Improving the Assisting Efficiency of Ankle Robot through Energy Harvesting of Achilles Tendon

Background: The construction of lightweight robots poses one of the major challenges in the field of active robots since bearing the weight of an active robot significantly increases metabolic cost. However, few studies have achieved a substantial reduction in the robot weight. The primary reason is that the weight of the actuator, which comprises the main weight of the robot, is limited by the specific power, power requirements and assisting efficiency. Methods: In this paper, we propose a new method that is utilizing the energy harvesting function of the Achilles tendon to improve the assistance efficiency of ankle robots to reduce the weight of the actuator and we design a novel ankle robot to test the validity of the method. The robot works with the ankle plantar flexor at 43%-60% of the gait cycle and has no other effects on the joints or the tendon of the lower limb. Healthy subjects were recruited to test the prototype in three conditions: free walking, power-on walking, and power-off walking. Data on the robot assisting power, metabolic cost and kinematics in different conditions were collected and analyzed. Result: The results showed that the ankle robot can deliver forces at the controlled assistance timing. The average assisting power of 0.0650±0.0054 W/kg per leg resulted in an 8.7±8.1% and 19.0±6.4% net reduction in metabolic cost in power-on walking compared to free-walking and power-off walking, respectively. Conclusion: Compared with the results of some of the best research, our initial result supports the validity of the method. This method can help to reduce the weight of active robots and the technical innovative method to determine the assistance timing more accurately and the novel design of the ankle robot can provide a reference for future research. To the best of our knowledge, this method is the first to use the human physiological structure to optimize the design of active robots.


Background
Humans take more than 10,000 steps per day on average [1], and the metabolic cost of walking is higher than those of other daily activities [2]. Therefore, reducing metabolic cost is of great significance to the preservation of physical strength.
To reduce the metabolic cost of walking, much research has been performed regarding designing and using wearable lower-limb robots to assist the motion of the joints [3]. In general, wearable lower-limb robots can be concluded as devices that use structural parts to connect with the segments of the lower limbs of humans and use a mechanism to actuate the relative motion between the structural parts. According to whether there is 2 an external power source or not, the lower limb robots can be divided into active robots and passive robots [4]. The principle of passive robots is using elastic elements between the segments of the lower limb to imitate or replace the tendon to absorb the negative work and assist the positive work of joints during walking. However, the stiffness of the elastic element needs to be designed strictly according to the subject and motion conditions [5], which limits the practical use of passive robots. Active robots use external controllable power sources such as electricity or pneumatics to actuate robots, which makes active robots adaptable to various requirements in practical applications.
A variety of active robots have been developed in the past two decades. Early robots were designed to assist the multiple joints of the lower limbs in walking [6][7][8].
These robots had a typical rigid structure in which the segments of the robot align with those of the lower limb and the joints of robots are coaxial with those of the lower limb.
However, the heavy weight of the mechanical structure greatly increased the metabolic cost [7] and the inconsistency of the human-machine structure affects the movement of humans and function of the tendon [3].
Some of the later studies fixed the robots on a frame to reduce the burden on humans, which reduced metabolic cost. Philippe and his colleagues designed a robot that assisted the movement of the ankle joint by pneumatic muscles in 2013 and realized 6±2% reduction in the metabolic cost [9]. In 2017, Galle and his colleagues designed a robot with a similar structure. They tested different timings of assistance and trajectories of assisting power and found that assisting the ankle joint at 42% to 60% of the gait with 0.4 W/kg assisting power can achieve an optimal reduction in metabolic cost(21%) [10]. These robots still had the problems of the inconsistency of human-machine structure. In the same year, B.Quinlivan and his colleagues tested a multi-joint soft exosuit and achieved 22.83 ± 3.17% net metabolic cost reduction with an assisting torque of 38% biological ankle moment [11]. The robot was joint-less and did not restrict the joint movement structurally, but the results showed that the dorsiflexion of the ankle decreased greatly, which affected the function of the Achilles tendon. Although these robots reduced the net metabolic cost, most of these robots are only suitable for applications on a treadmill such as the walking training of patients and cannot be used in actual moving conditions [12,13], since they are fixed to a frame.
Recently, some lightweight autonomous robots have been designed to assist joints during walking. In 2014, Alan and his colleagues designed a multi-joint soft exosuit to assist the movements of lower limbs, but the results showed an increase in the metabolic cost [14]. An important reason was that the excessive weight of the robot increased the metabolic cost, and there were also some possible reasons that the fabric was so flexible that it consumed energy from the motors [15], and the negative power of the robot affected the ankle dorsiflexion [16]. In 2015, Luke and his colleagues developed a novel joint-less robot to assist the plantarflexion of the ankle joint [4]. Specifically, in a robot driven by motors, the weight of motors and reducers always constitutes a large proportion of the weight.
However, simply reducing the weight of the motor and reducer causes the robots to fail to meet the power requirements because the weight of the motor and the reducer is related to the specific power.
Considering that the Achilles tendon can store and return 61% of the energy of the ankle joint [20] and that this function has been ignored or affected in previous studies,

Method Prototype design
The prototype is designed to assist the ankle plantarflexion which takes 43% of the positive work [21] and the main forward driving force [16]. The prototype can be divided into three parts by function: controlling part, driving part and actuating part (   Ergonomics is considered throughout the design of the prototype. The weight center of the backload is placed at the approximate level of thoracic vertebrae 1-6, which is the optimal load-bearing position regarding energy [23]. The weight of the backload is limited to less than 10% of the person's bodyweight to reduce the impact of the weight of the backload on kinematics and avoid the appearance of subjective weight-bearing sense and back pain [24]. A waist belt is used to improve the load distribution [25], to reduce the load feeling and improve the perceived stability during walking [26]. To increase the contact surface and reduce the pressure [27], the shape of the calf connector is similar to that of the gastrocnemius muscle. considered the optimal assistance timing in the study [10]. Foot switches [10] or inertial sensors [28] are usually used to judge the gait event which is used to determine the assistance timing with the gait cycle in the traditional method. The

Control method
Studies have shown that greater assisting force leads to lower metabolic cost [11].
However, excessive tension can cause excessive pressure on the calf, affecting the wearing comfort and blood flow [27]. To balance these two situations, we control the EtherCAT) to track the set current value, and the parameters of the PID controller are tuned manually to achieve a quick response (Fig.6).
According to the characteristics of the DC servo motor, the torque of the motor is proportional to the current. The relationship between the torque and the current can be written by the formula: = • where is the torque of the motor; is the torque coefficient of the motor and is the current of the motor. A small initial current value is set and then increases slowly according to the requirements of the subjects. During the robot assistance, the motor output torque is controlled by the above method. After that, the motor quickly resets to the initial position and waits for the next assistance timing. Human-machine interaction peak force limit.
When the human-machine interaction force  Fig.7.  and is the reduction ratio of the reducer.

Metabolic cost
The metabolic cost was measured by a portable gas analysis system (K5b

Statistical methods
SPSS (SPSS Inc., USA) was used to conduct the statistical analysis [33]. A paired-samples t test with the three conditions was used to verify the effect of the device on human walking. p <0.05 was set as the level of a significant difference.

Result Mechanical power
The average peak interaction force was 126±  (Table S2).

Metabolic cost
The average metabolic power during standing was  (Table S3).

Discussion
The initial result supported the idea that utilizing the energy harvesting function of approximately one-fourth of that in study [11] and one-third of that in study [33], but we obtained a similar reduction in metabolic cost.
If the reduction in the metabolic cost in power-on walking compared to that in power-off waking is regarded as the net effect of robot assistance, the 19.0 ± 6.4% reduction in the metabolic cost caused by the 0.0650±0.0054 W/kg per leg assistance in our study is 31% more efficient than the results of the study where 0.4W/kg assisting power cause 21% reduction in metabolic cost [10].
In terms of the amount of the net reduction in metabolic cost, our robot achieved a result similar to those of some of the best research at present. The average net metabolic cost reduction of 8.7±8.1% in power-on walking compared to free walking was slightly more than the 7.3±5.0% in study [33] and slightly 13 less than the 11±4% in study [4] and the 9.3% in the study that designed a hip robot [19].
The average net metabolic cost reduction of 19.0±6.3% in power-on walking compared to power-off walking is slightly less than that of 21% in study [10] and 22.84±3.17% in study [11].
It should be noted that the effect of the assistance of the active robot on the metabolic cost during walking seems to vary from person to person. In this study, there was a "super subject", who showed a large net reduction in metabolic cost with robot assistance compared to all the other subjects.
There are "super subjects" in other studies [4,19], although the variance in other studies is not as large as that in this study. In a sense, The reaction force of the interaction force is borne by the calf, so the peak interaction is limited by the pressure that the calf can withstand. The results showed that the maximum peak interaction force was 161 N, which was far less than those in other studies [10,11]. Since the robot has no joints, the reaction force of the interactive force cannot be transmitted to the ground as in study [10].
If the reaction force can be borne by the waist, the peak interaction force will be larger. A previous study has demonstrated that a large interaction force can lead to a large decrease in metabolic cost [10,11], potentially indicating that there is still much potential for the metabolic cost to decline.
This paper used a small assisting force to achieve a large metabolic cost reduction. One advantage is that the small assisting force has little effect on gait [30]. As far as the current consensus is concerned, it is always good to not change the walking gait. After all, changing the gait will cause an increase in metabolic cost. The other advantage is that smaller motors and reducers can be used because of the improvement in efficiency. It has been reported that a 1 kg reduction in load can lead to a 1% to 2% reduction in metabolic cost [17]. Our assisting principle provides a solution to the lightweight design of active robots.

Future work
In this paper, we only performed a basic verification of the proposed assisting method, which only involves the interaction force, interaction power, metabolic cost, and kinematics, but does not involve the performance of the muscles and tendons which are the lowest and most basic layers.
Future work should explore the performance of muscles and tendons before and after the assistance of a prototype to validate the proposed assisting principle.
The gait analysis system is used to judge the timing of assistance. In the future, we can develop a portable plantar pressure sensor to judge the timing of assistance so that the robot can become a real autonomous robot. Table S1: weight of the prototype. Table S2: mechanical power. Table S3: metabolic cost.