DESIGN AND CONTROL OF COMPLIANT WEARABLE ROBOT MECHANISMS

Transcription

1 DESIGN AND CONTROL OF COMPLIANT WEARABLE ROBOT MECHANISMS Seminar 1 Miha Dežman Supervisor: Andrej Gams Approved by the supervisor: (signature) Study programme: Information and Communication Technologies (Doctoral degree)... Ljubljana, 2016

2 Contents Contents 2 1 Abstract 3 2 Problem definition 4 3 Research overview Compliant actuation Fixed stiffness and variable stiffness actuators MACCEPA DLR - VSJ Joint Exoskeletons Exoskeleton design Exoskeleton control HARMONY exoskeleton ESA exoskeleton Ekso exoskeleton Passive ankle exoskeleton Outlook and future work 19 5 References 21 2

3 1 Abstract In order to successfully operate in everyday human environments and homes, robotics is shifting actuation to a new, biologically inspired type of actuation, which is mechanically compliant. The new compliant actuators consist of a classical electrical motor or a hydraulic system, and a serially connected elastic element. They have low impedance and friction, which enables implementation of high quality force control. These characteristics make them suitable for robots that operate in unstructured environments, such as workshops, hospitals and homes. Their main advantage is that they allow deflection from the equilibrium position when the external load changes. In contrast, conventional stiff actuators allow no deflection from a set position, which is in the ideal case independent of external force. Other advantages are the possibility to store energy, intrinsic shock protection, higher safety, masking of nonlinear effects and possibility of measuring the external force based on the deflection of the elastic element. Such compliant actuators are extremely suitable for actuation of exoskeletons - robotic wearable mechanisms. Because exoskeletons work in parallel with the user and are usually attached with straps and form in essence a stiff interaction between the robot and the user, classical high feedback gain (stiff) actuation and control are not applicable, because the robot can not always adapt to the human movement. Exoskeletons differ from conventional robots in both design and control. Conventional robots are designed and optimised for their intended task. On the other hand, the exoskeletons design is focused on ergonomy, comfort and most notably the safety of the user. Their intended task comes only second to these aspects of the design. Because of a novel actuation type and a high level of human-robot interaction, they also differ in control. For successful control of an exoskeleton, every movement of the human needs to be predicted in order to successfully provide assistance, which provides a considerable challenge. Keywords: exoskeletons, wearable robotics, active orthosis, serial elastic actuator, variable stiffness actuator, compliant actuation, compliant design, compliant control. 3

4 2 Problem definition Robots hold a great potential for the improvement of peoples everyday lives. Nowadays, they are slowly making their way into our everyday. The amount of human-robot interaction is thus growing. The problem of conventional industrial robots is that they cannot simply leave their industrial robot cells and operate in unstructured environment, along humans, or in physical contact with humans. Safety standards, as well as the abilities of most industrial robots, demand that no human is present in their workspace when they operate. Internal and external models of the robots typically do not predict obstacles that were not defined in advance. In case of an unpredicted collision, the robot will not stop or it will stop too late. As they are typically very stiff and heavy, in order to achieve greater position precision, industrial robots can generate very high forces even through a very small motion. Therefore, for successful control, every movement must be planned in advance. All control errors result in the position error of the end effector. If we introduce an unpredictable element, such as a human, who is nonlinear and dynamic, then classical robotic control will not suffice. Compliant behaviour can be simulated as a virtual spring behaviour on stiff actuators, as in conventional active control [1]. However, this approach does not completely solve the problems of conventional robotics, because the robots safety, performance and compliance depend on its control structure, electric system and sensors. In its essence, the robot acting as a virtual spring is still a mechanically rigid object. Some level of mechanical compliance is needed. This is especially true for a special type of robots called exoskeletons. These wearable robotic mechanism or active orthoses are meant to work directly with a human. They are usually connected in parallel with the user using straps and have a wide array of potential applications: assisting humans, returning their mobility, easing certain tasks or improving their skills. Because of this, in essence stiff connection, consisting of sensors and straps, the amount human-robot interaction is therefore very high. This is where and why the exoskeletons requirements differ from conventional industrial robots. The first major difference is in the design and construction. Conventional robots are designed to be rigid and are therefore very heavy, which demands high-power motors that further increase their weight. Because of that, they cannot effectively adapt to unpredicted external stimuli or obstacles and produce high forces of inertia. The risk of injury to mechanical parts, surrounding environment and humans in close proximity, is high. Such a rigid structure and conventional position control is therefore not appropriate for exoskeletons. The robot needs to be designed so, that it can adapt to human movement very fast. This demands a different design approach in comparison with conventional robots. The most important thing for the exoskeleton is the safety and comfort of the user which is the first design goal. This calls for a light-weight design and exoskeleton structure that adapts structurally and mechanically to the human body and its movement. Only after that the robot can be designed for a specific task. On the other hand, conventional industrial robots are normally designed foremost for the task they will do, for example speed or accuracy. Because of these shortcomings of industrial robots and rigid design and actuation, compliant actuation is being introduced to the latest generations of robotic mechanisms, both industrial as well as exoskeletons. The main difference between a stiff and compliant actuator is that a stiff actuator holds its position regardless of the external load, where the compliant actuator allows a deflection from the set position, when external load changes. Compliant actuators are not yet widely used and are therefore commercially 4

5 relatively difficult to obtain. This new technology demands new ways of control that differ from conventional techniques. First the control needs to be designed at the lowest level. We need to control the equilibrium position and the stiffness separately. Second, the design of control at the higher level has to take the first level into consideration, but must also consider task-dependent behaviour. Here, other benefits of the compliant actuator can be exploited, for example force control and energy storage and release, to produce robust control and energy efficiency. Advantages of compliant actuators make them suitable for implementation into exoskeletons, where stiff actuators are not appropriate. Consequently, an interaction on a higher level can be achieved at a smaller impedance, higher transparency and comfort. This makes the exoskeleton easier to adapt and the user more likely to accept it. 5

6 3 Research overview 3.1 Compliant actuation When comparing humans and robots we can notice that apart from an incredible neurological control system, humans differ also in the actuation system. Human body uses actuators that differ from conventional drives, be it an electrical motor or, a hydraulic system or something else, and the difference is not only in the materials and structure but also in the principle itself. An ideal force actuator is able to provide a specific force at a specific position [2]. It has low impedance and is completely backdrivable, has no stiction and and no friction, and endless bandwidth. The human muscle gets very close to the ideal case, since it has a very low impedance and stiction, and a medium range of bandwidth. The human muscles are arranged in an antagonistic configuration because they can exert forces in only one direction. This way the human can not only move a limb, but also change its stiffness. The actuation is thus very compliant. Changing of stiffness has significant advantages in interaction with the environment. The other difference is a high efficiency rate of actuation. The human body uses very little energy to perform walking and other cyclic movements. It uses elastic elements called tendons, which can store energy and improve the efficiency of movements and achieve higher power in extreme cases [3]. Serial elastic actuators are compliant actuators, that come very close to a perfect force source [2]. They consist of a conventional actuator, which is either an electrical motor or a hydraulic system, and an additional serially connected elastic element. The addition of a elastic element has many other benefits: efficient energy storage and release can be achieved in cyclic and explosive tasks, the system impedance can be lowered, the effects of stiction, friction, backlash and other non-linearities is reduced, which means that cheaper lower quality components can be used, work and power output of a motor can be increased, by choosing an appropriate series elasticity according to a specific task, elastic element allows for shock tolerance, which protects the motor and its gearbox, safety is improved even from the view of the human, since the forces generated during a collision are reduced, force measurement can be approximated by measuring the deflection of elastic element and by knowing its properties, which means that we can exchange the delicate, expensive and chatter-full measurement cells with a compliant elastic element that is robust, cheap and stable, it is possible to implement robust torque/force control, that is based on the feedback loop achieved with the measurement of the deflection of the elastic element. Of course such an actuator also has some flaws. The actuator has unfortunately a more limited bandwidth and the occurrence of oscillations. However, the human movement is still slow enough, so the actuator can follow. The next step are actuators with a variable stiffness, that can be changed on the fly. These variable stiffness actuators introduce some new advantages: changeable natural system dynamics and using this to aid in cyclic tasks, more optimal energy storage, 6

7 in case of walking robots, the forces can be distributed among the legs to achieve an active suspension, that is robust on rough terrain Fixed stiffness and variable stiffness actuators The definition of an ideal stiff actuator or an non-compliant actuator is that such an actuator moves to the desired position and holds it or follows a desired trajectory, regardless of external load. In contrast, a compliant actuator allows a deflection of its equilibrium position, if the external force has changed. In a state of no external load, the actuator holds the equilibrium position. The deflection oscillates around this equilibrium position. Different types of compliant actuators differ in energy needed to change the stiffness, the deflection range, amount of deflection torque and mechanism complexity. The perfect variable stiffness actuator should be small, light and compact with a wide range of deflection, big area and speed of stiffness variation, high torque transmission and the possibility to store energy. In the seminar only a few compliant actuators are mentioned. The reader can find more about other kinds of variable stiffness actuators in [4] in [5]. In their work, the author Ronald Van Ham et al. [4] divide compliant actuators among those with an equilibrium-controlled stiffness, antagonistically-controlled stiffness, structurally-controlled stiffness and mechanically-controlled stiffness. Compliant actuators with equilibrium-controlled stiffness use a spring with a fixed stiffness, that is not changed during usage. This type of actuator is often called a serial-elastic actuator (SEA) [2], determined by its serially connected elastic element, as shown in Fig. 1. This principle uses only one motor. Instead of an electrical motor, a hydraulic systems can also be used [6]. The compliance does not change and is dependent on the integrated elastic element. The elastic element can also be connected in parallel, to achieve a parallel elastic actuator (PEA). Generally, the serial elastic element can reduce the peak power demand of the motor and the parallel elastic element can reduce the torque demand on the torque source [7]. Dynamic adaptability of the actuators is the next step of compliant actuation. These new variable- Figure 1: An example of a serial elastic element actuator [8]. stiffness actuators (VSA) or variable-impedance actuators (VIA) introduce another motor, that enables new options as double actuation units that adapt compliance of the whole actuator system. The first type of actuators with variable compliance are the actuators with antagonistically-controllable stiffness. The inspiration for this type of actuators comes from nature, where a good example is the antagonistic configuration of muscles in the human body. In this type, two identical actuators/motors are connected antagonistically, so that they can work one against the other, or together. Both actuators have a fixed compliance and use an elastic element with a nonlinear relation of deflection and force. With proper control of both actuators, variable stiffness of the system can indirectly be achieved. The 7

8 characteristics of spring or an elastic element must be nonlinear in order to achieve a linear stiffness of the system. An example of such an actuator is depicted in Fig. 2. The problems of this group of actuators are a higher complexity of the system, usage of harder to get nonlinear springs, nonoptimal energy storage and in usage of identical actuators. Both actuator are the same and each one has to Figure 2: Antagonistic compliant actuator and a mechanism to change a linear spring into a nonlinear spring [9]. have enough power to work by itself. Another problem is the stiffness setup, which is not mechanically separated from equilibrium position setup. The position and stiffness variation of the system result from a complex control scheme and is not mechanically separated. In the example of Fig. 2, the nonlinear characteristic of a spring is achieved through a special mechanism in combination with conventional linear tension springs. Actuators with structurally-controlled stiffness are the next group of compliant actuators. The compliance can here be changed directly through the change in structure of the elastic element. In this case the load is not distributed through the whole structure of the elastic element. An example is the Jack Spring actuator, which is shown in Fig. 3. The compliance of the actuator can be changed through a special mechanism that blocks a certain amount of spring length or number of coils. This way the spring characteristic as a whole is changed. The problem here is that a high quality mechanism is hard to achieve. A custom made elastic element is needed, else problems with backlash and other nonlinearities occur. All this can make the mechanism quite complex, which gets only more complex with higher rated loads. Figure 3: The principle of Jack Spring actuator [4]. The next group of actuators with variable stiffness achieves stiffness variance through mechanics. These are the mechanically-controlled stiffness actuators. The principle is similar to the structure-controlled stiffness, the difference is only that here the whole elastic element is always under load. In structure controlled variable stiffness mechanisms, the elastic element can be partially under load. One way to change the stiffness is by changing the pre-load of the elastic element and using a mechanism that loads the spring when the link moves. Good examples are the MACCEPA [10] and VS-Joint [11] mechanisms. 8

9 In the last group of actuators belong other pneumatic systems and all kinds of other hybrid systems, that can not be assigned to previous groups. The pneumatic systems are quite interesting, because the compressibility of air enables design of a compliant actuator. The McKibben pneumatic muscles [12] are a known example. Some of the advantages of such actuators based on pneumatic system are a high power/weight ratio and a convenient elongated shape. One disadvantage is the nonlinear pressure/force behaviour of air, which makes control more challenging. Pneumatic muscles generate force in only one direction, that is why two of them in an antagonistic setting are required. Other flaws are also a slow deflation rate, hysteresis, minimal starting pressure and the need for a pressure tank or a gas compressor. An example of the actuator can be found in Fig. 4. Figure 4: Example of a pneumatic compliant actuator [4] MACCEPA 2.0 Mechanically Adjustable Compliance and Controllable Equilibrium Position Actuator, referred to as MACCEPA, is an actuator with mechanically adjustable compliance, proposed at Vrije university in Brussels [13], [10]. The main advantages of MACCEPA are: - simple design, - usage of linear extension springs, - usage of simple off the shelf components, - mechanically separated stiffness and equilibrium position set-up, - possibility to adapt the stiffness characteristics by changing the shape of the deflection pulley. The variable stiffness with MACCEPA actuators is achieved via a mechanism that provides pretension of the elastic element. Different versions exist. The principle of the first version of MACCEPA actuator [13] is shown in Fig. 5. The actuator has three levers that rotate around the same point. Lever A can be moved with respect to lever B using a position motor. This way the equilibrium position can be set. Equilibrium position is the position of the lever in a state of no external load. Lever B is connected to lever C with a linear extension spring. When moving lever A, the lever B also moves. This way the spring starts to extend and returns the applied force to lever B. After the external force is removed, the lever A starts to oscillate, if the motion is not damped. The variable stiffness is achieved on the lever C, where an actuator with a special mechanism can pretension the spring. This second motor is called the stiffness motor. By changing the pretension, the stiffness of whole system and the oscillation speed can 9

10 b c a Figure 5: The principle of the first version of MACCEPA actuator [13]. be changed. A second version of MACCEPA called MACCEPA 2.0 was already developed. It is shown in Fig. 6. The c b a Figure 6: The principle of MACCEPA 2 actuator [10]. concept is similar to the previous one. Levers A and C are practically the same. Lever B is replaced with a deflection mechanism. The deflection pulley has a special shape, which deflects a specific amount of cable depending on the angle of the pulley, with respect to lever C. By designing this shape accordingly, different characteristic of stiffness can be achieved in combination with a linear tension spring. The stiffness and position are here mechanically separated DLR - VSJ Joint Institute for robotics and mechatronics of the German Space Agency DLR (ger. Deutsche Zentrum für Luft- und Raumfahrt) developed its own collection of compliant actuators. Some examples are the VS-Joint (Variable Stiffness Joint) [11], FSJ-Joint (Floating Spring Joint) [14] and QA-Joint (Quasi- Antagonistic Joint) [15]. Actuators were later integrated into a robotic arm, that has compliant actuation in each joint [16]. In the following we provide details on the VS-Joint mechanism and an improved version called the FSJ- Joint. The principle of these mechanisms is complementary to the MACCEPA mechanism. MACCEPA mechanism uses a linear tension spring. The FSJ-Joint and VS-Joint, however, use a linear pressure spring. The main parts of the mechanism are a cam mechanism, a linear elastic element and a linear guide. VS-Joint is a less complex version of both mechanisms. It is shown in Fig. 7, where the important parts are marked. The mechanism allows for deflection of the lever, which is stopped by a counteractive torque generated by the spring. When the lever deflects out of the equilibrium position, the lower rotation disc is rotated. A special deflection grove is designed in the rotation disc. That grove serves as a cam curve to the cam mechanism. Because of the rotation of cam disk and cam rollers, the translation disk deflects up, which causes pressure on the spring. The pressure spring generates a counteracting force, which stops 10

11 Figure 7: VS-Joint mechanism [11]. the deflection and rotation. If the external torque is removed, the lever starts to oscillate. The VS-Joint uses three cam mechanisms because of symmetry. The mechanism would also operate with only one spring and one cam mechanism. The position motor moves the whole system and designates the equilibrium position. Stiffness motor changes the stiffness of the system by rotating the spindle, which moves the stiffness disk and pretensions the spring. A bigger pretension of the spring results in a lower compliance of the mechanism. This mechanism is more compact than the MACCEPA mechanism. It Figure 8: FSJ-Joint, the improved version of the VS-Joitn mechanism [14]. uses a linear pressure spring. Because of its round shape, the mechanism is convenient for integration in robot joints. A big advantage of this mechanism is again the custom-shaped cam curve. A custom stiffness characteristics can be achieved through the mechanical design of the cam curve. The mechanism uses two motors. The first sets the equilibrium position and is stronger, bigger and heavier. The second motor sets the stiffness and can be weaker, smaller and lighter. The stiffness and position are here also mechanically separated. In Fig. 8 the next version of VS-Joint is shown - the FSJ-Joint. 11

12 3.2 Exoskeletons A wearable robot mechanism or an exoskeleton, sometimes also called an active orthosis, is a special type of robot [17], that works in close interaction with the human user. Its main task is an active assistance in different human movements and human tasks. The exoskeleton either makes the human movement possible, supervises it, or improves it. The robot and the user are connected through a human-robot interface, which is composed of an array of sensors and straps for attaching. The exoskeleton is with the user connected in parallel. The main area of application is in rehabilitation, where they can assist with movements of users with weaker muscles, older people and the disabled. They can assist workers performing different heavy tasks, where the exoskeleton can take some weight off of the user and thus makes the work easier and protect the users from injuries. The exoskeletons can also assist healthy people in everyday lives Exoskeleton design The design of exoskeletons differs from the design of conventional robots. Conventional robots are often as stiff as possible and are consequently heavier. They are often designed and optimised for the task they will be performing. The exoskeletons, however, are foremost designed to fit on the user, to be comfortable and safe. Only after that comes the task, which they will be doing. When designing an exoskeleton, one must pay attention to safety, natural area of movement, ergonomics, motor power, backdrivability, weight limit and moments of inertia. The users safety is always a priority. In this case the safety can be improved with redundant sensors, safety switches, mechanical movement limits and backdrivability. The robot, however, should also be protected. Unexpected shock loads can damage the gearboxes of the motors. The addition of compliant elements can prevent such damage. The robot also has to be comfortable and ergonomic. This means that it has to have a suitable amount of degrees of freedom and suitable movement range so that the robot can successfully assist human movement and not impede on the user or in worst case cause an injury. It is important that the axes coincide with the exoskeleton joints. This way a non intrusive eksoskeleton movement can be achieved. Because of nonlinearity of human body, this proves to be quite a challenge. In Fig. 9(a) we can see the human Lateral Medial Extension Flexion>90 (a) (b) Figure 9: (a) Scapulohumeral Rhythm [18], (b) Knee [19]. arm shoulder, which can be approximated as a 5 DoF joint system [20]. The shoulder joint complex is essential part that enables humans to throw projectiles very fast and relatively accurate [21]. This of 12

13 course makes the whole system more complex and the design of an exoskeleton, that will be able to follow the movement of the shoulder accordingly, very challenging. Another example is the knee, shown in Fig. 9(b), a type of condylar joint, that because of ligaments behaves more like a hinge and is often called a modified hinge joint [19]. A simple 1 DoF rotational joint can not replicate its movement. Because the human and the exoskeleton operate in parallel, the formation of mechanical singularities is more likely. The mechanical singularity occurs when two levers coincide and the mechanism or a robot loses a degree of freedom. The movement is blocked and high forces can be generated in the joints. The exoskeleton should be designed so that the possibility of singularities is minimal and, if possible, that they occur outside of the working range. Most of the time however, we can not completely prevent them. It is also important to limit the mass and moments of inertia, which consequently lower the energy usage and make control easier Exoskeleton control Exoskeleton control is a big challenge and differs from conventional control. The movement of the user must be predicted, else the robot can not assist the user correctly and in time. Exoskeleton control can roughly be divided in three groups. The first exoskeletons are controlled via the push-button principle. This type of exoskeleton operate in open loop. The user controls and operates the exoskeleton using push buttons, located directly on the exoskeleton or somewhere else. The next type is normal control with a conventional feedback loop [22]. The feedback loop uses sensors, that are integrated in the exoskeleton or in the human-robot interface. Most common are the position encoders, accelerometers, force measurement systems, contact sensors, goniometer and inclination meters. These sensors gather information on position and posture of the robot. To achieve a real time response of the robot, a part of the movements of the user must be predicted, based one previous posture, link positions and forces in human-robot interface. The last type of control is based on biological feedback loop. Most commonly the electromyography (EMG) [23], electroencephalography (EEG) [24] and force-sensitive resistance sensors (FSR) are used. EMG is based on the measurement of muscles electric activity using electrodes, attached to the human skin. EEG measures the neural activity of the brain with electrodes attached around the head. FSR sensors measure the force that is generated on a strap around a human limb, when the muscles contract and change its shape. Biologically based feedback loop enables design of control with minimal movement prediction, since the neural activity, the activity of nerves and muscle contraction, occur some time before the limb movement [25]. Other methods exist, that measure biological data, however they are more invasive and therefore demand human-robot interface that is more permanent. Current control system schemes consists of some standard components: a system that reads the outputs of different sensors and converts them to internal states, which is most of the time based on a conventional logical (if-then), fuzzy-logic controller or an artificial neural network [25], a kinematic and dynamic mathematical model, which enables a more predictable control and usage of different compensation methods that ease the movement of the user, like gravity and inertia compensation [26] or posture stability, 13

14 a database of pre-recorded trajectories of repetitive movements [27], a system that interpolates between different movement patterns [27], a finite state machine, which connects the movement patterns on different levels, like walking and standing or leg swing and stepping [27]. Below, some interesting examples of exoskeletons are further explained HARMONY exoskeleton Exoskeleton named HARMONY [18] was build for rehabilitation of upper extremities. It is the result of development at the RENEU labs in the department for mechanical engineering at the Austin Texas university. The novel thing about this exoskeleton is that it can assist both arms simultaneously and a b c Figure 10: (a) Photo of the exoskeleton [18], (b) serial elastic actuator [18], (c) user during therapy [18]. enables a more natural movement of the shoulder complex joint in a relative wide range of movement. It uses a serial elastic element in each joint, which makes force control more robust. Force control is essential for rehabilitation and therapeutic applications. The exoskeleton can adapt to different body ratios. It can generate up to 30 Nm torque in the shoulder. Together it has 14 degrees of freedom. Five in each shoulder, one in each elbow and one in the lower arm. Its speciality is a parallelogram mechanism that more closely approximates the movement of the shoulder complex joint. It is positioned on the back side of exoskeleton. The special arrangement of the three rotation joints in the shoulder maximise the available working area despite the presence of a mechanical singularity. Control of this exoskeleton is based on a torque feedback loop, impedance control and compensation of forces due to the exoskeleton dynamics. The exoskeleton does not compensate for inertia forces, since the link movement during therapy is relatively slow. The effects of inertia can thus be neglected. The generated torque in each joint is composed of compensation torque, connection torque and the task torque. Compensation torque compensates the forces that result from exoskeleton dynamics. The inertia is neglected. The connection torque is calculated using elastic and damper constants based on the reference trajectory. Its purpose is to control the serial elastic actuator. Task torque is the rest of the torque, that is needed to complete the task. The goal of control is to produce force/torque in joints, so that the robot moves the shoulder complex gently to the desired trajectory. Position control is therefore not suitable. Even a small kinematic 14

15 variation in coordinated movement can result in forces, that can cause too much stress on, or even risk injury of the musculoskeletal system of the user ESA exoskeleton Exostation is an ESA sponsored project, where the 7 degree of freedom haptic robot was developed. The portable robotic arm is anthropomorphic and can control another distant robot and simulate the resultant force back to the user. SAM [28](Sensoric Arm Master) has serially connected links. These are isomorphic to the human arm. All 7 degrees of freedom are actuated, which is minimal for successful work of the operator. The exoskeleton can be adapted to human arms of different size and ratios via a sliding mechanism integrated into aluminium links. Special kinematic design maximises the working area of the robot and avoids internal singularities. Current weight of the device is 6.5 kg., mainly because of the actuators. Most of the mass is located in the shoulder complex. Each joint comprises of a brushed DC Figure 11: SAM haptic exoskeleton [28]. motor with a cable transmission and a gearbox. The cable transmission is very common among the haptic robots, because of it high level of backdrivability and low level of friction. The gearbox transmission makes the actuator more compact, but produces somewhat more friction and backlash. The goal is to combine both and achieve large enough power in a compact design with little friction and backlash. In each joint is a position and a torque measurement system in the form of an incremental encoder and an integrated torque meter using strain gages positioned in the cable transmission. The cable drive pulley has a special shape, where the deformation of internal part of the pulley is linear. The deformation is measured using strain gages. With this information external loads can be approximated and predicted. The control architecture is implemented on a real-time controller located on a personal computer and on a set of small controllers distributed among the joints. These are connected among each other with an internet cable and communicate with a frequency of 500 Hz. The motors are controlled using PWM (Pulse-Width Modulation) in a H-bridge architecture. The controller collects data from potentiometer and the strain 15

16 gages. At this stage only impedance control strategy has been implemented, where exoskeleton sends position to a remote robot and receives feedback information on torques, which the exoskeleton then emulates on the user Ekso exoskeleton The following example is a product of the company Ekso Bionics [29], which was formed in year In 2012 they presented exoskeleton Ekso as the first commercially available exoskeleton. Their exoskeleton (a) (b) Figure 12: (a) Exoskeleton during therapy [29], (b) photo of the exoskeleton [29]. returns the ability to stand up and walk to the user, that have either weaker lower extremity muscles or disabilities of the lower body. The exoskeleton weights approximately 23 kg and transfers all its weight to the ground. It has 2 actuated degrees of freedom in each leg. The motors are positioned in the hip and in the knee. Other degrees of freedom are passive. The motors are powered with two lithium batteries, that provide autonomy up to 6 hours. The suit is attached to the user via straps, that can be positioned over clothes. Exoskeleton can be adjusted to a user in under 5 minutes. It costs over $, however, the patients can test and use them in clinics and therapeutic programmes. Walking speeds from 0.45 m/s to 0.9 m/s can be achieved, they are, however, dependent on the user. The exoskeleton uses different sensors, including a gyroscope and sensors to measure trajectory and torque. Based on this information the robot detects the amount of assistance necessary, with a frequency of 500 Hz. The walking can be initiated with the shift of the users center of mass, which activates the sensors and initiates the first step. The overall control is comprised of different strategies. The first is named FirstStep and represents robot operation, where the therapist initiates the steps with push buttons. The user progresses first from sitting to standing and then to walking with crutches. This transition usually happens already on the first session. The second strategy is called the ActiveStep, where the user itself controls the initialization of the steps via buttons on the crutches or on the exoskeleton. 16

17 ProStep is another way of control, where the user initializes the step with the hip movement and leaning forward. The exoskeleton recognizes the pattern and checks that the user is in the correct position and in the right phase of gait. The ProStep Plus allows initialization of stepping, when the user leans the body forward and starts to move his leg. The exoskeleton has another mode for training, where the exoskeleton gives a voice feedback information whether the user is in the correct position to initiate a step. This mode enables the exoskeleton to find the optimal position of the users center of mass to initiate the step. The last mode is called Smart assistance. It is a program that automatically determines the amount of assistance the user needs Passive ankle exoskeleton The next exoskeleton [30] example is the result of the development at the university of Carnegie Mellon in Pennsylvania in collaboration with the department for biomedical engineering at the university of North Carolina State and the university of North Carolina. This exoskeleton is completely passive and can assist the walking of a healthy person. This light weight exoskeleton imitates the working of the Soleus muscle and Achilles tendon in the human calf. The exoskeleton has one degree of freedom in the ankle. It uses a spring that is connected in parallel with the Achilles tendon using a light composite frame. A mechanical clutch is connected serially with the spring and the frame. The clutch closes when the foot makes contact with the floor and opens, to enable free movement, when the leg is in the air. The inspiration for the (a) (b) (c) Figure 13: (a) Exoskeleton scheme, (b) clutch scheme, (c) exoskeleton during walking [30]. clutch came from a study where they observed a clutch-like behaviour of the calf muscles with ultrasound imaging. They saw that the calf muscles behave more like a clutch than a motor. This way the muscle can store energy and enable a higher efficiency of positive mechanical power on the joints during walking. The clutch has no motor, no battery or a controller and weights 0.06 kg. The whole exoskeleton has a mass of 0.4 to 0.5 kg for each leg, based on the size of the user. The process of storage and release of the energy is as follows. The mechanical clutch closes when the heel touches the floor. Next the clutch takes up excess string that connects the clutch and the spring, and then locks up. As the leg moves forward, the spring starts to extend and to store energy. At the heel-off phase the spring returns the stored energy. In the rest of the gait the clutch remains open. The process is shown in Fig. 14. The authors performed a study on the exoskeleton effect on human walking energy expenditure. The subjects were walking o a treadmill with the average gait time around 1.15 s. The usage of a spring reduced 17

18 Figure 14: The working principle of the exoskeleton [30]. the overall metabolic energy. Walking with the exoskeleton without the spring did not significantly increase the metabolic cost. The usage of the exoskeleton reduced the torque needed in the ankle for 14%, lowered the EMG activity of the muscles for 22%, and reduced the overall metabolic cost for 7%. The effect can be compared by decrease of metabolic power, as when a human stops wearing a 4 kg backpack. 18

19 4 Outlook and future work The usage of compliant actuators significantly improves the safety of humans and robots. Compliant elements allow for easier and more robust force measurement and force generation. The robot arm can be deflected in case of unpredicted errors and collisions. The elastic element filters high frequencies, which protect the mechanical parts of the robot. All this makes the tasks in unstructured environment easier. The biggest problem of stiff robots is that a relatively small movement can result in enormous forces. Stiff, heavy and strong robots will generate huge forces during collisions and that is why they can be dangerous to the surrounding environment. They are also dangerous to themselves. Their mechanical elements(motors, gearboxes) can be damaged. Robots with plastic mechanical parts (plastic motors and gearboxes) are especially prone to such damage. These plastic gears can be damaged even when a robot, for example a humanoid toy robot, falls to the ground. The addition of an elastic compliant element could drastically lower the chances of damage to mechanical parts. Although the forces of the collision still remain regardless of the elastic element. The spring extends the duration of the collision and lowers peak forces. Safety is crucial. This is the reason why a conventional industrial robot can not work in an everyday environment. Even if the robot system is robust and dependable, in the end there is still the human factor. Apart from mechanical errors, electrical errors or sensor errors, the origin of errors can also be purely from the user. The user can make a mistake and send the wrong orders. We must also anticipate this. Active control of stiff actuators, for example on the Kuka LWR, is based on the measurement of torques in each joint. Using the torque data and a precise mathematical model, a feedback loop can be implemented. This is a simulated compliance or an active compliance. These LWR robots present the matured technology of human-friendly robots. The danger of collision was decreased with a lighter arm design using composite materials and before mentioned torque sensors in each joint. The robots however still have some flaws of conventional stiff robots in that their safety is dependent on the sensors and electric circuits. Active control is suitable for slow link movement. By higher speeds it becomes unsuitable, because the probability for collision get higher and the collision forces stronger. The most problematic here are the high frequency parts of the impulse forces, which the robot can not compensate. The sampling rate of the sensors and the response time is just not fast enough. The Kuka LWR robots work at a sampling frequency of 3 khz. This are the current limits of stiff actuators. Stiff actuation is not appropriate for the actuation of the exoskeletons, because human often performs fast movement and because his movements are not exactly linear. We are considering an exoskeleton, that is attached to the users arm and assists by different tasks. The exoskeleton should assist by fast and slow movements and by any movement. Because of this, the compliance should change according to the movement speed of the links. This way higher levels of safety can be achieved. In case of exoskeletons, the human-robot interface is in essence stiff, however, by using straps, some minimal movement between the user and the robot is still possible. If we want to achieve assistance that will not impede the user, we would need to predict all movement of the human links. Considering that human movement is not easy to predict, this is a huge challenge. For a very accurate prediction, a complex simulation of a human musculoskeletal system would be needed. By using stiff actuation and a stiff human-robot interface, if the exoskeleton makes an error in its position, naturally the human body will adapt to it. This is course something we do not want. From here follows, that some amount of 19

20 mechanical compliance is needed. A special example of exoskeleton usage could be in sports, where the number of unexpected collision and fast movement is extremely large. Introduction of compliant actuation into the exoskeletons could significantly improve their performances, safety and comfort. With a correctly designed control strategy, the intrinsic compliance between human and the exoskeleton could be achieved. At that stage the user would completely thrust the robot and think of it as an extension of his body. This is of course the end goal of research of exoskeletons that improve the users performance. Compliant actuators are not yet commercially available. We could not find any distributors in our literature overview. However, even if they were available, they would probably not be appropriate for our purpose, since they would likely be expensive and not prone to modifications. These are the main reasons that we will design such an actuator by ourselves. The design of such an actuator will be a bigger challenge, compared to the design of conventional drives. The mechanism will most likely be a combination of MACCEPA and FSJ-Joint, because of convenience, compactness and our capacities. The actuator must also be light and compact. It must have a relatively large range of compliance and be able to change it very fast. It should have integrated sensors to measure the moment and position. The goal is to develop an actuator that will be modular and will enable integration into the exoskeletons joints. Another challenge is the design of the control strategy for the compliant actuator. Which will be the main focus of the research. A special control scheme needs to be developed, that will optimise the working of the exoskeleton for the task the user is doing. Later, a prototype exoskeleton is built that will use the compliant actuator. From that a lot of applications can follow. The exoskeleton could assist by any user tasks. Some applications for the upper extremities can be throwing, as in throwing a ball. Rehabilitation is also always a viable application. A lot of people have problems with upper extremities, here such exoskeletons can ease the rehabilitation process. Lower extremity exoskeletons are perhaps even more useful. Walking with an exoskeleton or perhaps running is still pretty much an open field and offers many research possibilities. A lot of research is currently invested into completely passive devices like [30], [31]. They are completely mechanical, which means that they can be more affordable and available. An interesting solution is to develop an exoskeleton, which has passive and active degrees of freedom. A hot topic is also control of the exoskeletons. The new compliant actuators demand new control techniques. We can expect a lot of results in this area. New wearable technology also demands new types of wearable sensors that could help in prediction of human intention and movement. 20

21 5 References [1] A. Albu-Schaffer, O. Eiberger, M. Grebenstein, S. Haddadin, C. Ott, T. Wimbock, S. Wolf, and G. Hirzinger, Soft robotics, IEEE Robotics Automation Magazine, vol. 15, pp , September [2] J. E. Pratt and B. T. Krupp, Series elastic actuators for legged robots, in Defense and Security, pp , International Society for Optics and Photonics, [3] R. Alexander, R. Ker, M. Bennet, S. Bibby, and R. Kester, The spring in the arch of the human foot, Nature, vol. 325, no. 6100, pp , [4] R. V. Ham, T. G. Sugar, B. Vanderborght, K. W. Hollander, and D. Lefeber, Compliant actuator designs, IEEE Robotics Automation Magazine, vol. 16, pp , September [5] N. L. Tagliamonte, F. Sergi, D. Accoto, G. Carpino, and E. Guglielmelli, Double actuation architectures for rendering variable impedance in compliant robots: A review, Mechatronics, vol. 22, no. 8, pp , [6] A. B. Zoss, H. Kazerooni, and A. Chu, Biomechanical design of the berkeley lower extremity exoskeleton (bleex), IEEE/ASME Transactions on Mechatronics, vol. 11, pp , April [7] M. Grimmer, M. Eslamy, S. Gliech, and A. Seyfarth, A comparison of parallel- and series elastic elements in an actuator for mimicking human ankle joint in walking and running, in IEEE International Conference on Robotics and Automation (ICRA), pp , May [8] N. Paine and L. Sentis, A new prismatic series elastic actuator with compact size and high performance, in IEEE International Conference on Robotics and Biomimetics (ROBIO), pp , Dec [9] S. A. Migliore, E. A. Brown, and S. P. DeWeerth, Biologically inspired joint stiffness control, in IEEE International Conference on Robotics and Automation (ICRA), pp , April [10] B. Vanderborght, N. G. Tsagarakis, C. Semini, R. V. Ham, and D. G. Caldwell, Maccepa 2.0: Adjustable compliant actuator with stiffening characteristic for energy efficient hopping, in IEEE International Conference on Robotics and Automation (ICRA), pp , May [11] S. Wolf and G. Hirzinger, A new variable stiffness design: Matching requirements of the next robot generation, in IEEE International Conference on Robotics and Automation (ICRA), pp , May [12] G. K. Klute, J. M. Czerniecki, and B. Hannaford, Mckibben artificial muscles: pneumatic actuators with biomechanical intelligence, in IEEE/ASME International Conference on Advanced Intelligent Mechatronics, pp , September [13] R. V. Ham, B. Vanderborght, M. V. Damme, B. Verrelst, and D. Lefeber, Maccepa: the mechanically adjustable compliance and controllable equilibrium position actuator for controlled passive walking, in IEEE International Conference on Robotics and Automation (2006), pp , May

22 [14] S. Wolf, O. Eiberger, and G. Hirzinger, The DLR FSJ: Energy based design of a variable stiffness joint, in IEEE International Conference on Robotics and Automation (ICRA), pp , May [15] O. Eiberger, S. Haddadin, M. Weis, A. Albu-Schäffer, and G. Hirzinger, On joint design with intrinsic variable compliance: derivation of the DLR QA-joint, in IEEE International Conference on Robotics and Automation (ICRA), pp , May [16] M. Grebenstein, A. Albu-Schäffer, T. Bahls, M. Chalon, O. Eiberger, W. Friedl, R. Gruber, S. Haddadin, U. Hagn, R. Haslinger, H. Höppner, S. Jörg, M. Nickl, A. Nothhelfer, F. Petit, J. Reill, N. Seitz, T. Wimböck, S. Wolf, T. Wüsthoff, and G. Hirzinger, The DLR hand arm system, in IEEE International Conference on Robotics and Automation (ICRA), pp , May [17] A. M. Dollar and H. Herr, Lower extremity exoskeletons and active orthoses: Challenges and stateof-the-art, IEEE Transactions on Robotics, vol. 24, pp , Feb [18] A. D. Deshpande, Two Exoskeletons for Upper - Body Rehabilitation, in IEEE International Conference on Robotics and Automation (ICRA), May [19] J. Hamill and K. M. Knutzen, Biomechanical basis of human movement. Lippincott Williams & Wilkins, [20] J. Lenarcic, M. M. Stanisic, and V. Parenti-Castelli, Kinematic design of a humanoid robotic shoulder complex, in IEEE International Conference on Robotics and Automation (ICRA), vol. 1, pp , April [21] N. T. Roach, M. Venkadesan, M. J. Rainbow, and D. E. Lieberman, Elastic energy storage in the shoulder and the evolution of high-speed throwing in homo, Nature, vol. 498, pp , June [22] H. Kazerooni, R. Steger, and L. Huang, Hybrid control of the berkeley lower extremity exoskeleton (BLEEX), The International Journal of Robotics Research, vol. 25, no. 5-6, pp , [23] K. Kiguchi and Y. Hayashi, An EMG-based control for an upper-limb power-assist exoskeleton robot, IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, vol. 42, no. 4, pp , [24] T. Noda, N. Sugimoto, J.-i. Furukawa, M. Sato, S.-H. Hyon, and J. Morimoto, Brain-controlled exoskeleton robot for BMI rehabilitation, in IEEE-RAS International Conference on Humanoid Robots (Humanoids), pp , IEEE, December [25] K. Kong and D. Jeon, Design and control of an exoskeleton for the elderly and patients, IEEE/ASME Transactions on Mechatronics, vol. 11, no. 4, pp , [26] A. Gams, T. Petrič, T. Debevec, and J. Babič, Effects of robotic knee exoskeleton on human energy expenditure, IEEE Transactions on Biomedical Engineering, vol. 60, no. 6, pp , [27] S. Wang, L. Wang, C. Meijneke, E. van Asseldonk, T. Hoellinger, G. Cheron, Y. Ivanenko, V. La Scaleia, F. Sylos-Labini, M. Molinari, et al., Design and control of the mindwalker exoskeleton, IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 23, no. 2, pp ,