High fall risk individuals are constantly threatened by the unpredictability of the occurrence of gait per-turbations, which can happen in a wide range of scenarios during the everyday living. Although these subjects are able to produce reactive responses to counteract theLOB, they are generally not agile and strong enough to avoid falling [40]. According to the overwhelming prevalence and harmful consequences associated with the occurrence of slips or other types of fall, recent literature has attempted to implement fall prevention strategies to mitigate fall incidence among the elderly. Thus, different assistive devices have been robotised personalising the fall prevention techniques according to the user needs and level of mo-bility. This review aims to present current fall prevention solutions for robotic canes, smart walkers and wearable assistive devices (except exoskeletons), as well as their limitations and challenges.
2.3.1 Search Strategy
An exhaustive literature review based on three databases was conducted to have a better knowledge of robotic canes: Scopus, Web of Science and IEEE. In a primordial phase, an iteration table was created for each database, with a set of keyword combinations in order to obtain the best search terms for a successful review. The chosen keyword combinations were “Robot cane” OR “Robot walking aid” OR “Cane type robot”
for Web of Science and IEEE Xplore, and “Robot cane” OR “Robot walking aid” OR “Cane-type robot” OR
“Cane-type robot” AND NOT “sugar cane” for Scopus. A comprehensive survey was also carried out on Scopus and Web of Science to identify smart walkers who have fall prevention or detection strategies. The following keywords were used: ”Walking support”AND fall, ”Smart walker”AND fall, ”Smart rollators”, and
”Walking-aid”AND fall. Due to the limited number of wearable robotic devices for fall prevention, specially to prevent slips, a more cursory investigation was conducted. The points of interest during the analysis of the systems were the following: i) the sensors used and their location; ii) the strategy and algorithm implemented to prevent falls or imbalance events; and iv) the algorithm developed.
2.3.2 Search Results
Taking into account the research on robotic canes, a total of 1506 results were found using the mentioned keyword combinations. The review of these articles was performed based on thePRISMA flowchart, as illustrated in Fig. 2[72]. After removing 423 duplicates and 1026 articles based on title and abstract, resulted in 57 full-text articles assessed for eligibility. Following a complete reading of the eligible articles, 17 were excluded for not meeting the proposed eligibility criteria, which revealed that they were not relevant to the study in question. This selection concludes with 40 articles included in the final review, however only 9 articles present fall prevention strategies. To round out the information, 8 more articles on smart walkers and 3 articles on wearable robotic devices for slip avoidance were included.
2.3.3 Robotic Canes
To protect the user’s safety, the robotic cane must intervene to prevent a fall when an emergency situa-tion is detected. To accomplish successful fall prevensitua-tion, efficient approaches must evaluate numerous characteristics within a small time frame capable of giving the opportunity to the user to regain the bal-ance, namely: i) an analysis of the direction of the fall; ii) the relative position and orientation of the cane at the beginning and conclusion of the fall; and iii) the interacting forces applied to the cane. Electronic sensor-based devices are implemented in these robotic canes to acquire important information of: i) the user, such as the user’s gait status and gait phase recognition; ii) the robot-user interaction by detecting the user intention of movement with forces applied on the cane; as well as iii) important information about
Figure 2: PRISMA flowchart of the Review on Robotic Canes.
the cane itself, such as the cane relative position, velocity and orientation. The following sensors are com-monly implemented: i) axial force/torque sensor; ii)IMU; iii)Laser Range Finder (LRF); iv) force sensor; v) tilt sensor; vi) camera; vii) ultrasonic sensor; and viii) infrared sensor.
Four canes implemented fairly similar fall prevention approaches [43, 73–75], since they shift the robotic cane to a favourable and strategic position based on the user’s attitude and the direction of the fall. As a result, the users will be able to support their body weight on the robotic cane, allowing them to maintain their balance and prevent falling (Fig.3.a). In the case of canes made by [43,74,75], which have a universal joint for the first two and a revolution joint for the last, the angle of the robotic cane rod can be altered and regulated to give extra help. This strategy can strengthen the stability of the cane during the fall, preventing it from toppling over owing to a strong shove. According to Di et al. [43,76], the optimum method for preventing falls using a cane’s tiltable rod is to place it with the falling direction and the tilted direction of the cane stick in the same line but in the opposite direction.
Canes from Fujimoto [77] and Van Lam [78] are based on the inverted pendulum model in order to have a robotic cane in self-balance. Basically, when the user applies forces in the cane, as in a fall event, the cane will assure a favourable position to aid the user throughout the fall. This procedure is accomplished by moving the cane in the direction of the applied forces, allowing the cane to retain its balance and remain upright. Cane from Fujimoto [77] employs the angle of the cane in respect to the plane of displacement as a key parameter, whereas cane from Van Lam [78] uses the sensing forces applied to the cane’s rod as its main parameter (Fig.3.b). These canes, which must self-balance to remain upright, have minimal strength, instability, and safety, making them unsuitable for older users and those with restricted mobility.
The apparent dynamics of the robotic cane can be altered to prevent the user from falling by changing the braking torques of the wheels. Braking system of the Suzuki’s cane [79] limits the speed of the movable
a) b)
Figure 3: Fall prevention methods. (a) Movement of the robotic cane to a favorable and strategic position [43]. (b) Cane self-balance and its capability to maintain balance and move alongside the user [78].
base, so assisting and preventing the user from falling. This passive system is innately safe since they cannot move accidentally; nevertheless, due to the absence of active movement when a fall is detected, it may not be optimal for fall prevention. Because it can only keep the cane from moving, there may be times when it needs to shift position in order to provide a steady and firm support to the user, ensure the balance and stability of the system’sCentre of Mass (CoM), and avoid the fall. Although Ito’s cane [80] is a prototype with certain drawbacks in terms of its not-so-sturdy structure and motion control, which requires the user to lift the 1.2 kg cane during walking, it proposes a novel way for ensuring cane stability when walking in unstable and irregular circuits. Overall, this walking assistance technology, which also attempts to avoid falls, has limited capacity to give security and stability to sustain the body weight in a fall event.
2.3.4 Powered Orthosis
Wearable robotic devices can also stop falls from happening, particularly slips. It does, however, demand certain requirements for detection and actuation. Wearable robotic devices must have technology capable of detecting perturbations in milliseconds, as well as a motorised joint with time-effective actuation. In the literature, there are some examples related to fall prevention. Monaco et al. [40] developed an Active Pelvis Orthosis (APO) to aid with balance recovery following unanticipated slip disturbances. The scientists based their approach on the idea that increasing stiffness at the hip joints could aid participants in recovering from treadmill slip-like disturbances. Their system presented: i) an assistive torque with the ration of 0.2 Nm/kg and duration of 0.25 s; and ii)LOBdetection time between 0.3s and 0.4s. Hopf oscillators were integrated in the detection system. Perturbations were recognised by comparing the actual APO’s hip angles to the hip angles predicted by a pool of adaptive Hopf oscillators in real time [13]. The error between both signals was used as input for an adaptive-threshold algorithm, i.e., the responsible for the detection of perturbations.
With the same purpose, Mioskowska et al. [37] presents a wearable knee assistive device capable of actively extend the trailing leg’s knee by means of a knee brace once a slip perturbation has been detected. The idea is to quickly restore foot contact with the ground and thus extend subject’sBoS. The
(a) (b) (c)
Figure 4: Literature slip-related fall prevention actuation systems. (a) Monaco et al. [40]. (b) Mioskowska et al. [37]. (c) Trkov et al. [81]
average actuation time for device extension was 0.082 and 0.072 seconds, respectively, from the initial 90 to 0 degrees and 60 to 0 degrees. Furthermore, the technology was demonstrated to stretch a human knee more than 30 degrees in 0.4 seconds. There was no indication of detection time.
Like Mioskowska, Trkov et al. [81] developed a Robotic Knee Assistive Device. They do, however, provide assistive knee torque to the leading leg rather than the trailing leg, as Mioskowska did [37]. The magnitude of the desired torque (up to 40Nm) is established by the linear feedback between the actual and desired knee angle positions and velocities, i.e., operates with an impedance and torque feedback control. It can generate the maximum torque magnitude within less than 0.2 s. Despite the absence of detection timings, the authors reveal that this device employsIMUdata as input to the detection algorithm. Figure4depicts a general overview of the actuation system from the 3 previously described studies.
2.3.5 Smart Walkers
Through the search process just a few smart walkers were discovered. Stereo cameras, LRF or force sensors are the main information collected to be used as input in these algorithms. RT Walker [82–84], a passive device equipped with rear wheels with powder brakes, was the subject of experimental tests with different sensors. In [82], twoLRFwere used to create a 7-link human model, which allows the generation of a stability region considering the support polygon formed by the walker and the user’s feet. OneLRF was located at the same height as the user’s hip to measure the distance along theVdirection between the walker and the user. The other laser was placed at the base of the walker and measure the distance between the user’s leg and the walker. Further in [83], two stereo cameras tracked the head, hands, shoulders, and hip to get theThree-Dimensional (3D) upper body model to classify activities including falls. The upper body centroid position extracted by a depth camera was then used to categorise falls regarding their direction [84]. AHidden Markov Model (HMM)detected 98.75% of the falls.
Other studies used similar sensors to gather data for their algorithms. Mou et al. [85] used aLRF and force sensors on the handle to classify three kinds of gaits (festinating gait,Freezing of Gait (FOG)
and normal gait). When a sudden push is detected, the walker stops. Azqueta-Gavaldon et al. [86] used depth cameras to measure the distance between the user’s leg and the rollator. Healthy subjects simulated forward falls, namely freezing of limbs, stumble, andLOB. When the distance between the user and the rollator is higher than a threshold, the rollator stops, preventing a fall (brake activation: 80-90ms) with an overallACCof 95%. Xu et al. [42] also used aLRFand force sensors on the handle. However, theSupport Vector Machine (SVM) was the approach used to classify the state of gait and, consequently, detect a possible fall. If the user is falling the walker stops moving.
Irgenfried et al. [87] developed a walker only with a 6D-force/torque sensors, which information is used as input to a mathematical model of the human body. Data from simulated falls revealed a peak in the sensor values that can be used to detect a possible fall. Once again, the authors suggested stopping or slowing down the walker. Contrarily, Huang et al. [88] used wearable and non-wearable sensors to detect possible falls. The user is instrumented withIMUin the lower limbs and trunk, while the walker is equipped with force sensors on the handlebar. This combination of information allows estimation of the position between the midpoint of the feet and theCentre of Pressure (COP). Different types of falls were simulated and when the possible fall is detected, the walker brakes and stops.
2.3.6 Clinical Highlights and Future Directions
Studies about robotic canes present some limitations. Clinical gait experiments involving the elderly or those with conditioned mobility are lacking, demonstrating that the systems developed were not tested in the real environments. Furthermore, only one cane offers an active fall prevention method with a mecha-nism that uses cane movement that does not require the user to wear wearable sensors. Another significant drawback is that the sensor systems of many robotic canes in the literature do not identify the user’s gait phases, which is an important element for gait monitoring and possibly fall detection. Considering pow-ered orthoses, the ideal scenario would include the assistive torque supply to all the lower limb joints from both legs upon a slip. Trkov et al. [81] and Mioskowska et al. [37] assisted only one joint. In fact, such an approach would increase the computationally and mechanically complexity of the fall prevention strategy possibly contributing to an ineffective fall prevention. However, scientific literature is poor regarding real-environment tests, mechanical information about the actuation units, importance quantification of each joint in the biomechanical response to a slip event. Smart walkers’ prevention strategy consisted in stop the robotic device when the possible fall was detected. Although it ensures safety for forward falls, it is required to investigate lateral and backward falls to mitigate fall occurrences. Tests in a real setting are necessary, just like with robotic canes and powered orthoses. Furthermore, it is feasible to confirm that there are just a few available options for these robotic devices on the market.