State of the Art Technologies for Exoskeleton Human Lower Extremity Rehabilitation Robots

Corresponding author: S.K. Hasan Department of Mechanical Engineering, University of Wisconsin Milwaukee, Milwaukee, WI 53211, USA Email: skkhasan@uwm.edu Abstract: The World Health Organization reports that about 1 billion people worldwide have some form of disability. Of these, 110-190 million people have significant difficulties in functioning. Providing rehabilitation services to a large number of disabled people with limited resources is quite challenging. Robot-based neurorehabilitation is an appealing solution for providing rehabilitation services without the downside of provider fatigue. Despite the importance of robot-based rehabilitation, a limited number of resources are readily available related to the technical challenges that need to be addressed in designing human lower extremity rehabilitation robots? This paper tries to fill up the technical gaps by explaining lower extremity anatomy, probable causes of lower extremity disability, rehabilitation protocols and fundamental safety and ergonomic features that need to be considered during the design/selection of an exoskeleton robot. A comprehensive review of the state-of-the-art sensing and actuation technologies involved in human lower extremity exoskeleton robots is also presented to familiarize the reader with current technologies. It is hoped that a comprehensive coverage of all relevant anatomical, anthropometric, actuation and sensing aspects of lower extremity exoskeleton robots will help the designer come up with the best possible solution for a specific application.


Introduction
Physical disability is defined as a physical state that influences a person's mobility, physical capacity, endurance, or dexterity. It is a severe socio-economic problem in the world. Globally, approximately 15% of the total population is considered as physically disabled. Many people experience a temporary or permanent lower or upper limb impairment sometime in their lifetime. Exoskeleton robot-based exercises are a practical approach for neurorehabilitation. Exoskeleton robots offer an array of possibilities in human upper and lower extremity rehabilitation by providing customizable physiotherapy and evaluating performance during all stages of recovery.
In recent years, research in bio-robotics has received a lot of attention because of its effectiveness in the field of upper and lower extremity rehabilitation. Exoskeleton based physical therapy, prosthetics, artificial limbs are major applications of bio-robotics. This paper focuses on Human Lower Extremity (HLE) robotic rehabilitation which has been an area of active research during the last two decades.
A number of review papers have appeared over the last five years that address different aspects of lower extremity assistive devices. Yan et al. (2015) presented a comprehensive review of all HLE assistive devices developed after the 1990s. In their review, all analyzed devices were classified based on four criteria, namely, the robot control system, mechanical structure, validation with users and classification based on mechanical structures and sensory devices.
The review article by (Rupal et al., 2016) describes the construction and technological features of existing lower extremity exoskeleton robots and classifies them based on human power amplification and rehabilitation. The existing rehabilitation devices were further classified as commercial or prototype versions. Meng et al. (2015) reviewed recent developments in mechanisms and control strategies for robot-assisted lower limb rehabilitation. This work addressed comprehensively various control strategies for robot-assisted human lower limb rehabilitation. Chen et al. (2016) described developments in lower extremity exoskeletons, including their structure, control algorithms used, sensing technology to determine wearer's motion intention, as well as remaining challenges in this field.
The review article by (Chen et al., 2013) focuses on lower extremity assistive robotic exoskeletons used in rehabilitation therapy. The authors classified gait rehabilitation exoskeletons into two groups: Treadmillbased gait rehabilitation exoskeletons and exoskeletons for over ground gait training/assistance. Another comprehensive review by (Shi et al., 2019) summarizes the current state-of-the-art of HLE exoskeleton robots by focusing on human gait analysis, as well as actuation system design and control of the lower extremity exoskeleton robot.
Even though the review articles mentioned above summarize the state-of-the-art for HLE robots, it is seen that a majority of these articles have focused on an overall description of the robot, mechanism and control strategies, application (power assist or rehabilitation) and commercial aspects. Design considerations for these robots which are based on HLE anatomy, training methods and the state-ofthe-art in sensing and actuation technologies have received limited attention in these works. To overcome these gaps in the literature, this review article addresses probable causes of HLE disability, rehabilitation protocol, lower extremity anatomy and basic safety and ergonomic design requirements that need to be considered during design/selection of an exoskeleton robot. A comprehensive review of the state-of-the-art sensing and actuation technologies is also presented to familiarize the readers with current technologies. This paper is divided into nine sections. Section two discusses human lower extremity anatomy. An understanding of the lower extremity anatomy is important in understanding the causes of lower extremity disabilities and to design a suitable lower extremity assistive device. Section three describes the probable causes of human lower extremity disabilities. Fundamental safety and ergonomic features requirements for rehabilitation robots are discussed in section four. Section five describes the robot-assisted training methods, including active, passive, active assist and active resist forms of physiotherapy. Section six presents a review of existing exoskeleton robots. It includes a classification of existing 25 exoskeleton robots based on robot's Degrees of Freedom (DoF), actuator type, mechanical power transmission, feedback sensing method and control algorithm used. Sections seven and eight deal with actuators, drive systems and mechanical power transmission systems respectively. Lastly, section nine describes the intention sensing system that helps the user interact with the robot.

Human Lower Extremity Anatomy
Knowledge of HLE anatomy is essential to understanding the causes of lower extremity disability, developing a suitable rehabilitation protocol and designing an exoskeleton robot for HLE rehabilitation. Extracting features from the HLE anatomy serves as a starting point for an ergonomic exoskeleton rehabilitation robot. For designing, developing and performance evaluation of a HLE exoskeleton robot, the following information is needed: The structure of bones and joints, nervous system, muscle map for HLE, degrees of freedom, ranges of motion and anthropometric parameters. These aspects are discussed briefly in the next couple of subsections.

Bones and Joints
The HLE consists of three major segments: Thigh, shank and foot. Figure 1 shows the major bones and joints present in the HLE. The thigh connects with the vertebra (pelvis bone) through the hip joint, the shank connects with the thigh at the knee joint and the foot is attached to the shank at the ankle joint. The femur, the longest and heaviest bone in the human body, is located in the thigh segment. The shank consists of two parallel bones, the tibia and the fibula. Tibia is located on the medial side and fibula is located on the lateral side of the body. The foot consists of a total of 26 bones. The two significant bones in the foot are the talus and calcaneus. Tibia and fibula connect with the talus. Calcaneus is the largest bone in the foot joint and produces cushioned action during human movement. Figure 2 shows the ankle joint and bones in the foot.

Nervous System
The human body's nervous system is similar to the electrical wiring in a robot. It is a collection of different types of nerves that are made of neurons. There are two types of neurons: Sensory neurons and motor neurons. The role of the nervous system is to communicate between the Central Nervous System (CNS) and different parts of the body. Typically, it takes a fraction of a millisecond to transfer signals between two points. The complete nervous system divided into two parts, the CNS and the Peripheral Nervous System (PNS). Figure 3 shows the central and peripheral nervous systems of the body. The brain, spinal cord and retina of the eye together comprise the central nervous system, which works as the main controller of the body. The PNS communicates between the CNS and different parts of the body. The sensory nerves carry stimuli signals from different receptors of sensory organs to the brain, whereas the motor nerves carry activation signals from the brain to various muscles and glands.  Patients with nerve impairment may suffer from epilepsy, multiple sclerosis, amyotrophic lateral sclerosis, Huntington's disease, Alzheimer's disease, etc. The nervous system may also be affected by vascular disorders. Nervous system disorders may lead to upper or lower extremity disabilities.

Muscles
The muscles act similar to linear actuators and are made of soft tissue. The muscle cells consist of two protein filaments, actin and myosin, which can slide with respect to each other. The sliding of actin and myosin layers produces contraction that changes both the length and shape of the cell. Muscles produce forces and motion and are responsible for posture, locomotion and the movements of the internal organs. There are three types of muscles: Skeletal muscles, smooth muscles and cardiac muscles.
A skeletal muscle, also known as voluntary muscle, is attached to the bone by tendons and works for skeletal movements. Although postural control is often done by unconscious reflex, skeletal muscles are responsible for conscious control. Muscles receive actuation command from the brain via the motor neurons. To diagnose a lower extremity movement disorder, knowledge of muscle location and their contribution to different movements is needed. A list of all skeletal muscles involved in HLE movements is given in Fig. 4 (Florimond, 2008) (Florimond, 2008) A smooth muscle, also known as an involuntary muscle, forms the walls of the organs such as the stomach, intestine, bronchi, uterus, bladder, blood vessels, etc. Smooth muscles are not under conscious control. Finally, the cardiac muscle is an "involuntary muscle," structurally similar to the skeletal muscle and is found only in the heart.

Human Lower Extremity Degrees of Freedom
The HLE has a total of 7 DoF. The femur attaches to the pelvis through a ball socket joint ( Fig. 7) that allows it to rotate along 3 individual axes which allow for:
Hip internal and external rotation At the knee joint, the femur connects with the tibia and fibula via an elliptical-shaped end called femoral condyle. Russell et al. (2018) presented the shape and size of the femoral condyle with its instantaneous center of rotation. As the connection surface has an elliptical shape, it allows for rotational as well as linear displacement at the knee joint. The leg length increases during flexion and decreases during extension (Hasan and Dhingra, 2020). Figure 8 shows all seven DoF of the HLE.
Shank internal and external rotation The tibia and fibula are attached to the talus at their lower ends that provide 2 DOF. The allowed motions are:
Ankle pronation and supination

Range of Motion
Maintaining a proper Range of Motion (RoM) is essential for the exoskeleton robot. Most healthy people have a similar range of motion. In the case of physical therapy, the RoM varies with the patient's health condition. A failure to meet this criterion may result in serious adverse consequences. The reported HLE ranges of motion from various sources are given in Table 2.

Anthropometric Parameters
Anthropometrical properties play an essential role in the development of human lower extremity kinematic and dynamic models. Nikolova and Toshev (2008) reported the anthropometrical properties based on the data collected from 5290 individuals (2435 males and 2855 females).
Some empirical equations are also available in the literature to determine lower extremity anthropometrical parameters based on the weight and height of the subject (Contini, 1972) In Equation (2)-(4), Td is the thigh density, Sd is the shank density and Fd is the foot density.
The whole body volume (Bv) is calculated using the body weight and body density Approximate weights of the thigh, shank and foot were calculated as given in Equation (9)-(11): The length of the thigh (Tl), shank (Sl), foot (Fl) and ankle to lower face of the foot (Ag) were calculated as shown in Equation (12) The locations of the center of the mass from the proximal joint for the thigh (Tcm), shank (Scm), foot (Fcm) are given by Equation (16)

Probable Causes of Physical Disability
The reasons behind physical disability can be classified into two categories: Hereditary/congenital or acquired. Congenital disability occurs before or during birth whereas acquired disability may appear at any moment during the lifetime. Congenital forms of disability are hard to overcome using physical therapy.
There are multiple causes for the acquired type of physical disability. Any kind of injury/damage in the nervous system may lead to acquired physical disability. These injuries may include injuries in the central nervous system (brain, spinal cord) or the peripheral nervous system. The primary causes that damage the central nervous system include accidental brain injuries, strokes, trauma, infections, insufficient oxygen supply to the brain cells, etc.
Stroke is a common cause of neurological injuries which often paralyzes one or more limbs of the body.
Physical therapy is the only treatment for post-stroke recovery (Gibson, 1990). Epilepsy, an abnormal electrical activity of the brain, causes short term or long-term physical disability (Brodie et al., 2012). Cerebral palsy is due to brain cell damage or malformation, which affects a person's movements and leads to physical disability. Cerebral palsy occurs during birth (due to obstructive labor), but the symptoms appear during early childhood (Harstad and Albers-Prock, 2011). Multiple sclerosis is an unpredictable disorder of the central nervous system, which freezes communication between the central nervous system and the peripheral nervous system (Waksman, 2020). Muscular dystrophy weakens the muscles; as a result, physical disability appears (Bushby and Marini-Bettolo, 2020).
Any type of fall that causes injuries to the central nervous system or peripheral nervous system also leads to long or short-term physical disability. Fall also causes fracture of the bones and often requires surgery; after surgery, the patient needs physical therapy to improve mobility disorders. Thus, any type of injury that disrupts communication between the CNS and PNS causes disability.

Requirements for Exoskeleton Rehabilitation Robot
Specific safety standards should be maintained for HLE exoskeleton robots to protect the user from any dangers. By following ergonomic design rules, both the user's comfort and the physiotherapy performance can be improved. The next two subsections briefly explain the desirable safety and ergonomic design features that need to be considered while designing an exoskeleton robot.

Safety
Safety is the most critical requirement for an exoskeleton or wearable robot. Exoskeleton robot interacts very closely with the human body. Any runtime failure in mechanical, electrical and electronics, or control algorithms may cause serious adverse consequences. Rehabilitation exoskeleton robot works parallel with the human joints with a corresponding range of motion for every joint. For healthy subjects, the RoMs have some standard values, whereas the RoM for impaired people are different, depending on the severity of impairment. Every exoskeleton robot should have a mechanism to limit the RoM of the joints. It may be in the form of a mechanical stopper or an electromechanical sensor-based stopping system. Further, every exoskeleton robot should also be equipped with an easily accessible emergency stop button. The function of the emergency stop button would be to stop the movement immediately and hold the robot against gravity.
Exoskeleton robots are designed by considering a maximum payload and exceeding the maximum limit may cause electrical or mechanical failures. A higher safety factor can reduce the chances of both electrical and mechanical failures. Robot payload specifications should be very clearly mentioned in the robot's user manual.
The robot should be operated at a low voltage to minimize the risk of electrocution. The severity of the electrical injuries depends on the amount of current flowing through the human body. The skin resistance varies with sweat gland activities and body temperature. Based on ohm's law, lower skin resistance allows more current flow through the body. While skin resistance is variable, the risk of electrocution can be reduced by avoiding higher operating voltages in a robot.
During the mechanical design of the robot, special attention should be given to avoid sharp edges and corners, which increase the risk of injuries. Dangling wires on moving parts increase the risk of an accident. The designer should allow a suitable passage for electrical wiring.

Ergonomic Mechanical Design
The utilization of ergonomic design rules improves the effectiveness of the rehabilitation exoskeleton robot. Since the exoskeleton robot is attached firmly to the wearer's body, an ergonomic design can improve the user's comfort and allow the user to exercise for a longer duration. The major focus of applying ergonomic design rules in rehabilitation exoskeleton robots are to minimize the macro and micro misalignments. The robot's axes of rotation should be closely aligned with the user's joint axes of rotation. Exoskeleton robots with anthropomorphic structures help to improve joint axes alignment (Cenciarini and Dollar, 2011). Careful consideration of all DoF helps to make the exoskeleton robot ergonomic. For instance, the knee joint is elliptical, which provides both rotational and linear movements to occur simultaneously. None of the existing 25 lower extremity exoskeletons reviewed in section 6 consider the linear motion of the knee; this affects the user's comfort and puts extra pressure on the knee joint during exercise (Hasan and Dhingra, 2020). The robot trajectories should be based on natural human movements. A lot of researchers have developed natural trajectories based on the data collected from healthy subjects (Jezernik et al., 2003;Neuhaus et al., 2011;Kwa et al., 2009;Farris et al., 2013).

Robot-Assisted Training Methods
The effectiveness of the physical therapy depends on delivering the right form of exercise to the patient, at the right stage of recovery. It depends entirely on the competence of the caregiver. Currently, based on the patient's health condition and recovery stage, the therapist selects the proper form of physiotherapy. The primary role of the robot is to follow the therapist's command and perform a task repetitively. It is expected that in the near future with the help of artificial intelligence and natural language processing, the robot will be able to select the appropriate exercise(s) based on the patient's symptoms.
The conventional physical therapy-based recovery process consists of three stages: Primary, intermediate and advanced stage. During the primary stage, the therapist moves the limb along a specific trajectory to improve the joint's range of motion, blood circulation and reduce disuse atrophy. In the robotic rehabilitation system, the robot does the same thing. This mode of therapy is also referred to as passive form of physical therapy. During the intermediate stage of physiotherapy, the subject is also engaged in the exercise. During conventional physical therapy, the patient is asked to move a limb voluntarily. But in the robot-based physical therapy called the active form of physical therapy, the robot moves the limb along a trajectory based on the user's intentions. In the active assist form of physical therapy, the subject is asked to move voluntarily. If the subject fails to complete the task based on specific criteria, the robot will assist in completing the task. The final stage of physical therapy is active resist form of exercise. Some resistive force is applied to the limb and the patient needs to work against the resistive force. The active resist form of exercise helps to increase muscle strength. Most of the time, viscous friction is applied to the joint to create specific resistive forces. During robot assistive physiotherapy, a physiotherapist may be present in person or may use haptic technology. Table 3 summarizes different forms of robot assistive exercises, their expected outcome along with representative research works.

Review of Existing Human Lower Extremity Exoskeleton Robots
During the last two decades, significant advances have been made in establishing robot-assisted physical therapies. The major advantage of robot-based physical therapy is that the robot can efficiently provide passive or active forms of physical therapy without the downside of provider fatigue. The patients can receive therapeutic exercises based on their schedule. The performance evaluation is easier and accurate compared to conventional forms of physical therapy.
Based on our literature search, 25 rehabilitation robots were identified that are used to assist patients with lower extremity disorders. Some robots are specific to gait pattern rehabilitation whereas some are more focused on active or passive forms of robot-assisted physical exercises. In this study, the focus is on the sensing and actuation mechanism used for existing robots. Table 4 enumerates existing rehabilitation robots along with their DoF, actuator type, mechanical power transmission, feedback sensing method and control algorithms.
The existing HLE rehabilitation robots can be classified in many ways based on purpose of the device, actuation system, control scheme used, type of exercise provided by the robot, the sensing method used, etc. Figure 9 illustrates the classification criteria for existing human lower extremity rehabilitation robots.
The next section discusses the actuators and drive system principles along with representative works.

Actuators and Drive System
Actuator selection and placement plays a vital role in the robotic system. Based on the working principle, an actuator can be active or passive. The actuator is selected based on the peak torque and power requirements during dynamic simulation which in turn depends on the input trajectory, the robot's shape, size and payload considered during the simulation. Based on the application requirements, an actuator can be electric, hydraulic, pneumatic, or hybrid type. The robot's mechanical structure also has a strong influence on the actuator's shape and size. The power requirement of the actuator is proportional to the robot's bandwidth (maximum range of operating frequency). Higher bandwidth robots require more powerful actuators. The efficiency and total power consumption of a robot largely depends on the actuator's characteristics. In the case of robots with electric actuators, the power transmission wire gauge depends on the actuator's specifications. For a specific power, the higher the actuator's operating voltage, the lower are the current requirements, with a corresponding influence on the wire gauge. A proper selection of actuators increases the power-weight ratio.    In the next couple of subsections, different types of actuators commonly used for exoskeleton robots are discussed along with their selection criteria and control strategies.

Passive Actuators
Passive actuators don't require any power to operate. Based on the application requirement(s), a passive actuator is used individually or in combination with an active actuator. Passive actuators play a vital role in the design and development of energy-efficient robots. Springs are commonly used as the standard form of passive actuators in robotics applications (McGeer, 1990). Collins et al. (2001) presented applications of passive actuators in bipedal walkin. Some existing robots which have used spring as a passive actuator are discussed in (Wang et al., 2013;Banala et al., 2007;Strickland, 2019).

Active Actuators
A majority of the actuators used in robotics applications are active actuators. Active actuators need a power supply to function. They are comparatively expensive and need a sophisticated control system to operate. The major types of active actuators include the electric actuator, the hydraulic actuator and the pneumatic actuator. Every actuator type has its advantages and disadvantages. A proper selection of actuators plays a vital role in robot design and development. Presented next is a brief description of the working principles of different types of actuators, their applications, along with representative works that employ these actuators.

Electric Actuator
The most common type of actuator used in robotics applications is the electric actuator. High peak torque, the flexibility of motion control, low operating and maintenance costs, reliability, workability in a high bandwidth system, compact shape and size have made electric actuators very popular for robotics applications. There are different types of electric actuators available (Krishnan, 2001). Based on the application requirements, a suitable electric actuator is selected. Presented next is a brief description of different types of electric actuators.

DC Motor
A DC motor is one of the most common types of actuators used in robotics applications. Based on the electrical configuration between the stator and rotor magnetic field coil, the DC motor can be classified as a series-wound, shunt-wound, or compound wound motor. In the case of Permanent Magnet (PM) DC motor, the stator field coil is replaced by a set of permanent magnets. Each type of motor is characterized by its torque-speed relation curve, which expresses the operating characteristics at different loading conditions. Figure 10 shows the characteristics curves of different types of DC motors. In the case of PM DC motor, since the motor speed decreases proportionally with the load, such motors are not suitable for constant speed applications. A shunt-wound DC motor is less sensitive to the load around the rated motor capacity but at loads exceeding two times the rated load, the speed falls very sharply. From the motor characteristic curve, it can be seen that the series-wound DC motor is capable of providing a high starting torque (as high as eight times of the rated torque). The speed of a series-wound DC motor is more sensitive to the load. They are good for the applications that require variable speed and higher starting torques. In compound wound motors, the field coils are connected in both series and parallel combination with the rotor coils. They exhibit characteristics between the shunt and series-wound motors.
The DC motor speed is controlled by varying motor torque as the torque is proportional to the current through the motor windings. Electronic driver circuits control the current to control the torque. A use of DC motors as the major actuator in the robots can be found in (Zeilig et al., 2012;Esquenazi et al., 2012;Kong and Jeon, 2006;Colombo et al., 2000;Bernhardt et al., 2005;Araujo et al., 2015;Gloger et al., 2015;Mori et al., 2006;Marcheschi et al., 2011;Chen et al., 2017;2019).

AC Motor
AC motors are less expensive compared to the DC motor. AC motors can have different torque-speed characteristics curves, which makes them an excellent choice for various applications. If AC power is supplied directly from the AC supply line, then these motors are limited to a few different speeds because the motor speed has a direct relationship with the supply voltage frequency. The running speed can be varied by using an inverter. Another option is to use a mechanical gearbox to change speeds and available torque. AC motors can be subdivided as a synchronous motor and asynchronous motor. Synchronous motor speed has a relation with the AC line frequency ( 120 s f n p  , where ns is the synchronous speed, f is line frequency and p is the number of motor poles), whereas asynchronous motor has a small amount of slip (typically 3 to 10%).

Brush Less DC Motor (BLDC)
BLDC motor is also known as Electronically Commuted motor (EC) or synchronous DC motor. The BLDC motor is popular for robotics applications because of its higher efficiency, low acoustic noise and longer operating life. In a BLDC motor, the rotor consists of a set of permanent magnets and the stator is based on the electric-magnetic field. The stator magnetic field rotates continuously driven by the electronic switching circuit. The rotor always tries to synchronize with the stator's rotating magnetic field and as a result, it rotates. Like the permanent magnet DC motor, no electric medium required for carrying current to the rotor. Many of the existing lower extremity rehabilitation robots use a BLDC motor. A list of representative works include (Neuhaus et al., 2011;Sanz-Merodio et al., 2012;Kim et al., 2013;Tagliamonte et al., 2013;Moltedo et al., 2016).

Stepper Motor
The working principle of the stepper motor is similar to the BLDC motor. Typically, it runs based on open-loop control architecture. As no feedback system exists in an open-loop control system, the rotor position can be out of phase at any instant. To deal with this problem, often, little extra power is added with the calculated power of the motor. The main advantage of a stepper motor is that they can sit energized for an extended period (although the temperature rises as high as 150°F). The number of magnetic strips arranged around the circumference of the rotor determines the resolution of the motor (typically 200 steps/rev). With the help of a micro-stepping stepper motor driver, the resolution can be increased as high as 2000 steps/rev. The main difference between the BLDC motor and stepper motor is that the BLDC motor runs more smoothly than the stepper motor, while the stepper motor is very good at providing holding torque.

Servomotor
Servo motor refers to a complete actuation system that includes the motor (based on the application requirements AC/DC motor can be used), driver, motor position/velocity sensor (encoder/tachometer) that runs based on some close loop control algorithms. AC motor based servo systems are more popular because they can be run at a wide variety of speeds by varying the supply voltage frequency. AC synchronous motor speed becomes fixed at a specific value based on the AC line frequency, which is called the synchronous speed of the motor. A servo motor is capable of providing a programmed function of position, velocity, acceleration and torque. A servo system is popular because of its fast response, high efficiency and high torque to inertia ratio. The following references used servo motors as the major actuator for their robots (Kawamoto and Sankai, 2004;Suzuki et al., 2005;Kawamoto et al., 2009;Jianfeng et al., 2013).

Series Elastic Actuator
The Series Elastic Actuator (SEA) is a relatively new concept introduced by (Williamson and Robinson, 1995). In the conventional robotic system, the actuator is rigidly attached to the load. A force or a torque sensor is placed in between them to measure the flow of force or torque to feedback to the control system. In some cases, the motor current is measured to estimate the motor torque. But in the case of a series elastic actuator, instead of using a force, torque, or current sensor, the spring's displacement is used to measure force/torque passively. The spring allows the load to move freely. Figure 11 shows a common configuration of a series elastic actuator. A pair of springs is attached in antagonist configuration. With the help of two encoders, the change of spring length is measured. The Hooke's law defines the relationship between the force (F) and displacement of the spring (X) as: In the Equation (22), F is the force, K is the spring stiffness (constant for a specific spring) and X is the displacement. In the above equation, the force can be measured with the help of displacement. A similar principle is applicable for torsional spring. Nowadays, SEA is widely used in rehabilitation robotics. The following references use series elastic actuators (Kwa et al., 2009;Veneman et al., 2007a;2007bWang et al., 2013Veneman et al., 2005;Munawar et al., 2015;Tagliamonte et al., 2013;). Krishnan has discussed the modeling, analysis and control techniques of all the above mentioned electric actuators in-depth (Krishnan, 2001).

Hydraulic Actuation System
The hydraulic actuation system is more suitable for high power-weight ratio applications. Both linear and rotary hydraulic actuators are available. The efficiency of the hydraulic system is less than that of an electric system. The hydraulic system does not require reduction gears. It can work for a wide range of bandwidths. The hydraulic system requires a pump, reservoir, motor and hoses. Operating temperature affects the system performance (since fluid viscosity varies with temperature). The following references discuss some robots actuated using hydraulic actuation system (Strickland, 2019;Zoss et al., 2006;Kazerooni et al., 2006;Lee et al., 2020;Scheidl et al., 2018).

Pneumatic Actuation System
The pneumatic system runs using compressed air. Pneumatic systems are comparatively inexpensive and have been used in many exoskeleton robots. Position control applications are challenging in the pneumatic system than other types of actuation mechanisms. The main drawbacks of the pneumatic systems are that they are noisy and very difficult to control. The pneumatic system deforms under constant loading conditions. It provides the lowest power-weight ratio. The McKibben type of artificial muscle is one of the best examples of a pneumatic actuator.

McKibben Artificial Muscle
The McKibben artificial muscle was introduced in the 1950s (Gavrilović and Marić, 1969). It is one of the best examples of a pneumatic actuators for an exoskeleton robot (Kobayashi and Hiramatsu, 2004;Noritsugu et al., 2007). Currently, most of the Pneumatic Artificial Muscles (PAM) is based on McKibben artificial muscles. It consists of an internal bladder plaited by a flexible, but non-extensible mesh. The ends of the bladder are firmly attached to both ends of the mesh. When the inner bladder pressurized, the bladder expands to maximize the volume. Due to the action of non-extensile mesh, the whole system expands radially and shortens axially. McKibben artificial muscle possesses the following mechanical properties: Spring with variable stiffness, nonlinear passive elasticity, flexible and very lightweight (Hannaford and Winters, 1990). Chou and Hannaford (1996) developed a linear model of the McKibben pneumatic artificial muscles. Wickramatunge and Leephakpreeda (2010) studied different properties of pneumatic artificial muscles, including the relation between pressure, force, length, stiffness, elastic force. The relationship between the activation level and length, velocity, tension varies with the physical construction of the artificial muscle and needs to be modeled for the individual prototype. McKibben artificial muscle is used in pairs and antagonistic configuration to increase the joint stiffness. Teng et al. (2016) characterized the activation, displacement and velocity in the case of the opposing pair configuration. Zhang et al. (2008) presented the modeling and control of pneumatic muscle actuator for the elbow exoskeleton. They developed a fuzzy logic controller for controlling a pair of antagonistic pneumatic artificial muscles for the elbow exoskeleton.

McKibben Artificial Muscle Using Shape Memory Polymer
The main drawback of McKibben Artificial Muscle is that it needs continuous control for position tracking, which makes it challenging for real-life applications. Using a shape memory polymer can solve this problem (Sokolowski et al., 1999;Tobushi et al., 2001;Ishizawa et al., 2003;Behl and Lendlein, 2007;Huang et al., 2005;Baer et al., 2007;Lendlein and Langer, 2002;Lan et al., 2009;Xie, 2010). Shape memory polymer consists of two phases, the hard phase and the soft phase. At any temperature lower than the glass transition temperature (Tg) temperature, it has higher stiffness and is considered a hard phase or "glassy" phase. At temperatures above the glass transition temperature, it exhibits a lower stiffness. This is called the soft phase or "rubbery" phase. Figure 12 shows the temperature stiffness relationship of the shape memory polymer. Takashima et al. (2010) developed McKibben artificial muscle using shape-memory polymer and showed its effectiveness in robotic applications.

Mechanical Power Transmission System
A majority of the robots are actuated by electric actuators. Most of the electric motors run at high speed and deliver a small amount of torque. But for the robotics application, low speeds and high torques are required. To fulfill these requirements, mechanical power transmission is used. Mechanical power transmission can be done in several ways such as using gears of different types/sizes, belt drives, chain drives, sun planet gearbox, strain wave gearing, synchronous belt and timing pulley, etc.
Each type of transmission mechanism has its pros and cons. The selection of suitable mechanical power transmission depends on many factors such as the robot end-effector accuracy and precision requirements, overall cost, shape and size factors, mechanical power loss, cost of the actuator, environment, total power consumption, direction of power flow, mechanical noise and vibrations, etc. A brief description of different types of mechanical power transmission options along with their pros and cons are given below.

Gear Drive
The gear drive is the most common type of mechanical power transmission system with applications ranging from small precision settings to large applications. Gear drives can be used for controlling speed, torque and rotational direction. The most common type of gear is the spur gear. The gear teeth are parallel to the shaft on which the gear is mounted. Spur gears transmit power between two parallel axes. The main drawback of a spur gear based power transmission system is that it provides a significant amount of backlash. As robotics applications require high precision, an antibacklash mechanism is widely used to compensate for mechanical backlash.
A helical gear is very similar to spur gear. The main difference between the spur gear and helical gear's teeth is that helical gear teeth are not parallel to the power transmission axes. Helical gears also have reduced backlash compared to spur gears. They make less noise and induce lesser vibrations. The bevel gear is used for power transmission between two nonparallel/intersecting shafts. It affords a compact arrangement. There four types of bevel gears: Straight bevel gear, spiral bevel gear, zero bevel gear and hypoid bevel gears. The worm gear is used for power transmission between two nonparallel nonintersecting shafts. Figure 13 presents different types of gears widely used for mechanical power transmission. An application of bevel gears in the robot power transmission system can be found in (Kong and Jeon, 2006;Mori et al., 2006;Chen et al., 2017;2019).

Belt Drive
The belt drive is used to transmit power between two shafts. There are three types of belts: Flat belt, V-belt and the toothed belt. Flat belt and V-belt are based on friction between the belt and the pulley; as a result, belt slippage can cause problems. The synchronous belt (toothed belt) eliminates the risk of slippage. Such belts are widely used for robotics applications.

Sun Planet Drive
The sun planet gear system, also known as the epicyclic gear train, is a compact way to provide a higher gear ratio (higher torque and lower speed). The epicyclic gear system has three main parts, sun, planet and the outer ring gear. Based on the internal architecture of the epicyclic gear train, it is two degrees of freedom system, i.e., the user can control the output, including torque, speed, the direction of rotation by controlling two input variables. In the case of robotics application, only one degree of freedom is used as the ring gear is kept stationary. Many times, a multistage epicyclic gear train is used to get a more compact gearing system. Some representative works on the use of epicyclic gear train for robotic applications include (Tagliamonte et al., 2013;Chen et al., 2017;2019). Figure 14 and 15 show the internal architecture and a multistage sun planet gear system.

Strain Wave Drive
Strain wave gearing is also known as a harmonic drive. It is widely used for robotics applications and possesses many excellent qualities such as: (i) Almost zero backlash, (ii) compact design, (iii) lightweight, (iv) high gear ratio, (v) good resolution, (vi) excellent repeatability, (vii) high torque capability, (viii) coaxial input-output shaft and (ix) very high gear ratio in a single stage (typically 30:1 to 320:1).

Spur Gear
Worm Gear Bevel Gear Hypoid Gear Screw Gear The strain wave drive is built based on the elastic/flexible properties of the metal. It has three major components: A wave generator, flexible-spline and a circular spline (Fig. 16). The source is connected to the wave generator and the output is connected to the flex spline. A wave generator is an epicyclic disk covered by ball bearings. The outer cage of the ball bearings is made of flexible metal that allows changing of the shape during rotation. The wave generator is inserted into the flexible spline (a flexible shallow metallic cap with a finite number of teeth on the outer side). The flexible spline is made of flexible material and due to the action of the inside epicyclic disk, the flex spline continuously changes its shape. The major axis of the flex spline rotates with the wave generator and the flex spline meshes with the circular spline only at a few teeth. For every rotation of the wave generator, the flex spline advances a very small amount. Some references that have used strain wave generator based gear reduction system include (Kwa et al., 2009;Kawamoto et al., 2009;Kawamoto and Sankai, 2004;Suzuki et al., 2005;Kim et al., 2013;Jianfeng et al., 2013).

Bowden Cable
In the conventional power transmission system, the motor and the gearbox stay close to the joint. The main drawbacks of the conventional power transmission system are: They make the exoskeleton robot heavy and large. The robot consumes extra power for carrying the motor and the gearbox and requires a strengthening of mechanical structures. All these problems can be solved with the help of the Bowden cable-based power transmission system. In the Bowden cable-based power transmission system, the actuators are located at a distant place. Mechanical power is transmitted to the joints via cables. The main drawback of the Bowden Circular spline

Flexible spline
Wave generator tube-based transmission system is that it introduces a significant amount of backlash. If the gearbox is attached to the motor at a distant location, then the system gives more backlash, whereas if the gearbox is located at the joint and the motor is at a distant location, the backlash gets divided by the gear ratio and becomes small. There are many robotic rehabilitation system where the Bowden cable-based power transmission system is used such as (Kong and Jeon, 2006;Veneman et al., 2005;Munawar et al., 2015;Marcheschi et al., 2011;Asbeck et al., 2013;Veneman et al., 2007b;Ekkelenkamp et al., 2005).

Ball Screw
The ball screw is a linear actuator that converts rotatory motion into linear motion. The basic components of a ball screw mechanism are a screw with helical grooves on its body, balls and a nut. The ball stays between the screw and the nut. The balls are routed in a continuous path. The following specifications characterize the ball screw mechanism: Lead accuracy, axial play, preload and life/load relationship. The ball screw is a popular mechanism in robotic applications as it provides high accuracy and efficiency. A use ball screw based power transmission in robots can be found in (Colombo et al., 2000;Bernhardt et al., 2005;Mori et al., 2006;Marcheschi et al., 2011;Moltedo et al., 2016).

Intention Sensing System
The user's intention sensing, measurement and proper interpretation are vital for selecting active, active assist or active resist forms of physical therapy. For the active form of robotic neurorehabilitation, the exoskeleton robot's assistance or degree of involvement depends on the user's intention. There are many ways to sense and interpret user's intentions, such as sensing force or torque, sensing different forms of motion, or by analyzing biological signals from a particular or a group of muscles, including surface electromyography (sEMG), mechanomyography (MMG), pressure pattern (Force myography), etc. In the following subsections, different types of intention sensing methods are discussed that can be used for exoskeleton assisted rehabilitation applications.

Force Sensing
The force/torque sensing method is frequently used for determining the user's motion intention. Often force/torque sensors are attached between the exoskeleton robot's mechanical structures and the lower extremity to measure the force/torque exerted by the user. In many cases, force sensors are also attached at the shoe sole or the bottom of the crutches to determine the state of walking. The torque sensor is placed between the motor and the output links to feedback the output torque to the controller or to measure torque assistance provided by the robot. Depending on the output of the force/torque sensor, the robot controller makes maneuvering decisions. In the case of series elastic actuators, force/torque is measured passively with the help of displacements. The following references used force or torque sensors for interacting with the users (Veneman et al., 2007a;2007b;Kong and Jeon, 2006;Colombo et al., 2000;Bernhardt et al., 2005;Veneman et al., 2005;Banala et al., 2007;Agrawal et al., 2007;Kim et al., 2010;Stegall et al., 2012;Kim et al., 2013;Tagliamonte et al., 2013;Jianfeng et al., 2013;Mori et al., 2006;Marcheschi et al., 2011;Yoshimitsu and Yamamoto, 2004;Asbeck et al., 2013;Zoss et al., 2006;Kazerooni et al., 2006;Chen et al., 2017;2019).

Motion Sensing
Another popular way to sense the user's intention is to measure the motion of a specific point on the limb. The motion can be in the form of linear or rotational displacement, velocity, or acceleration. Rotatory encoders, accelerometer, inertial measurement unit, gyroscope and inclinometer sensor can be used for sensing motion. A rotatory encoder converts angular motion (angular position and velocity) into equivalent analog voltage or digital signals. The modern accelerometers are microelectromechanical devices, tiny in size, widely used for measuring acceleration forces or inclination angles. A gyroscope measures the angular velocity. Inertial Measuring Unit (IMU) is a combination of 3 axis accelerometer, a three-axis gyroscope and a magnetometer and is capable of providing absolute orientation. IMU can be used for measuring the absolute orientation of a moving body in the space. An inclinometer also can be used for measuring the absolute angles. A use of these devices can be found in (Kwa et al., 2009;Zeilig et al., 2012;Esquenazi et al., 2012;Colombo et al., 2000;Bernhardt et al., 2005;Kim et al., 2013;Jianfeng et al., 2013;Mori et al., 2006;Marcheschi et al., 2011;Zoss et al., 2006;Kazerooni et al., 2006;Moltedo et al., 2016;Chen et al., 2017;2019).

Different Forms of the Switch
Mechanical switches can be used as sensors. Often single pole single throw switches are used as limit switches. For the user's intention sensing, different types of switches can be used, including reed switch, pillow switch, mercury switches. Pillow switch is handy for detecting the interaction between the user and the exoskeleton robot as discussed in (Neuhaus et al., 2011;Wang et al., 2013;Strickland, 2019;Sanz-Merodio et al., 2012;Munawar et al., 2015;Araujo et al., 2015;Gloger et al., 2015;Asbeck et al., 2013).

Surface Electromyography
Electromyography (EMG) is an effective way to evaluate and record electrical activity produced by skeletal muscle during movements. Surface ElectroMyoGraphy (SEMG) is a non-invasive form of EMG. Instead of using needle, surface electromyography uses surface electrodes for picking up muscles' electrical activities. Typically, the amplitude of the signal varies from 0 to 10 mV (peak-to-peak). The frequency of the signal varies between 0 to 500 Hz, with the dominant energy being in the 50 to 150 Hz range. SEMG signals can be used for estimating the force/torque developed by an individual or group of muscles. There are different types of SEMG instruments are available in the market. Sometimes they come with additional functionalities such as real-time acceleration measurements. Modern SEMG systems use Bluetooth wireless communications, which provide more accessibility to users. SEMG applications for intention sensing is discussed in (Kawamoto et al., 2009;Kawamoto and Sankai, 2004;Suzuki et al., 2005;Wang et al., 2013;Hyon et al., 2011;Morimoto et al., 2012).

Pressure Pattern (Force Myography)
In force myography, a cuff with arrays of pressure sensors surrounding thigh, shank, or foot is used for determining different muscle activation levels. The user's movement intention can be classified based on the pattern of the force myography as explained in (Yoshimitsu and Yamamoto, 2004;Yamamoto et al., 2004).

Digital Camera and Depth Sensor
By using modern digital image processing and machine vision technologies, the individual joint angles can be measured with great precision. Microsoft Kinect combines both the digital camera and image depth sensors and includes an Application Program Interface (API) to get the readily available human upper and lower extremity joints angle (Lun and Zhao, 2015). Kinect was developed for gaming purposes, but it can be used for rehabilitation purposes. The primary application of Microsoft Kinect in rehabilitation robotics is to determine the user's intentions and to measure joint angles at any instant. Even without using Microsoft Kinect, by combining data from two sensors and by using custom algorithms, the same objective can be accomplished.

Discussion
A comprehensive review of the state-of-the-art for HLE exoskeleton robots is presented that includes detailed discussions on lower extremity anatomy, possible causes of HLE disability, rehabilitation protocols, and basic safety and ergonomic design requirements that need to be considered during design/selection of an exoskeleton robot. Both electrical and mechanical sensing and actuation technologies are discussed in depth to familiarize the reader with current technologies. A lack of commercial names and/or model information of previously used sensors and actuators should not be considered as a limitation of this paper. Instead of focusing on a specific brand, the sensor family's working principles are discussed. Attempting to explain all the previously used sensors and actuators requires a considerable amount of information which presents challenges in keeping this article of reasonable length. A comprehensive coverage of all relevant anatomical, actuation and sensing aspects of lower extremity exoskeleton robots will help a designer come up with the best possible solution for a specific application.

Conclusion
The goal of this paper was to extract and analyze information from existing lower extremity exoskeleton robots, identify the gaps and utilize them for further improvement(s) in the design of exoskeleton rehabilitation robots.
To achieve this goal, 25 existing human lower extremity exoskeleton robots were examined carefully based on published articles and relevant review papers. It was noticed information pertaining to human lower extremity anatomy was missing from exoskeleton robot design and development. The gap has been filled herein by explaining human lower extremity bones, joints, nervous system, muscles, anthropomorphic DoF, RoM and anthropometric parameters. Knowledge of human lower extremity anatomy is vital to understanding the causes of human lower extremity disability. For design, development and performance evaluation of human lower extremity exoskeleton robot, anatomical features play a crucial role.
It was also noticed that in the existing literature, there is insufficient explanation of the causes of human lower extremity disabilities. An explanation of the probable causes of human lower extremity disability is provided herein along with an explanation of how they are related to the nervous system and the muscular system. Safety and ergonomic design are two critical aspects that need to be addressed during the design and development of human lower extremity exoskeleton robots. Safety is the first requirement for any device that deals with humans. As the lower extremity exoskeleton robot firmly attaches to the human lower extremity, the designer needs to consider safety issues seriously. Different ways to improve safety aspects from multiple dimensions were explained. Ergonomic design features help improve the user's comfort thereby helping users to exercise for a prolonged period. Techniques for incorporating ergonomic features in the exoskeleton robot were discussed. Different phases of neurorehabilitation and the associated physical exercises, including passive, active, active assist and active resist forms of physical therapies, were also explained.
After reviewing the existing literature, it was also noticed that no review article has addressed in depth the sensing and actuation system used for human lower extremity exoskeleton robots. The actuators act like muscles of the robot whereas the sensors help to interact with user and control of the robot. A detailed description of different types of sensors and actuators used in existing exoskeleton robots, their working principle, along with representative works, is presented.
It is hoped that a comprehensive treatment of all aspects that need to be considered while designing a lower extremity exoskeleton robot will help designers develop a good understanding of human anatomy which in turn, will facilitate the design of safe and ergonomic robots. Knowledge of disability causes will help in the selection of the best form of exercise for the patient. Comprehensive knowledge about all the sensors and actuators that are used for existing lower extremity exoskeleton robots will help the designer choose the best sensor or actuator for a specific application.