Lower limb rehabilitation robotics: The current understanding and technology

4.1 Therapeutic robots

Emphasis has been placed on the use of robotic devices to achieve consistent and reproducible therapeutic movements. Since neuro-rehabilitation is both time and labour intensive involving a large number of repetitive movement practices by the patient, the application of rehabilitation robot can release the therapist from heavy training tasks. Due to the advantages of repeatability, accuracy and reliability, therapy robots can provide an effective means of improving rehabilitation outcome as well as reducing health care costs [22].

Assessment of the human sensorimotor function and retraining of the human brain are two key capabilities of such devices in improving the patient’s quality of life. The rehabilitation robotics fulfil these functions by applying three types of therapeutic techniques: priming, augmenting and task-specific practices. Priming techniques include therapeutic interventions (e.g. passive movements, motor imagery, etc.) that prepare sensorimotor system for increased plasticity through direct stimulation of the tissues or joints; augmenting techniques (e.g. EMG biofeedback, constraint-induced movement therapy, etc.) that emphasizes on improving the sensorimotor interactions during the therapy; and task-specific practices which include the repetitive exercise/movement to gain the plasticity and strength [23].

At present, much of the therapeutic robots are focused on retraining the movement abilities of people who have suffered a stroke or spinal cord injury. The main reasons for this focus is the relatively large number of patients with these conditions and the high costs of rehabilitation associated with them. Section 5 provides details of various types of lower-limb rehabilitation robots which have been developed in recent years to assist locomotor training to improve the gait function and therefore patient independence.

4.2 Assistive robots

Body movement is one key area where assistive devices have emerged to improve the activities of daily living (ADL) in patients with SCI, neuromuscular disorders and ageing. Most of the researches in assistive robotics are observed towards seamless integration of assistive technology with the user. In recent times, this integration has been improved by developments in the mechanics of assistive devices (both software and hardware), the physical interface between technology and user, and intuitive control strategies for device operation.

Robotic assistive devices are either used as “alternative devices” which are used in case of total incapacity of mobility, or as “augmentative devices” for patients with residual mobility capacity. Wheelchair navigation systems support patients with a total incapacity to mobility with an autonomous or semi-autonomous mode of operation depending on the user ability to manipulate the controls for steering and obstacle avoidance. Wheelesley [24] and NavChair [25] are two mostly explored smart wheelchairs for their possibilities in providing mobility assistance. Smart wheelchairs are augmented with IR sensors, SONAR and machine vision to provide obstacle avoidance and manoeuvring. Apart from joystick-based control, bio-signals such as electromyography (EMG), electroencephalography (EEG) and electrooculography (EOG) has also been used to control the wheelchair movement and steering [26, 27].

However, stagnant posture due to the continuous use of a wheelchair poses a health problem to the user in the form of bone loss, osteoporosis, muscle atrophy, skin sores, etc. For these reasons, it is important to encourage the use of augmentative devices in place of alternative devices, whenever possible, considering the mobility capacity of the user [28].

Augmentative devices for assistive aids include external aids such as smart walkers and wearable robots. Smart walkers have emerged as a robotic alternative to conventional walkers, promoting better assistance in gait and balance disorders. Guido is an example of smart walker developed by Haptica (Dublin, Ireland) that provide navigation and obstacle avoidance assistance to frail, elderly, visually impaired and Parkinson patients [29]. These smart motorised walkers include sensorial setups such as sonar, wheel encoders, laser, IR sensors and computer vision for navigation and obstacle avoidance, and force sensors and haptic feedback for steering. Walkers have great potential for assistance; however, smart walkers show lack of acceptance and adoption by patients. Ergonomics and aesthetics of these walkers need to be explored to improve and achieve total acceptance [30].

Nowadays, the use of wearable robots, in the form of exoskeletons or active orthosis, is rapidly increasing toward lower limb assistive aids, where the robot is designed to promote the functional activities at home, society and workplace. The term ‘exoskeleton’ is generally referred to a device that augments the performance of an able-bodied person, whereas the term ‘orthosis’ is used for a device used to assist a person with pathology or disability [31]. However, there is no clear cut difference between orthosis and exoskeleton, and the two terms for most of the part are used interchangeably. In this article, the two terms are used as they are referred in the associated references.

Exoskeletons are used to actuate specific joints and thus provide an assistive advantage to the user. Typical objectives of wearable robots are to establish stable weight bearing, to control the speed or direction of limb motion, and to reduce the energy required to ambulate. Tibion PK100 bionic leg (Tibion Bionic Technologies, USA) is an example of partial lower limb exoskeleton, which actively supplements the knee movement for ambulatory assistance of patients affected by neurological conditions. The device includes a shoe with built-in force sensors to determine the user-required assist torque at the knee. The device can be configured to operate in assist mode (continuous assistance during the task), therapeutic mode (providing assistance as needed by the patient) or passive mode (for slow repetitive motion of the knee) [32].

In many cases, it is difficult to have a clear cut distinction between the therapeutic and assistive robots, as many devices offer the promise of combining both the assistive and therapeutic technology together. For example, ReWalk (ARGO Medical Technologies, Germany), is not only an overground gait trainer for patients with paraplegia but also facilitate them with gait assistance to ambulate [33]. In the future, the difference between assistive and therapeutic aids will mitigate. With the advancement in robotic exoskeletons, patients will be able to wear them in the home and community, receiving both therapeutic interventions and supportive assistance when needed. With this viewpoint, it is essential to stress the research and development of exoskeletons for the assistive and therapeutic needs of the patient whenever possible.

In a treadmill-based exoskeleton (TBE), the patient training is accompanied by the exoskeleton while walking on the treadmill. They generally consist of body weight support (BWS) system for ensuring safety, reducing the joint loading due to body weight and maintaining balance [34]. Lokomat, LokoHelp, AutoAmbulator and LOPES are clinically explored treadmill-based rehabilitation robots.

Lokomat is an early TBE developed by Hocoma (Zurich, Switzerland) consisting of cable BWS system for gait rehabilitation. It consists of 2 degrees of freedom (DOF) at each leg, assisting the hip and knee movement in the sagittal plane. The joints are driven by DC motors using ball screw transmission [21]. Woodway and LokoHelp group, Germany, developed LokoHelp robot which was similar in construction to Lokomat. The LokoHelp device is placed in the middle of the treadmill, and the foot is tied to the device to perform the simulated gait action [35]. AutoAmbulator developed by Healthsouth, US, is another cable BWS system for gait training. It uses robotic arms strapped to the ankle and thigh of the patient to simulate the movement at the knee and thigh. The use of robotic arm reduces the overall weight of the device and ease donning and doffing as compared to Lokomat [36]. LOPES (University of Twente, Netherlands) is another TBE consisting of 8 DOF (two for pelvis translation and three revolute joints at each leg) actuated through SEA [37].

5.2 Orthosis-based exoskeletons

Leg orthoses are actuated wearable lower limb exoskeletons that can provide power assistance during walking as well as act as overground gait trainers. These are full and partial lower limb devices based on the number of human joints assisted by the device. Blaya and Herr [38] developed an Active Ankle–Foot Orthosis (AAFO, Massachusetts Institute of Technology) for gait training of drop foot patients. In AAFO, the impedance is modulated throughout the gait cycle by biomimetic torsional spring control using SEA. The variable impedance control at the ankle joint was shown to reduce slap foot. In another approach, pneumatic muscles were used for actuation of knee and ankle joint in Knee-Ankle–Foot-Orthosis (KAFO, University of Michigan) developed by Sawicki and Ferris [39]. The KAFO used proportional myoelectric control with flexor inhibition algorithm to reduce pneumatic muscles co-activation during the simulated gait cycle.

The use of myoelectric signals is, however, difficult in case of neurologic patients due to weak EMG signals. To cater this problem, Suzuki et al. [40] used ground reaction force as another measure in Hybrid Assistive Limb (HAL-5, Tsukuba University) to estimate the user intent in full body exoskeleton for assisting gait movement.

While the devices discussed above are used to assist the user’s muscle, Berkley Lower Extremity Exoskeleton (BLEEX) is a lower body exoskeleton developed by Zoss et al. [41] to support external payload mainly for military application. BLEEX consists of 7 DOF consisting of hip, knee and ankle actuated by hydraulics.

5.3 Foot plate-based end-effectors

In foot plate-based devices, patient’s feet are positioned on the preprogrammed foot plates to stimulate various walking phases. Unlike the exoskeleton where the patient is fixed to the robot kinematics, providing no room for the therapist to physically access the patient’s limb, the foot plate-based devices only support the patient feet while the therapist physically accesses the patient’s limb during training. Gait Trainer, GT-I (Reha-Stim, Germany), uses a planetary gear system to drive the independent foot plates to simulate the foot motion while the patient is supported by a cable BWS system. The device does not constrain the patient’s knee, thereby allowing the therapist to perform correction on the patient’s movement [42].

A major redesign to GI-I is HapticWalker. Developed by Schmidt et al. [43], HapticWalker is a reprogrammable foot plate-based system with the capability to simulate different terrain for comprehensive training of day to day activities such as walking on a rough surface, stair climbing, etc. The system also has a virtual reality (VR) mode, where the user wears the VR display helmet and interact with the virtual scene to augment the training [44]. To compensate for the larger size and high voltage requirement of HapticWalker, Hesse et al. [45] developed G-EO-System. The system was devised for maximum step length of 0.55 m and gait velocity of 0.6 m/s.

5.4 Platform-based end-effectors

Platform-based devices allow the patient to be stationary while the affected limb is attached to the platform, which in turn is controlled to execute the training program. These include stationary gait trainers as in case of MotionMaker [46] and Lambda [47], as well as stationary systems for ankle and knee rehabilitation such as Rutgers ankle [48] and High Performance Ankle Rehabilitation Robot [49]. Parallel robots have become quite common in platform-based rehabilitation devices. Compared to exoskeleton-based devices, end-effector-based robots can easily adapt to different patients. End-effector robots typically make contact with the patient’s body at specific points, making it easier to design and control these types of robots [50].

The key distinctive aspect of wearable robots is their intrinsic dual cognitive and physical interaction with humans. In wearable robotics, a cognitive human-robot interface (CHRI) is explicitly developed to support the cognitive interaction (possibly two-way) between the robot and the human. Information is the result of processing, manipulating and organizing of data, and so the CHRI in the human-robot direction is based on data acquired by a set of sensors to measure bioelectrical and biomechanical variables. Similarly, a physical human-robot interface (PHRI) is explicitly developed to support the flow of power between human and robot. The PHRI is based on a set of actuators and a rigid structure that is used to transmit forces to the human musculoskeletal system [54]. The close physical interaction through this interface imposes strict requirements on wearable robots towards ergonomics, safety and dependability.

6.2 Neuromotor control

There are three levels of cognitive interactions: one related to reasoning and planning, one related to muscle activity and one related to the wearer’s motion. The planning level interaction can be accomplished by monitoring brain activity using different techniques, e.g. EEG or brain-implanted electrodes [55]. The muscle activity level uses muscle electrical activity, i.e. EMG, to command the devices [56]. The movement-related level of interaction uses kinematic and kinetic information from the subject as control inputs [57]. Multimodal approaches propose diversified use of these channels to gather more realistic and robust information and to gain a better understanding of the phenomena through data fusion techniques [58].

7.1 Degree of freedom

The hip is considered to be ball and socket joint with three degrees of rotations. The major design challenge is posed in terms of abduction/adduction and internal/external rotation at the hip joint. MIT exoskeleton used cam mechanism to adapt for the difference between biological and exoskeleton length [59], while in BLEEX the researchers opted to position the center of rotation for abduction/adduction in the rear part of hip joint mechanism [41]. LOPES also opted for a solution similar to BLEEX for providing abduction/adduction DOF [37]. Internal/external rotation is provided in BLEEX by a single axis of rotation in the middle of the hip joint at the rear attachment [41].

The knee is condyloid joint with two degrees of freedom (flexion/extension and internal/external rotation) [60]. Despite this, for simplicity, the knee joint is often modelled for flexion/extension due to very limited longitudinal rotation. Most of the knee exoskeleton considers single DOF at the knee [61–64]. However, due to polycentric motion of the knee in the sagittal plane [39], one purely rotatory DOF is argued to cause misalignment thus over constraining the design and causing undesirable forces at the point of attachment and knee [65]. Several designs have been sought to achieve knee alignment. Kim et al. [66] implemented a four bar linkage mechanism to follow the polycentric motion of the knee. Liao et al. [67] used a five bar linkage mechanism at the knee to drive the shank part using a rack. Wang et al. [65] used the rolling knee joint with a double hinge joint to reduce the knee misalignment in the frontal and sagittal plane. Celebi et al. [68] used Schmidt coupling, which self-align with the translation of the instantaneous centre of rotation of the knee. In another design, soft inflatable exosuit was designed by Sridar et al. [69] to eliminate the misalignment and improve the compliance between the device and knee movement.

The ankle is a complex joint having 3 rotational DOF (internal/external rotation, plantar/dorsiflexion and inversion/eversion). The use of all the 3 DOF is advocated in design because the foot makes contact with the ground [41]. However, to simplify the mechanical design, the ankle is assumed to be hinge joint having 1 DOF (plantar/dorsiflexion) as in the case of ALEX (active leg exoskeleton) [70]. Multi DOF at ankle joint has been realized for various mechanical designs as in the case of ankle rehabilitation orthosis developed by Agrawal et al. [71] and Zhang et al. [72].

7.2 Range of motion

Range of motion (ROM) requirement is dependent on the application of the device and the driving mode. Even though the exoskeleton bears anthropomorphic design in most cases, the ROM didn’t need to be same as the ROM of the biological limb. Zhang et al. [73] reduced the range of motion at the hip, knee and ankle to reduce the extra loading at the passive hinges, reducing vibration and improving the safety of the wearer. Singh et al. [74] replaced the hinge joint at the knee with four bar linkage joint to obtain a desired range of motion at the knee. Kubota and Hasegawa [75] studied the physical feature of the exoskeleton, which may interfere with the wearer’s motion. ROM reportedly tends to reduce with the thickness of exoskeleton if the medial part of the thighs increases beyond 20 mm. ROM for exoskeletons designed for overground walking is generally greater than the actual ROM of the limb to ensure wearers safety and avoiding any restriction to the movement [41].

7.3 Joint torque

To assist the wearer, a robotic device must provide adequate movement torque either passively [37] or actively [76]. Three methods are mostly used in determining the joint torque requirements- human body modelling [77], clinical gait analysis [78, 79] and experimental test using torque sensors [80]. Mostly, the most significant joint torque is observed in the sagittal plane, primarily in flexion/extension DOF except for the hip joint where the greatest torque is obtained in adduction/abduction. For this reason, BLEEX, LOPES and MIT exoskeletons have provided assistance at the hip to augment hip adduction/abduction [41, 82]. The joint torque data is essential in determining the actuation system required at the joint of interest. The system must be able to deliver the necessary motion velocity and adequate frequency response for the dynamic task augmented by the device [81]. The joint torque data is also used to evaluate the muscular effort. Hwang and Jeon [80] computed the active muscular effort from the measured torque at the knees.

8.1 Sensing

The sensors allow feedback from the device and regulate the force, torque and position of the joint necessary to perform the desired movement. Robotic rehabilitation devices developed in the past have used multiple sensors and their combinations such as encoders, inertial measurement units (IMU), potentiometers, sensors to measure bio-signals (EMG and EEG), force and torque sensors, etc.

Ground contact force (GCF) is importantly measured in assistive devices to classify movement [83], diagnosis of abnormal movement [84], model limb [85] and for fall detection [86]. Force sensitive resistors (FSR) and load cells are commonly used for the measurement of ground force. Sahin et al. [87] developed hydraulically actuated exoskeleton which uses FSR for measurement of ground reaction force while load cells were used for the measurement of piston force. The force measurements were used to drive the exoskeleton using a proportional and integral (PI) controller. In commercially developed exoskeleton HAL, a semiconductor type floor reaction sensor was used for GCF measurement for the movement control [40]. Fiber-based force sensors have also been used to improve comfort while walking [88].

Both active and passive range of motion (ROM) measurements are employed in rehabilitation robots. These measurements largely employ potentiometers and encoders at the joints to measure joint angles [89, 90]. However, these require to have their axis aligned with the rotational axis of the joint. To overcome the issue of misalignment in case of joints having multi-axis rotation, inertial measurement units (IMU) are used [91]. Strain gauge-based flexible goniometers [92] and optical-based goniometers [93] are also tested for their usability in providing multi-axial measurement and lightweight application, however, they are fragile and not usable in daily usage. A newer area is also being researched on the use of teleceptive sensing (using stereo RGB cameras, laser, radar, sonar, ultrasonic sensors) for wearable assistive robotic devices [94].

Bio-signals have been incorporated in rehabilitation robotics with their potential to estimate the human intent. Surface EMG electrodes have been used for the measurement of myoelectric signals from the superficial muscles for classification of movement [95, 96] and to estimate joint torque [97, 98]. Fusion of EMG with mechanical sensors such as IMU is found useful in improving the classification accuracies [58]. Examples following the use of EEG signals in exoskeletons are relatively few compared to EMG due to the complicacies in EEG signal processing and data mining. EEG-RoGO (Robotic Gait Orthosis) [99] and BCI-MAFO (motorized ankle foot orthosis) [100] are few EEG driven lower limb exoskeletons developed for paraplegia patients. Mechanomyographic (MMG) signals, picked up from dermally fixed accelerometer, were also used for the control [101].

8.2 Actuation

Various actuation systems are being employed in rehabilitation robotics in the past. These are classified as electric motor, pneumatic and hydraulic actuators. Advantages and disadvantages of these actuation drive systems have been documented by Zhang et al. [44]. The actuation system must suffice in both performance and physical aspect to comprehend the workability of the exoskeleton. The performance requirement includes high specific power, back-drivability, ease of control and efficiency, while the key physical requirements are low mass, low cost, modularity, and noise [102]. A comprehensive review of actuators of various lower limb robotic rehabilitation devices have been documented by Huo et al. [18].

Research has been shifted towards the use of compliant actuators to improve the safety, efficiency and comfort of the exoskeletons. Three types of compliant actuation systems are commonly used –series elastic actuators (SEA) [103, 104], variable stiffness actuators (VSA) [105, 106] and pneumatic artificial muscles (PAM) [39, 107]. The use of SEA in exoskeletons has demonstrated improved human-robot interaction, and also aids in the measurement of the joint torque based on the deformation of the elastic element. Linear springs, torsional springs, spiral springs, Bowden and steel cables have been used as elastic element in SEA and VSA [108]. In VSA, the stiffness of the elastic element can be modulated to change the actuator characteristics thus adapting to the environment and task [109]. The PAM has the advantage of low weight, back-drivability and high specific force [110]. Antagonist configurations of PMA is also been used to obtain bidirectional actuation [90]. However, these are particularly inefficient in producing a large range of motion and high torque due to non-linear force contraction behavior [111].

Both soft and rigid smart materials have found a promising utility as actuators in robotic rehabilitation devices. Shape memory alloy (SMA)-based artificial muscles have been tested in ankle foot orthosis [112]. The use of SMA improves the tracking frequency of PMA by 12 times and reduces the tracking error by 82%. Hyeon et al. [113] demonstrated the feasibility of graphene/carbon nanotubes (CNT) yarn PMA in robotic devices for actuation. The tensile actuation of these electrochemical muscles was found to be twice than coiled CNT muscle. Dielectric elastomer and polymeric molecular actuators are also promising for actuation of robotic rehabilitation devices because of their stress, strain and speed similar to that of a human muscle [102].

8.3 Control strategies

With the complex interaction between human and robot, both cognitive and physical aspect of human-robot interaction (CHRI and PHRI) have been taken into consideration for controlling the device [114]. Different control strategies are required to suffice the various operating mode of a rehabilitation device (active mode, passive mode, active assist mode and active resistive mode) [115]. During early stages of rehabilitation, when patient’s voluntary movements are absent, a control strategy must provide passive training, which must be changed to active mode after the patient would have regained the lost limb movement. These two modes have also been reviewed in literature as trajectory control or position control and assist as needed (AAN) control [116]. Assist as needed control have been obtained with the aids of various sensors and their fusions, categorizing it into force control, impedance control, bio-signal-based control and adaptive control [117]. Table 1 gives an overview of control strategies for lower limb exoskeletons developed in recent years.

Hussain et al. [118] devised an orthosis for gait rehabilitation based on trajectory control. Boundary layer augmented sliding control was implemented to guide the patient leg. Mathematical models and predefined gait trajectories have been used for devising training [119, 120]. Emken et al. [121] demonstrated the feasibility of teach-and-replay method in which the therapist first teach the system by assisting the patient and then the device replays the movement to provide repeated therapy to the patient. In another approach relating to the rehabilitation of hemiparetic patients, Vallery et al. [122] used the Complementary Limb Motion Estimation (CLME) approach to map the movement of unimpaired leg on to the training trajectories of impaired leg.

To augment patient motivated training (active movement training), thereby improving the therapy outcome, real-time assessment of performance is required [123]. Hybrid position and force control strategy access the position as well as the force between the device and user to control the assistance during training. Simon et al. [124] used a novel force control scheme to force leg symmetry during extension exercise. The device responds by increasing the load beyond target resistance whenever the leg asymmetry is detected by the force plate. In gait training exoskeleton, ALEX, developed by Banal et al. [70], a force-field controller was used to provide assistive torque at the hip and knee. This method was also called virtual tunnel approach, as the patients were guided to be in the tolerance limit of gait trajectory. A similar approach was adopted in developing patient-cooperative control for Lokomat [21]. Impedance control was employed to regulate the dynamic relation between the device and the user. Many devices such as LOPES [37] and Lokomat [125] have used impedance control for regulating the patient gait and promoting patient participation. A major drawback of impedance controller is that the impedance parameters must need to be changed with the rehabilitation progress of the patient [117]. To address this problem, an adaptive impedance control was proposed by Hussain et al. [126], which adapts itself according to the disability level of the patient.

Bio-signals have also been used in developing user intended control paradigm. Surface EMG and non-invasive EEG have been used for the control. Do et al. [99] developed a BCI interface orthotic device, RoGO. Alternating epochs of idling and walking motor imaginary were analysed to generate the model for predicting the walking assistance. Feasibility of EEG-based control is, however, limited to EEG sensitivity and classification accuracies [18]. EMG control strategies are mainly classified into two types: onset detection-based control and pattern recognition-based control. For HAL systems, EMG was used to measure the joint torque assistance [127, 128]. Hassani et al. [129] used EMG with knee musculoskeletal model to control the exoskeleton.

9.1 A promising approach

Apart from the other rehabilitation robotic architectures, the exoskeletons, whether tethered or untethered, have better embodiment with the user. This not only has the advantage of effectively integrating the cognitive ability of human being [18] but also promotes greater self-esteem compared to other aids [136], thus improving the overall quality of life of the disabled and elderlies [137, 138]. Based on the joint actuation, various exoskeletons have been researched for their efficacy towards assistance and rehabilitation in different lower limb motor impairments.

The primary thrust of exoskeleton research has focused on medical applications to support and rehabilitate neurological impairments. The notion of using rehabilitation robotics in neurological problems is supported by the high prevalence rate of mobility impairments caused by these disorders. WHO has estimated neurological disorders and in particular cerebrovascular disease, as the cause of maximum disability and healthy life year lost in the world [139].

Exoskeletons in the form of with and without body weight support (BWS) system have shown the efficacy in gait training and assisting mobility of SCI, traumatic brain injury (TBI) patients and rehabilitation of major trauma patients such as stroke, cerebral palsy (CP), Parkinson disease, etc. [140–142]. Apart from walking ability, exoskeletons have also shown their efficacy in improving secondary gait outcomes viz. balance, spasticity and pain [143].

9.2 Still a long way to go

Robotic rehabilitation devices, particularly exoskeletons, are still lying on the curve of inflated expectations in Gartner Hype cycle with at least ten years until reaching a steady level of productive use in the real world [144, 145]. Primarily, most of the lower limb exoskeletons are tethered and used in controlled environments such as hospitals and rehabilitation centres under the supervision of medical professionals. In the direction to restore the autonomy during training as well as gait assistance, only a few overground exoskeletons (untethered) have been researched such as ReWalk [33], Indigo [146] and Ekxo [147]. However, they are still prone to the user subjective criticism such as donning/doffing, movement speeds [143], aesthetics [148, 149], prevention of venous-lymphatic stasis and skin protection at the interface [136] as well as engineering aspects relating to sensors, actuators, control strategies, energy supplies, and materials [144, 150].

Although the use of robotics has shown a promising approach to rehabilitation, two major challenges are hampering the progress. First is the incompetence of the current system to achieve a compliant modulation of the neuromuscular activity while promoting a voluntary robotic control, and second is the limited understanding of the disability induced musculoskeletal changes that impede the understanding of how the patient’s motor intentions can be best formulated in a control strategy for robotic device [151].

9.3 Newer dimensions to explore

Advances in the sensors, actuators, information technology and control techniques are essential to accelerate the success of robotic exoskeletons. Wearable sensors will provide a new dimension to the human intent and motor functionality. Big data algorithms will need to be incorporated to mine the vast amount of physiological signals that will be available in the near future. Machine learning techniques may bring new control approaches to robotic exoskeletons. Additive manufacturing will allow customization and cost-effective approaches in manufacturing exoskeleton frames and components. The use of compliant actuators will add more safety to the operation of exoskeletons [144, 152].

Although, the use of exoskeletons for rehabilitation and assistance of neurological patients is clinically and economically justified [153], a wider approach needs to be explored for less researched non-medical use of exoskeletons as in case of occupational workers, healthy elderlies, sports, exercises and even for the astronauts for extra-terrestrial surfaces [154]. The use of exoskeletons for industrial usage will have direct consequence of reduction in the incidences of WMSDs among occupational workers by defining reachable spaces for the tasks and supporting joint movement.

Recently, due to the rapidly growing elderly population, attention has been started for providing assistance to elderly people for carrying ADLs. There emerged a growing urgency for assistive technologies to help elderly people remain independent. Healthy elderly people do not, as such, have serious problems which demand the intervention of medical personal except the reduced physical abilities. However, if left unattended, the frailty leads to disabilities over time and may even aggravate the progression of other musculoskeletal, cardiovascular and respiratory pathologies [155]. Exoskeletons will be useful in addressing the age related inabilities to promote active and assisted living (AAL). Not only can exoskeletons help the frail elderly, but they can also delay the onset of frailty by providing the required mobility, strength and endurance training [10].

9.4 Challenges

With the advent of wearable sensors and advancement in actuation technology, research in robotic rehabilitation and in particular lower limb exoskeletons have experienced a sustained pace over the past decade. Most of the rehabilitation exoskeletons (assistive/therapeutic) developed in the past were, however, aimed towards patients with neurological disorders (SCI and stroke), with very less emphasis on musculoskeletal disorders. Although wearable robotics and exoskeletons are considered as a key enabling technology for personalized ambulatory and assistive solution, the technology is still in the development phase and many factors still need to be addressed in this domain.

One broad issue in the domain of exoskeletons in the detection of human intention beforehand. Control strategies promoting high and low level human-robot interaction is very important in the drivability of device and further to encourage patient’s recovery. Efforts have been made in enhancing the human-robot interaction using EMG signals. However, the use of EMG has several limitations that are required to be addressed in novel ways. Firstly, The EMG signals are prone to inter-subject variability that limit the use of EMG in making a generalized muscle model. Second, EMG is not suitable for patients with muscle disorders. Third, controller dynamic performance for real-time EMG computation. Fourth, errors associated with rapidly changing direction of the muscle extension during exercise/therapy. It is expected that EMG must be introduced in the whole cycle of robot control, which is only possible by hybrid control strategies.

Other challenges that hinder the realization of exoskeletons are: requirement of compact actuators, long term power supplies, improved ergonomics and safety, use of lightweight materials in construction and miniaturization of electronics for reducing the weight.

Another factor that hinders the potential for effective improvement is lack of empathy. Empathy is an important factor in improving therapeutic outcome. However, machines cannot feel empathy, and this creates a hurdle in maximizing the full potential use of robotics in treatment and assistive needs. The exchange of emotional expression, perhaps with a comprehensive brain-computer interface between human and machine, will strengthen the formation of the desired assistance relationship.