springs, and a knee variable-damper [16]. Although the structure and control method of the robot was modified, the metabolic consumption had not been reduced significantly. In fact, many devices have been shown to have no significant effect and may even increase metabolism [17]. Nevertheless, these studies have positive guiding significance for subsequent research. For example, during walking, the ankle provides approximately 35% mechanical work but only accounts for 19% of metabolic consumption [18,19]. In view of such an important effect of the ankle, researchers have begun paying more attention to robot assistance with the ankle, and lightweight design has become a trend [20].

Gradually, some excellent studies have reported a positive decline in metabolic cost. In 2015, Massachusetts Institute of Technology (MIT) designed a no-joint mobile robot that assisted the plantar flexion of the ankle with its control and transmission mechanism integrated into a small box installed on the shank. The results showed that during subjects' walking with a 23 kg load, the device could provide an average power of 11.5 W for each ankle joint, resulting in a net metabolic cost reduction of approximately 8±3%. In contrast, the robot can provide an average power of 13.5 W under the condition of no-load walking assist, and the net metabolic cost reduction was 11±4% [21,22]. In 2014, Asbeck at Harvard designed a wearable walking assist robot named “exosuit” that helped humans move the ankle and hip to reduce metabolic expenditure. They made use of the transmission characteristics of Bowden cable, instead of using a traditional mechanical transmission structure. The ankle could be provided with an auxiliary force at the peak up to 200 N. It is worth mentioning that the part worn on the leg was much lighter, and this wearable device's weight distribution aimed to avoid the impact of the robot weight on the human metabolic cost as much as possible. However, the experimental results showed a 9.3% increase in metabolic expenditure. In the subsequent study, the auxiliary force increased to 330 N by improving the tension preload. The results showed that, when walking with the load of 30% mass of human body weight, the metabolic cost was reduced by 7.3±5% with the assistance of an exosuit [23,24]. Furthermore, in 2017, the Harvard team achieved a 22.8%±3.2% reduction in metabolic cost during walking by placing a very heavy power source and controller outside the body [25]. Such an effect confirmed the great potential of flexible wearable devices for walking assistance. Additionally, their research achieved good results in the field of rehabilitation [26].

Previous studies have confirmed that the mechanism of such robots to assist the human walking is to replace part of the muscle's function rather than augment its ability [27]. This gave us an inspiration for our research, whether the autonomous robot from MIT or the exosuit from Harvard, their successful experiment indicates that the performance of reducing metabolic cost was better in assisted load-carriage walking as opposed to no load-carriage conditions. Furthermore, this seems to involve a question of efficiency, because the primary goal of achieving a reduction in metabolic consumption is to counteract the effect of the constant weight of the device on human walking. The results in these research studies seem to suggest that assistance during weight-bearing can be considered a less efficient situation. Also, it is obvious that the auxiliary force increased, and with no consideration of energy, the performance of metabolic consumption reduction improved as a result [25]. However, when considering the power of portable devices in the field, the efficiency of the auxiliary becomes very important.

Therefore, it is of great value to explore the efficiency of lower limb assisted robots for evaluating the ability of robot assistance and guiding subsequent research on similar wearable walking assist devices. To achieve this goal, we designed and manufactured a lightweight flexible and wearable ankle-assisted robot. We creatively introduced the concept of critical forces, where at a certain load, the robot inputs a certain force to the human body, and with the assistance of this force, the positive effect of the robot on the human body exactly compensates for the negative effect. The critical force is used as an entry point to study the relationship between the efficiency of robot-assisted human walking and the size of the load carried by the human body. We hypothesize that the efficiency of robot-assisted human walking decreases as the load carried by the body continues to increase.

2 Methods

2.1 Experimental Platform Description.

During gait transition, the body expends a lot of energy in both transition from the swing to stance or the stance to swing. The forefoot first touches the ground, then the ankle joint starts to roll to reduce the strike, and during mid to late stance, some energy is stored in the Achilles. From initial ground touch of the forefoot, the Achilles tendon of the hindfoot releases energy and participate in plantarflexion, assisting the process of shifting the center of gravity and compensating for the loss of energy in the forefoot. When the hindfoot leaves the ground, the thigh flexors of the hindfoot provide energy to accelerate the limb swing [28].

Therefore, the elasticity of the tendon is very important to the efficiency of walking, and its releasing energy at the right time can reduce the energy loss due to the heel-strike. During walking, the ankle provides approximately 35% mechanical work, but only accounts for 19% of metabolic consumption [18,19]. From this point of view, assisting the rotation of the ankle seems to be a good option for a walking booster device. For our solution, to avoid affecting the function of energy storage and release of the tendon, and try to reduce the impact on the gait, we set the timing of the assistance to be between 40% and 60% of the gait cycle, because this is the stage when the ankle does positive work [29,30].

The experimental platform includes the walking assisting robot, a backpack with load, a treadmill, a portable gas analysis system (K5b2, Cosmed^®, Roma, Italy), and a motion capture (Vicon^®, Oxford Metrics, UK; 250 Hz). In the lower actuating part, two kinds of signal acquisition elements are included: a load cell and a trigger switch. The weight of the device is approximately 5.2 kg, and more than 95% of the weight is placed in the backpack.

As Fig. 1(a) shows, when the robot is on the working condition, the driver (EPOS4 COMPACT 50/15, Maxon, Switzerland) receives control instructions that drive the maxon motor (EC 647694, 150 W, Maxon, Switzerland, 24 V with a 26:1 gear-box) to rotate first, and then the gearbox shaft drives the wire-wheel to rotate to pull the wire rope. Then, the wire rope comes out from the lower end of the Bowden cable sheath, passing through the guide wheel and the limit wheels, in a direction parallel to the bone axis, and the end is connected to the anchor point of the shoe. The load cell (Tecsis F2811, WIKA^®, Germany; 5000 Hz) is mounted between these sections of the wire rope. At this moment, there is an interaction between the robot and human, and the sensor transmitter processed the signal from the load cell and sends it to the signal receiving terminal (terminal bus EL3124, Beckhoff^®, German). Then, the real-time tensile force is displayed on the personal computer screen.

Fig. 1.

(a) Control system. The complete route of the data stream and power flow of the device. Each block represents a function component of the device. Three downward arrows in the leftmost indicate the direction of the power flow from the device, while other arrows indicate the direction of the data stream that returns to the device. (b) The overall structure of the experimental platform.

The overall structure of the experimental platform is shown in Fig. 1(b). The main body of the device is worn on the human body, the backpack mainly contains the controller and load, and the actuating part is worn on the shank. The treadmill was used to maintain a constant walking speed as the human walked on it. The walking assist robot includes backpack above the waist and a lower actuator on the calf. The upper actuator, controller (C6015-0010, Beckhoff, Germany), and load are installed on the backpack together. We set the position of the backpack carried between c1–c6 of the spine, which is conducive for reducing the metabolic consumption of the human body [31]. The lower actuator includes the Bowdon cable sheath, lower wire rope, load cell, modified shoes with an anchor, and a soft shin pad (thermoplastic polyurethanes part). As a result, our equipment achieves the goal of minimizing the weight burden on the legs, only adding 0.14 kg of weight (2.7% of the total mass of the robot; the entire robot weighs 5.2 kg) to each leg. When the force is required to be increased, the input current will be added until the target tension is displayed on the personal computer. During our test, when the subject steps on the treadmill, the trigger switch is pressed, and this moment is taken as the beginning of gait. Our equipment primarily assists the ankle in 40–60% of the gait cycle, which is the phase of the gait cycle when the ankle is doing the most positive work.

We use the wire rope as the transmission medium of power and a flexible sheath of Bowen cable as the guider of transmission. There are many studies on the transmission loss and hysteresis of Bowden cables [32,33]. The functional relationship between the input and output forces is as follows:

$$k_t=\frac{F_{\mathrm{in}}}{F_{\mathrm{out}}}=e^{f\alpha }$$ (1)

$$F_{\mathrm{in}}=\frac{T}{r}$$ (2)

where f is the friction coefficient, α is the wrap angle between two ends, and r is the radius of the wire wheel.

According to the characteristics of the direct current servomotor, the output torque is

$$I=\frac{V-k_v\omega }{R}$$ (3)

$$T=NI-J\dot{\omega }$$ (4)

where I is the motor current, V is the applied voltage, k_v is the voltage constant, R is the resistance, ω and $\dot{\omega }$ are the velocity and acceleration, T is the motor output torque, N is the torque constant, and J is the motor inertia.

Figure 2 illustrates the calculation of the input work of the robot. Then, the applied mechanical power is estimated by

$${l_0}^2=a^2+b^2-2ab\text{cos}{\alpha }_0$$ (5)

$$\Delta l\left(t\right)=l_0-\sqrt{a^2+b^2-2ab\text{cos}\alpha \left(t\right)}$$ (6)

$$P\left(t\right)=\hspace{0.17em}{F}_{\mathrm{out}}\left(t\right)\frac{d\Delta l\left(t\right)}{dt}$$ (7)

$$W_{\mathrm{input}}={\int }_{t_0}^{t_{\mathrm{gait}}}P\left(t\right)dt$$ (8)

Fig. 2.

Dimensional models. The circle represents the ankle joint, a, b represents the two lateral sides, l is the length of the wire rope, and α is the angle formed by the length of the sides at this time, depending on the moment of gait the body is in.

where $l_0$, $\Delta l_0$, a, b, and α represent the length of steel wire, length change of steel wire, the long and short side of a geometric triangle, and the angle between the two sides, respectively, P is the power of the robot, W_input is the input mechanical work, and t_gait is the time of a gait cycle.

The change in the angle can be converted from the ankle motion captured by the Vicon, and we used the duration of a gait cycle as a criterion for evaluating power. In our calculations, we ignored the wearing part with the deformation of the human body limb, because in approximately 200 N under the action of tension, this deformation is small relative to the change in the length of the wire rope.

2.2 Effect of Equipment Weight on Human Load-Bearing Walking.

In the Introduction, we mentioned that the weight of the device on the metabolic impact with an increase of 1 kg at the waist will increase metabolism by 1–2%, for the legs, which will be 4–5 times the magnification effect. Flexible wearable devices have the advantage of high efficiency in the structural quality of the hardware. The main mass of the autonomous assistive robot developed by MIT is approximately 2.5 kg distributed in the position of the lower legs and feet of the human body [21], which is not an optimal structural design approach from the point of view of efficiency. The Harvard-designed flexible wearable exosuit uses the Bowden line drive to transfer power from the back to the lower limbs of the body. A multijoint assistive prototype weighs 6.6 kg with the vast majority of the weight on the back of the body [23]. Therefore, we investigated carrying the weight of the device primarily from the back as a way to improve the efficiency of the assistance.

The quantitative relationships between metabolism and gait, mechanical work, or weight-bearing are inconclusive, and most studies have focused on fitting the mathematical relationships between different variables and metabolism through experimental results [34,35]. In this study, we focus on the variable of mechanical work. The effect of mechanical work of the human body on metabolism satisfies the following mathematical relationship in terms of the mathematical formula:$\hspace{0.17em}\mathrm{AF}=\frac{p^{+}+p^{\mathrm{dis}}}{\mu }-\sum {\beta }_i{m}_i$. Since the weight of our equipment is not very large compared to the weight of the load, the calculation from the fitted results has a large error. Because the power from the device is input into the ankle, for output power, we also study the effect of device weight on the mechanical work of the human body from the perspective of ankle-joint power.

Research and technology on human kinetics modeling have been well established [36], and the inverse kinematic equation can be used to derive the joint torque from the kinematic information and plantar pressure of the human body as it walks. Then, the power of our joint torque can be obtained by integrating the torque with the angular velocity of the joint motion [37]. The effect of the weight of the device on the output of human mechanical work during a gait cycle is shown in the following equation

$$P\left(t\right)=\hspace{0.17em}T\left(t\right)\frac{d\theta \left(t\right)}{dt}$$ (9)

$$W_{\mathrm{ankle}}=\sum _{i=1}^k\left|P\right|·\Delta t$$ (10)

$$\Delta t=\frac{1}{f_{\mathrm{vicon}}}$$ (11)

$$k=t_{\mathrm{gait}}·f_{\mathrm{vicon}}$$ (12)

$$W_{\mathrm{output}}=W_{r-\mathrm{ankle}}-W_{0-\mathrm{ankle}}$$ (13)

where P, θ, T, and W represent the ankle moment, angle, mechanical power, and mechanical work, respectively, f_vicon is the frame number of motion capture, and k is the number of points sampled within a gait cycle. It is worth mentioning that W_output is a variable related to the load carried on the human back.

2.3 Experimental Protocol.

Eight heathy subjects (22.5 ±1.5 years; 170.5±2.5 cm; 63.5±3.5 kg) were invited to participate in the experiment. The length of the calf leg circumference, where the calf pad (made by three-dimensional printing, and the material is Thermoplastic PolyureThanes) were worn, the distance between the axis of the ankle joint and the plantar plane of the foot, and the length of the foot were all recorded for each eligible subject. Figure 3 shows the state of our experiment. There were three experimental states in this study: free walking with load (FWL), power-off with load (POFL), and power-on with load (PONL). In the PONL state, one minute before the assist force was applied, the subject needed to adjust the walking speed, and then, the tension was increased until the target force was reached. We divided the size of target force into three levels: 150 N, 200 N, and 250 N. In each test, we selected one of the three states above, the order of selection was random, and every subject needed to be tested in these three states. After 10 min of rest before each test, the subjects were asked to walk on the treadmill at a speed of 3.6 km/h for 6 min, and the one-minute time used for preparation is not included in the 6-min experiment duration.

Fig. 3.

Experimental condition: (a) side view of the whole experimental plate and (b) back view of the actuator on shank and ankle

People usually increase walking speed by increasing the step length and step frequency. Before the formal experiment, we recorded the subject's natural walking step frequency and used cell phone beat software to imitate the rhythm of this step frequency. By this method, in the formal experiment, we reminded the subjects to accommodate the speed of the treadmill by keeping the step frequency constant, thus keeping the frequency-dependent metabolic cost constant. We divided the weight-bearing into three grades: about 10%, 20%, and 30% of the body mass for every subject. In the subsequent descriptions, we used 10%, 20%, or 30%M to represent the size of the load carried by the subjects. Short-term load-bearing walking would not cause great damage to the joints. We set three levels of auxiliary force 150 N, 200 N, and 250 N, respectively. In any of the tests under three different load-carriage (10%M, 20%M, and 30%M, respectively) level, subjects were asked to experiment with these three different levels of assistance (150 N, 200 N, and 250 N, respectively) in the PONL state.

Metabolic cost during walking was assessed by K5 (K5b2, Cosmed, Roma, Italy). Carbon dioxide and oxygen rate were averaged across the last two minutes of each condition used to calculate metabolic power using a modified Brockway equation [38]. Metabolic power was normalized by the body mass of each participant. The Vicon was used to record the kinematics, and the treadmill (h/p/cosmos, Mercury, FDM-THM, Germany; 120 Hz) was used to record plantar pressure.

Statistical analysis was conducted using SPSS (SPSS Inc., Statistics 21, Chicago, IL). One-way analyses of variance with five group (FWL, POFL, PONL (150 N), PONL (200 N), and PONL (250 N)) were used to verify the effect of the device on metabolic expenditure. Tukey posthoc tests were performed to identify differences between conditions when a statistically significant main effect was identified by analyses of variance. The significance level was set at p < 0.05 for all analyses.

3 Result

3.1 Mechanical Work Input.

There is no quantitative conclusion as to how mechanical work affects human metabolism, so we use the critical force as our measure of input mechanical work. We assume that, when the robot assists the human body to walk, there is a force presenting that the mechanical work it inputs to the human body exactly counteracts the negative effect of the robot's weight on the human body. With the assistance of a force at this magnitude, the human metabolic consumption data for the human body in the PONL and FWL states are the same.

Figure 4 and Table 1 demonstrate that the weight of the device inevitably increases the metabolic expenditure while walking. Additionally, the greater the auxiliary force provided by the device is, the more pronounced the metabolic decline is, regardless of the load under which the body is walking. Average metabolic power during standing with loads of 10%M, 20%M, and 30%M on average 1.12±0.14 W/kg, 1.44±0.21 W/kg, and 1.63±0.13 W/kg, respectively. The results of net metabolic expenditure under three different experimental conditions are shown in Table 1. The results confirm that, on the one hand, the weight of the device increases metabolic consumption; on the other hand, an increase in the force decreases metabolic consumption. Table 2 shows the significance level of metabolic cost between any two groups test conditions. Overall, there were significant differences between a significant portion of the data. At the very least, the conclusion that the robot we designed can have a positive or negative effect on human metabolism is convincing.

Fig. 4.

Metabolic power in FWL, POFL, and PONL at three different levels of load: 10%M, 20%M, and 30%M

Table 1.

Metabolic power for subjects in the experiment

	FWL	POFL	PONL (150N)	PONL (200N)	PONL (250N)
Metabolic power with 10%M load (W/kg)	3.89 ± 0.38	4.12 ± 0.27	3.89 ± 0.31	3.80 ± 0.33	3.62 ± 0.22
Metabolic power with 20%M load (W/kg)	4.20 ± 0.41	4.40 ± 0.13	4.25 ± 0.40	4.20 ± 0.09	3.98 ± 0.30
Metabolic power with 30%M load (W/kg)	4.67 ± 0.64	5.01 ± 0.32	4.85 ± 0.59	4.78 ± 0.41	4.50 ± 0.33

Table 2.

Significance level of metabolic cost between any two groups of test conditions

10%M load-carried	FWL	POFL	PONL (150N)	PONL (200N)	PONL (250N)
FWL		0.0421	p > 0.05	p > 0.05	0.0375
POFL			0.0360	0.0453	0.0240
PONL (150 N)				p > 0.05	0.0127
PONL (200 N)					p > 0.05
PONL (250 N)

20%M load-carried	FWL	POFL	PONL (150N)	PONL (200N)	PONL (250N)
FWL		0.0357	p > 0.05	p > 0.05	0.0217
POFL			0.045	0.0371	0.0415
PONL (150 N)				p > 0.05	0.0330
PONL (200 N)					0.0397
PONL (250 N)

20%M load-carried	FWL	POFL	PONL (150N)	PONL (200N)	PONL (250N)
FWL		0.0316	p > 0.05	0.0304	p > 0.05
POFL			0.0248	0.0317	0.0125
PONL (150 N)				p > 0.05	0.0377
PONL (200 N)					p > 0.05

To further obtain the magnitude of the critical force, we performed a linear fit to the force and metabolic cost curves, as shown in Fig. 5. Every goodness of fit is greater than 0.7. The critical force magnitudes are 130 N, 160 N, and 215 N for the three different load states. According to the calculation of the input mechanical work in Sec. 2.2, combined with the information about the angle of motion of the human ankle in Sec. 3.1, we obtain the mechanical input work of our assistive device for one leg during a single gait cycle in three load states.

Fig. 5.

Fitting formula. The dots represent metabolic cost in this experimental state, the oblique straight line represents the fitted line, and the horizontal dashed line represents metabolic cost in the FWL state. The graphs from top to bottom represent the three load classes.

3.2 Mechanical Work Output.

Figure 6 shows the kinematics of the subjects' ankle in POFL with different load carriages at 10%M, 20%M, and 30%M, and in POFL, a weight of robot at 5.2 kg was added to the human body. The experimental results confirmed that an increase in loading resulted in larger peak angles of ankle-joint dorsiflexion and plantarflexion with 9.1% (1.2 deg) and 18.9% (2.5 deg) increase in the mean peak angle of ankle-joint dorsiflexion at 20%M and 30%M loading, compared to 10%M loading. Because the range of ankle motion is inherently small, angular variation of the ankle motion is not significant in absolute terms; however, the relative variation is still significant.

Fig. 6.

Ankle degree curve in POFL for three different levels of load bearing at 10%M, 20%M, and 30%M

This result confirmed the significance of assisting the ankle for weight-bearing walking. Much energy will be lost during gait phase transition [35,39], and human carrying too much weight on the body can increase the burden on the ankle joint during the gait transition phase, so it is well understood that energy consumption increases at a faster rate. In addition, the ankle angle data from the POFL tests can be used for the calculation of the input mechanical work in Sec. 3.1.

From the change in the ankle power curve, it is clear that weight bearing also increases the power output of the ankle joint. As Fig. 7 illustrates, the average peak power of plantarflexion at 20%M and 30%M of load increased by 10.23% and 25.02%, respectively, compared to 10%M of load. The robot weight resulted in an increase in the average peak ankle power by 0.2380 W, 0.2521 W, and 0.2134 W at three different load levels. We calculate the magnitude of the ankle-joint mechanical work for the increased weight of the device by using formula (9)–(13).

Fig. 7.

Ankle-power curves in FWL and POFL for three different levels of load bearing at 10%M, 20%M, and 30%M. The left side shows the FWL data, and the right side is the POFL data.

These test results demonstrate the effect of weight bearing on the kinematics and kinetics of the human lower limb, which is ultimately reflected in the energy metabolic expenditure, and our analysis verifies the significance of assisting the ankle and the robot's inevitable interference with human walking.

3.3 Mechanical Efficiency.

Because the human body is a very complex system, no research has proven an absolutely correct mathematical relationship between metabolism and mechanical work, except for the mathematical relationship fitted to experimental data. Using the tool of critical force, we can avoid the complex topic of energy consumption by human metabolism and calculate the robot-assisted efficiency from the mechanical work of the input and output. According to the definition of critical force, in the process of human walking to receive the assistance of the critical force of the robot, the mechanical work input by the robot is the work done by the critical force on the ankle, and the mechanical work output is the work done by the robot to offset its own weight on ankle kinetics. On this basis, we analyze the assistance efficiency of the robot.

The experimental results in Table 3 show that efficiency of the robotic assistance decreases as the level of load increases during the human walking process, and more mechanical work is needed to counteract the adverse effects of the weight of the device on the human body.

Table 3.

Mechanical efficiency of the robot at critical force

Load
Result		10%M	20%M	30%M
Critical force (N)		130	160	215
Input work of the critical force (J/gait cycle)		2.30	3.39	5.32
Increased work of ankle (J/gait cycle)		2.08	2.43	2.73
Efficiency		0.904	0.717	0.513

4 Discussion

This study aims to investigate the efficiency of ankle-assisted robots under different loads during walking. In this research, we designed a prototype where we can adjust the weight of the backpack and the amount of assisted force according to the experimental needs. The robot could monitor the real-time tension, and the experimental platform could obtain information about the metabolic cost and human kinematics. We introduce the concept of critical forces, where human walking at a certain load, the robot inputs a certain force to the human body, and with the assistance of this force, the positive effect of the robot on the human body exactly compensates for the negative effect.

We obtained the metabolic cost data under the assisted forces at 0 N, 150 N, 200 N, and 250 N in POFL and PONL tests and obtained the critical force under 10%M, 20%M, and 30%M load-carried by linear fitting method. The topic of robot mechanical efficiency is raised from the point of view of mechanical work, and the critical force makes the calculation of the robot's input mechanical work much easier. Experimental results confirm that the mechanical efficiency of robotic assistance to the human body decreases as the load increases.

In fact, as described in the Introduction, many studies have investigated the effect of increasing weight and robot assistance on metabolic expenditure. For example, Mooney et al. fit an empirical formula by combining data from other literature [22]. The formula is as follows:$\hspace{0.17em}\mathrm{AF}=\frac{p^{+}+p^{\mathrm{dis}}}{\mu }-\sum {\beta }_i{m}_i$. In their study, evaluating the effect of the device on human metabolism is also a common way to explore metabolic cost by fitting the formula through some reviews and experimental results, which cannot truly be quantified [35]. The empirical formula goes some way to explain the effect of robots on human metabolism, both in terms of assistance and weight gain, which is why we try to avoid robots that add weight to the legs. However, one thing that this empirical formula does not take into account is that the same devices, weighting factors for assistance ($\mu$) and weight (${\beta }_i$) can change when the load is different, taking the peculiarities of the human body into account [40]. Therefore, it is not possible to calculate the booster effect and the impact of the load completely separately.

Previous research has suggested that the significance of reducing the weight of a device is to reduce the elevated metabolism resulting from the increased load [5], and if this simple linear superimposed relationship can be satisfied, then the same device with the same assistance should have a nearly identical reduction in metabolic consumption under different loads, which clearly is not the case. Therefore, guiding robot lightweight design from a metabolic reduction perspective is a very important, but we cannot isolate the link with robotics assistance. Our results confirm that the lightweight design of devices becomes very important in terms of mechanical efficiency. The research on flexible wearable lower limb power-assisted robots presents important guidelines in terms of efficiency and structure.

On the other hand, in studying to evaluate the superiority of different devices, our research also provides a new idea. It is naturally good for a device with superior performance to be able to reduce the most metabolic consumption, but one has to consider economic reasons as well. If a device uses less energy to counteract the effects of the device itself on the human body, then the device can be considered superior from an efficiency standpoint.

There are some flaws in our study. The subjects came from our lab group were likely familiar with the device and anticipated study outcomes could significantly impact results. This has a detrimental effect on the generalizability of the experimental results. In future studies, we will recruit more heterogeneous samples and participants from outside of our lab group to prevent bias and improve generalizability of study results. In addition, in our study, it is worth mentioning that the actuator is mounted on the human calf, and there are counteracting forces that cause discomfort in the calf muscles when the device assists during walking. This flaw needs to be improved later. Moreover, we obtain the critical value of the auxiliary force by fitting, which is logically well understood, but for efficiency, we can only analyze the trend, and not calculate it precisely, because it involves human metabolism, and many problems become very complicated [40].

Funding Data

• National Natural Science Foundation of China (Grant No. 51575188; Funder ID: 10.13039/501100001809).

• National Key R&D Program of China (Grant No. 2018YFB1306201; Funder ID: 10.13039/501100013290).

• Research Foundation of Guangdong Province (Grant No.2016A030313492 and 2019A050505001; Funder ID: 10.13039/501100002912).

• Guangzhou Research Foundation (Grant No. 201903010028; Funder ID: 10.13039/501100012241).

Nomenclature

FWL =

free walking with load (experimental condition 1)

POFL =

power-off with load (experimental condition 2)

PONL =

power-on with load (experimental condition 3)

10%M =

10% of each subject's body mass

20%M =

20% of each subject's body mass

30%M =

30% of each subject's body mass