SMARCOS: Off-the-Shelf Smart Compliant Actuators for Human–Robot Applications

Abstract

With the growing popularity of Human–Robot Interactions, a series of robotic assistive devices have been created over the last decades. However, due to the lack of easily integrable resources, the development of these custom made devices turns out to be long and expensive. Therefore, the SMARCOS, a novel off-the-shelf Smart Variable Stiffness Actuator for human-centered robotic applications is proposed in this paper. This modular actuator combines compliant elements and sensors as well as low-level controller and high-bandwidth communication. The characterisation of the actuator is presented in this manuscript, followed by two use-cases wherein the benefits of such technology can be truly exploited. The actuator provides a lightweight design that can serve as the building blocks to facilitate the development of robotic applications.

1. Introduction

Human–Robot Interactions (HRI) is a fast-growing field in the world of robotics [1]. The potential of assistive devices covers a wide range of applications, such as exoskeletons, rehabilitation devices or prostheses. One of the main challenges of this emerging technology is the need to re-think the actuator design. Indeed, where industrial robots require precision, stiffness and repeatability, robotic-assistive devices put the focus on safety, transparency and good force/torque control [2]. Hence, a new type of actuator needs to be developed to allow humans and robots to work together: the Variable Stiffness Actuator (VSA) [3]. Examples of such actuators developed over the last decade are ARES [4], ARES XL [5], AwAS [6], AwAS-II [7], BAFSA [8], BAVS [9], CompAct-VSA [10], FAS [11], FSJ [12], MACCEPA 2.0 [13], MIRAD [14], SVSA [15], VSA-Cube [16] and VSA-HD [17].

More specifically, for rehabilitation purposes, VSAs offer the possibility to adapt the human-robot interaction stiffness for several applications, e.g., different stages of a rehabilitation protocol. During the early stages, VSAs can offer a high interaction stiffness such that the patient is heavily supported by the robot that is in charge of completing the rehabilitation task (“robot-in-charge”); in later stages, the interaction stiffness can be slowly decreased, thus allowing the user to take control over the robot and complete the rehabilitation task he/she has been assigned with (“patient-in-charge”) [18,19]. Therefore, VSAs are already implemented in several rehabilitation devices [8,14,20,21].

Human-centered robotics has the potential to counter several societal challenges, such as the ageing of the population or the rising of healthcare costs. However, at the moment, the number of these devices available on the market is quite limited, meaning that very few people have access to these technologies. The lack of easily integrable resources results in a high cost and a long time-to-market development for rehabilitation and assistive devices [22]. These challenges can be addressed by the use of smart actuators, which are able to perform as independent actuation modules. This type of actuator only needs to be provided with power and a communication channel, whereas the motor, transmission, sensors, low-level controller and drivers are all included in the module. Their modularity offers the possibility to reduce the price and development time of custom devices, such as rehabilitation equipment [23], robotic systems [16] or walking robots [24].

The idea of the Robotics and MultiBody Mechanics Research Group (R&MM) is to merge the technologies of smart actuators and VSAs to make Smart Modular Actuators for Robotic Compliant Systems (SMARCOS). As the name suggests, this modular actuator combines compliant elements and sensors as well as low-level controllers and high-bandwidth communication. The goal of this paper is to investigate if a novel variable stiffness actuator, for which the scope is stretched to a variety of different use-cases, can be a suitable candidate as an off-the-shelf solution to ease the development of new HRI devices.

This paper starts in Section 2 with a description of the actuator (in terms of design, sensors and electronics) and the procedures used to benchmark it. The results of the characterisation are shown in Section 3, as well as two robotic assistive devices realised with SMARCOS. A discussion of the results is conducted in Section 4, and finally, a conclusion and future research directions are presented in Section 5.

2. Materials and Methods

2.1. Mechanical Design

The SMARCOS is based on the concept of the Mechanically Adjustable Compliance and Controllable Equilibrium Position Actuator (MACCEPA). This type of actuator acts as a torsion spring able to independently control both its stiffness and equilibrium position. The working principle of this concept is illustrated in Figure 1, and more information about it can be found in [25].

The MACCEPA uses three links, namely the input arm (which is usually fixed), output arm and lever arm, all rotating around the same joint. The input link is connected to the lever arm through the motor-spindle mechanism, and a spring is attached in series between the lever arm and the output link through a routed-cable mechanism. The equilibrium position of the actuator is obtained when the output arm is aligned with the lever arm, namely when the deflection angle $\alpha$ is zero. This position can be adjusted by means of the spindle drive, which actuates the lever arm and therefore creates a deflection. As a result, the spring is compressed, and torque is generated around the rotation joint. This torque brings the output arm back to its new equilibrium position, aligned with the lever arm ($\alpha =0$). The joint stiffness of the actuator can be adjusted by changing the initial state of the spring P through a pre-compression mechanism. The deflection-torque characteristic of the MACCEPA is given by [25]:

$$\tau =f_{MACCEPA}(\alpha ,P)=K·B·C·sin\left(\alpha \right)·\left(1+\frac{P-|C-B|}{A\left(\alpha \right)}\right)with A\left(\alpha \right)=\sqrt{B^2+C^2-2·B·C·cos\left(\alpha \right)},where:$$

(1)

• $A\left(\alpha \right)$ is the length between the connecting points of the lever arm and output arm;
• B is the length of the lever arm (fixed);
• C is the length of the output arm (fixed);
• K is the spring stiffness (fixed);
• P is the spring pre-compression.

The developed SMARCOS is shown in Figure 2. The dimensions and parameters are detailed in

Table 1. Main design parameters of the SMARCOS.

Parameter	Value
Spring Stiffness K	118.2 N/mm
Length ($\phi =0$°)	276.5 mm
Lever Arm Length B	56 mm
Output Link Length C	64.5 mm
Width	60 mm
Weight	1.4 kg

.

Compared to the principle schematics, each link is composed of two bodies rigidly attached to each other. The motor-spindle (Maxon Motor EC-i 40, 70 W, 36 V with Maxon Spindle Drive GP32S 1:1 Ball-screw and the spindle nut were hinged to their respective axes to allow rotation. To be as compact as possible, the spring (SODEMANN ISO R204-505, K = 118.2 N/mm), was placed between the two lever arms. Similarly, the encoders used to measure the angular positions of the output arm and lever arm are located in the spacer between the two input arms, hence the name Smart Spacer (cf. Section 2.2.1). Next to the encoders, strain gauges were also glued to the part holding the motor to measure the axial force acting on the spindle and hence provide an estimation of the output torque (cf. Section 2.2.2). The connection between the lever arm and the output happens through a Dyneema rope (∅1.5 mm), fixed at the spring carrier (lever arm side) and the output connector (output side), as illustrated in Figure 3.

This rope offers a high load capability (break-load: 225 kg) for a low stretching, which is important to have a good torque transmission efficiency. The undeformed length of the rope can be adjusted by two pre-tension mechanisms, which allow for the spring pre-compression. On the output side, the connection to the environment is done by clamping a ∅28 mm cylinder. On the input side, this connection can be realised in the same way or by screwing an M10 bolt in the input connector, depending on the application.

Since the main purpose of this actuator focuses on HRI, the safety of the user is critical. Its intrinsic compliance makes the SMARCOS a safer choice compared to rigid actuators. Moreover, the choice to not gear down the motor before the ball-screw (1:1 connection) ensures low reflected inertia, which, combined with high efficiency of the ball-screw (94%), makes the actuator highly back-drivable. An attempt to estimate the back-driving torque of the actuator using the strain gauges is provided in Section 2.2.2. Furthermore, a telescopic cover has been developed to avoid interactions between the user and the spindle screw, as shown in Figure 4. Finally, physical hard stops ensure that the actuator does not exceed its operation range. Indeed, a first hard stop blocks the motor rotation at $\phi =0$° (cf. Figure 2a), to prevent the lever arm from running in the spindle screw and to avoid any collision between the spindle nut and the motor-spindle combination. A second hard stop prevents the lever arm from exceeding its maximum angular position of 104° (cf. Figure 4a).

The black cover is composed of five pieces, namely two fixed to the input arms, two attached to the output arms and one shutter, which slides between the latter. This shutter is moved by the output arm cover, which makes it extend (cf. Figure 4a) or retract (cf. Figure 4b) when needed, to ensure that the spindle screw remains inaccessible. Additionally, the cover also allows the electronics to be contained in a closed envelope, protecting the cables.

2.2. Sensors

2.2.1. Encoders

The lever arm and output arm angles are measured by two magnetic absolute encoders (AMS AS5048A, SPI type, 6 pins, 5 V, 14 bits, 0.0219 deg/c). To achieve a compact assembly, these encoders were placed at the inside of the actuator, as shown in Figure 5.

The Smart Spacer (1), containing the encoders (2), is fixed at the input link (3), which is the fixed link and therefore serves as a reference. To avoid magnetic interference between the two sensors, the latter ones are separated by a steel plate (4) kept in place by a setscrew. The central shaft is split in two in order to be able to independently measure the lever arm and output arm positions. Each encoder measures the rotation of a shaft (5) by means of a magnet (6) fitted into it. A cam (7) is used to rigidly couple the rotation of each shaft with the rotation of the link, whose position must be measured. Therefore, on one side, the cam is located between the shaft and the output link (8), whereas on the other side, it is situated between the shaft and the lever arm (9). Ball bearings are placed at the other contact points to guarantee a smooth rotation between the links and the shafts.

2.2.2. Strain Gauges

The axial force exerted on the spindle nut is measured by strain gauges (Micro-Measurements WA-13-125BZ-350) glued on the motor hinge part. These linear strain gauges are placed back-to-back to form a Wheatstone bridge to measure bending strain symmetrically with all the gauges [26]. These gauges can provide a reliable estimation of the torque generated at the output of the actuator, as shown in Figure 6.

The calibration of the strain gauges was realised in the test setup, presented in Section 2.4. A sine position target was generated, while measuring the output torque as well as the output voltage of the strain gauges, over a total of ten cycles. A simple linear model regression was then applied as shown in Figure 6. The residuals vs. fits distribution was analysed to evaluate the quality of the model, as displayed in Figure 6. The symmetrical distribution around the origin and the absence of high value points suggest a good validity of the model.

Besides the strain gauges, the current drawn by the motor can be measured using the ESCON drive and used to estimate the torque delivered at the output. A second position sine wave was generated for different amplitudes, while measuring the output torque, the strain gauges and also the motor current. The different estimations of torque are obtained and compared in Figure 7.

As expected from Figure 6, the torque estimation based on the strain gauges gives a more reliable result compared to the one obtained from the current measurement. Indeed, the latter overestimates the torque delivered at the output. This is due to the fact that all the losses occurring inside the actuator are neglected by this model, which would otherwise reduce this output value. The average of the quotient of the mechanical torque measured at the output and the one estimated from the input current give an idea about the efficiency of the actuator, which, in this case, is about 78.3%. Nevertheless, the shape of this model is similar to the one measured, meaning that it can be scaled by applying a linear regression to it, to have a better fitting with the output torque. In the end, the adjusted curve overlaps with the ones obtained by the torque sensor and by the strain gauges. According to these results, the models provide a redundant way to estimate the torque delivered without the need to have a dedicated torque sensor, which would be expensive and cumbersome.

The strain gauges can also be used to estimate the back-driving torque of the actuator. For this, the output arm is no longer mechanically grounded, so that it can move freely. The output arm is then actuated by hand to back-drive the disabled motor while the torque is measured by the strain gauges. The results of this experiment, realised at three different speeds, can be seen in

Table 2. Peak values of the back-driving torque observed by moving the output arm by hand while measuring the strain gauges for three different speed ranges.

Speed Range [rpm]	Peak Back-Driving Torque [Nm]
	CW	CCW
[0;5]	1.43	1.60
[5;10]	1.88	1.83
[10;20]	2.32	2.65

. Overall, these values are low for human-robot applications, considering the knee joint of a healthy 75 kg man generates around 30 Nm while walking [27], whereas a shoulder joint generates around 20 Nm for overhead working [28]. Note that these values are obtained with a disabled motor, meaning that they could be further reduced by using the motor and applying a zero-torque controller.

2.3. Electronics

To operate as an independent module, the SMARCOS contains embedded electronics which allow for:

• sensors interfacing, data acquisition and processing;
• low-level control of the motor;
• data communication with the master computer.

The market offers a wide variety of sensors that require different protocols to acquire and process data. Both hardware and software interfacing are necessary for the integration of these sensors. Therefore, the low-level control unit of SMARCOS offers several input/output and communication ports, as listed in Figure 8.

The Low-Level Controller (LLC) is also able to power and control the motor of the actuator by using the add-on Maxon ESCON 50/5 motor drive. For this, the requested trajectory has to be generated by the master computer, which provides the setpoints to the LLC to operate the actuator accordingly. This communication is realised by real-time protocols through EtherCAT.

The LLC does not only allow the synergy of the encoders, sensors, motors and higher-level controllers, but it also serves as a safety mechanism at the joint level [29]. Indeed, several parameters are monitored to guarantee functional and safe operations, such as joint limit positions, motor current or power stage temperature. This allows for a faster and more accurate reaction time.

2.4. Test Bench for Characterisation

With the aim to provide a standardised characterisation of the SMARCOS, the experimental procedures follow the work of [30]. The characterisation of the actuator is therefore composed of two experiments:

• quasi-static torque-deflection cycles (for several spring precompressions);
• position step response (for several spring precompressions).

The first one consists of a series of loading-unloading cycles for different spring stiffness presets, in this case from 10% to 90% in steps of 20%. This implies applying a known position profile to the motor, while measuring the deflection angle ($\alpha$), as well as the output torque ($\tau$). These tests are realised in quasi-static conditions, which means that the output arm is grounded in a certain position. The latter is chosen to be ($\vartheta =52$°), which corresponds to the central position of the spindle nut on the ball-screw, therefore offering a symmetrical behaviour for positive and negative deflection. The actuator parameters obtained from these experiments are the torque-deflection curves (see Figure 10), which can be compared to the theoretical model given in Equation (1), and the stiffness-deflection curves (see Figure 11) obtained through numerical differentiation.

The aim of the second test is to estimate the maximum speed of the actuator in no-load conditions. Since the actuator cannot provide continuous rotation, a step response is used instead of a quasi-static rotation regime. The experiment therefore consists of a position step response, over the whole range of motion, thus from hard stop 1 to hard stop 2, and back. This is again realised for different pre-compressions, the same as for the first test, as shown in Figure 12. The maximum speed of the actuator can be obtained from the results by numerical derivation.

Both experiments are realised in the test bench presented in Figure 9. The setup consists of:

• an aluminium cage, with mechanical grounding part;
• an ETH Messtechnik DRBK-200 Nm (0.0122 Nm/c) torque sensor;
• a Maxon EPOS4 70/10 Motor Drive;
• a Beckhoff EK1400 EtherCAT G Coupler;
• a Beckhoff EL3101 EtherCAT Terminal, 1-channel analog input;
• a TDK-Lambda 0–60 V, 0–10 A Power Supply.

The Maxon EPOS4 motor drive was used to control the motor in position. The different modules communicate to each others through a Matlab real-time target at 1 kHz with EtherCAT real-time communication protocol in the Beckhoff TwinCAT environment [31].

3. Results

3.1. Quasi-Static Characterisation

As mentioned earlier, the first series of tests consist of a quasi-static benchmarking for several spring pre-compression levels. To ensure repeatable spring compression levels along the tests, 3D-printed parts are used as length reference during the tensioning of the rope. The output arm is mechanically grounded, whereas the lever arm is actuated by the motor, which follows a sine position target. The Maxon EPOS4 drives the motor, using the Cyclic Synchronous Position (CSP) mode via EtherCAT. The parameters measured during the experiments are the output torque and the deflection, which is the relative position of the lever arm with respect to the output one (fixed in this case). For each pre-compression, a series of ten loading-unloading cycles were performed. The resulting torque-deflection characteristics are shown in Figure 10.

The experimental data follows the theoretical model of the MACCEPA given by Equation (1), but contain a hysteresis, which is not present in the model. The fact that every curve ends up around the same torque level, namely 30 Nm, is due to the power supply used, which can only deliver a maximum of 10 A. Without this saturation effect, the curves at lower pre-compression can reach a higher torque, since they allow for more deflection.

The stiffness-deflection curves can be obtained through numerical differentiation of the torque-deflection characteristics, as shown in Figure 11. These can be used to have an experimental estimation of the pre-compression level of the spring. Indeed, pre-tensioning the rope to the right pre-compression level is not always reliable, since this tension could be released over time, without this the user could notice it (since the spring is hidden in the cover). This could be observed by a decrease of the quasi-stiffness of the actuator at no deflection.

The vertical lines at maximal deflection are results of the hysteresis. Figure 10 shows that during the transition from the loading phase to the unloading one, the torque decreases, while the deflection stays temporarily constant, leading to an infinite slope in the stiffness curves.

3.2. Step Response

The second set of tests consists of position step responses over the whole range of motion in no-load conditions. Since the experiments do not require measuring the output torque, the torque sensor, connected to the ground in the previous test, is simply removed from the setup to allow the actuator to move freely. Again, the step response is repeated for different pre-compression levels, again from 10% to 90% in steps of 20%, as shown in Figure 12.

The stiffness of the spring is such that almost all the curves are matching, except the one at 10% pre-compression, which is a bit lower than the others. This is due to the fact that the lower pre-compression of the spring allows for more compliance between the output arm and the lever arm. Furthermore, since the lever arm is already at the end of its range of motion, it cannot go beyond to compensate for the error on the output arm. Therefore, the maximum speed, obtained through numerical derivation, is similar for every curve and is situated around 500°/s.

3.3. Case Study

The following sections present two use-cases of human-robot collaboration systems: a planar upper-body rehabilitation end-effector and an assistive knee orthosis. The development time of these two prototypes has been drastically reduced thanks to the modularity of the SMARCOS.

3.3.1. Arm Rehabilitation Device

A standard planar upper body rehabilitation device, such as in [32], is pictured in Figure 13. This type of device is often used for post-stroke rehabilitation therapy. The prototype is composed of two links, each of them actuated by a SMARCOS. The two links are connected thanks to standard compression pipe couplers, which ease the assembly and disassembly of the whole. Once disassembled, the device can easily be transported by the user for home-based rehabilitation.

Once assembled, the device can be clamped to a table at a suitable height for the user. The actuators are linked in series with a computer through an EtherCAT connection for the data acquisition. A game interface has also been created to make the rehabilitation sessions more fun for the user than a series of repetitive tasks, as shown in Figure 14. The goal of this game is not only to entertain the patients, but also showed benefits in motor recovery, since the users are more engaged during therapy [33].

The patient can choose among four different gaming scenarios to avoid monotony during the exercise sessions. The objectives of the game are highly adaptable to meet the needs of the user at the different phases of revalidation, as shown in Figure 15. Therefore, the patient can start with basic horizontal or vertical motion training, before moving on to advanced stars and circle patterns. Later on, more complex exercises can be proposed by adding an extra time constraint or by reducing the size of the objects.

Once the different parameters have been chosen, the revalidation session can start. Therefore, the cascade controller shown in Figure 16 is applied to the actuator. The torque trajectory generated by the master computer is converted by a PID controller into velocity setpoints sent to the LLC. In the inner loop, the Maxon ESCON add-on module interfaced with the SMARCOS LLC uses this velocity reference to operated the actuator. The torque generated at the output is then measured by the strain gauges and used as a feedback signal for the outer loop.

Different control strategies can be applied depending on the torque reference requested. For example, for sessions in which the patient is in charge, a zero torque controller can be applied to follow the motion of the user with low interaction, as shown in Figure 17. This allows tracking of the performance of the patient, while he/she is following the trajectories requested by the gaming software.

In this case, the patient can actuate the output arm with very low resistance from the actuator. Indeed, compared to the values obtained in Table 2, the zero-torque controller helps the user to counteract the back-driving torque. However, in some other cases, it can be beneficial for the patient to experience some resistance in order to have a more adequate training of the impaired muscles. This can easily be realised by using a force-field controller set to provide a certain level of torque, as shown in Figure 18.

In this case, the actuator provides resistive torque of approximately 2 Nm, while the user is operating the output arm. Instead of a constant level of resistance, an impedance controller can also be used in order to create an angle-dependent force field. Such a controller simulates the behaviour of a spring of which the equilibrium position and stiffness can be set by design choices. Figure 19 shows an example of impedance control acting like a spring of 0.4 Nm/deg around the $\vartheta =50$° position.

These simple controllers can be used as a basis to create more elaborate high-level controllers. A video of the experiments realised with this prototype is available at: https://www.youtube.com/watch?v=2bqL_9PlWv4, accessed on 20 October 2021.

3.3.2. Active Knee Orthosis

Wearable devices to assist people suffering from physical disabilities are currently of great interest. Therefore, a prototype of an assistive robotic orthosis has been developed with the SMARCOS, as shown in Figure 20. The prototype is still at an early stage of development, hence no results can be presented at the moment.

A video of the experiments realised with this prototype is available at: https://www.youtube.com/watch?v=ZY97ss9LiM8, accessed on 20 October 2021.

4. Discussion

The quasi-static characterisation showed that the actuator follows the theoretical model but is, nevertheless, slightly diverging from it due to hysteresis. For the experimental testing of the actuator (e.g., as knee orthosis), it is crucial to know the torque delivered by the actuator to the user at any time. The use of an external torque sensor is not possible in this case, but the torque can be estimated by measuring the deflection angle and pre-compression level of the spring, as well as through the embedded electronics of the actuator. This provides an additional safety layer to the system. Moreover, in the case that the hysteresis is too important for some applications, it is still possible to use the strain gauges to estimate the output torque.

Furthermore, experimental data regarding torque and impedance controllers have been presented. The performance of these controllers can be optimised but their purpose was to validate the use-case. The compliant element included in the actuator can have some negative impacts on its control. The elastic element causes a reduction in bandwidth compared to directly-driven actuators, which can increase the complexity of the control problem. However, the compliance introduced by these elements is crucial for the safety of the user for HRI.

Comparison among VSA Systems

The SMARCOS was designed as a modular compliant SVSA, which can be used for a wide range of human-robot applications. It proposes a wide active range of motion of 104°, with an extra 20° passive deflection at each side. The actuator weighs 1.4 kg on its own and a total of 1.7 kg with cover and electronics included. It can provide a peak torque of 30 Nm, and the output link can reach a maximum speed of 83 rpm. These specifications, summarized in

Table 3. Specifications of the SMARCOS.

Parameter	Value
Peak Torque	30 Nm
Max. Output Speed	83 rpm
Active RoM	0–104°
Weight (MACCEPA)	1.4 kg
Weight (SMARCOS)	1.7 kg

, were compared to the state-of-the-art Variable Stiffness Actuators in

Table 4. Comparison between the specifications of SMARCOS and state-of-the-art VSAs.

Actuator	Ref.	Weight	Peak Torque	Width	I/O
		[kg]	[Nm]	[mm]	Protocol
ARES	[4]	1.1	76	70	CAN
ARES XL	[5]	1.2	76	70	I²C
AwAS	[6]	1.8	80	130	Ethernet
AwAS-II	[7]	1.1	80	140	Ethernet
BAFSA	[8]	2.96	72	95	-
BAVS	[9]	0.75	8	56	Spacewire
CompAct-VSA	[10]	1.8	117	-	-
FAS	[11]	3.9	4.9	123.34	Biss
FSJ	[12]	1.4	67	118.55	Spacewire
MACCEPA 2.0	[13]	2.4	70	-	EtherCAT
MIRAD	[14]	1.4	15	112	EtherCAT
SVSA	[15]	2.4	25	85	EtherCAT
VSA-Cube	[16]	0.26	3	58.5	I²C
VSA-HD	[17]	1.7	14	142	CAN
SMARCOS		1.7 (1.4)	30 ${}^a$	60	EtherCAT

${}^a$ Obtained at maximum current (10 A) of the available power supply.

.

First of all, concerning the torque, the value measured was the one obtained at saturation of the current. The peak torque of the SMARCOS can easily be increased by exchanging its GP32S 1:1 spindle drive for a 3.7:1, similarly to what was done in the work of [21]. This could increase the peak torque to around 70 Nm, while keeping the same mass (according to Maxon’s catalogue). This would bring SMARCOS up in the ranking slightly behind BAFSA, ARES, ARES XL, AwAS, AwAS-II and CompAct-VSA, which is far in front of every other VSA. Concerning the weight, if we consider the case without cover, since the majority of these actuators do not have embedded electronics, SMARCOS only loses to ARES, ARES XL and AwAS-II when looking at the same torque level. VSA-Cube and BAVS weigh much less but also provide a much smaller torque. For the width, only VSA-Cube and BAVS have a lower footprint but again, for a much smaller level of torque, whereas, AwAS and AwAS-II, which provided a bit more torque, are now far behind with double the width of SMARCOS.

5. Conclusions and Future Works

The SMARCOS presented is a smart variable stiffness actuator proposed as an off-the-shelf solution for the development of human-robot interaction devices. It constitutes a powerful yet compact module that encloses all necessary electronics, sensors and low-level controller. Its EtherCAT compatibility offers an easy way of relocating, adding or removing nodes thanks to the intrinsic modularity and scalability of this protocol. The actuator was not designed to be an optimised solution to a specific application, but on the other hand to be a modular and versatile piece of equipment able to adapt itself to a wide range of requirement. The adaptability of the SMARCOS has been shown by means of two different application scenarios.

Future works include the conduct of pilot studies with subjects using the two prototypes presented. This will allow us to further test the SMARCOS capabilities and to gain knowledge about the aspects of the actuator which can still be improved. In parallel, a cost analysis study will be realised for the production of large quantities of actuators.