Fall inducing movable platform (FIMP) for overground trips and slips

Background The study of falls and fall prevention/intervention devices requires the recording of true falls incidence. However, true falls are rare, random, and difficult to collect in real world settings. A system capable of producing falls in an ecologically valid manner will be very helpful in collecting the data necessary to advance our understanding of the neuro and musculoskeletal mechanisms underpinning real-world falls events. Methods A fall inducing movable platform (FIMP) was designed to arrest or accelerate a subject’s ankle to induce a trip or slip. The ankle was arrested posteriorly with an electromagnetic brake and accelerated anteriorly with a motor. A power spring was connected in series between the ankle and the brake/motor to allow freedom of movement (system transparency) when a fall is not being induced. A gait phase detection algorithm was also created to enable precise activation of the fall inducing mechanisms. Statistical Parametric Mapping (SPM1D) and one-way repeated measure ANOVA were used to evaluate the ability of the FIMP to induce a trip or slip. Results During FIMP induced trips, the brake activates at the terminal swing or mid swing gait phase to induce the lowering or skipping strategies, respectively. For the lowering strategy, the characteristic leg lowering and subsequent contralateral leg swing was seen in all subjects. Likewise, for the skipping strategy, all subjects skipped forward on the perturbed leg. Slip was induced by FIMP by using a motor to impart unwanted forward acceleration to the ankle with the help of friction-reducing ground sliding sheets. Joint stiffening was observed during the slips, and subjects universally adopted the surfing strategy after the initial slip. Conclusion The results indicate that FIMP can induce ecologically valid falls under controlled laboratory conditions. The use of SPM1D in conjunction with FIMP allows for the time varying statistical quantification of trip and slip reactive kinematics events. With future research, fall recovery anomalies in subjects can now also be systematically evaluated through the assessment of other neuromuscular variables such as joint forces, muscle activation and muscle forces.


Background
On average, elderly population(s) (age≥65) encounter falls incidence at a weighted mean rate of 0.65 times annually, and this rate doubles for individuals above 75 years old [1]. The importance of fall related solutions increases as the world population ages. However, the relative rarity and variability of real-world fall incidence has greatly impeded the progress of fall pre-* Correspondence: jker@ntu.edu.sg 1 Nanyang Technological University, Rehabilitation Research Institute of Singapore, 11 Mandalay Road, #14-03, 308232 Singapore Full list of author information is available at the end of the article vention/intervention/mitigation research. It is impracticable to request the elderly to wear motion capture sensors all year round only to capture one instance of fall. Hence, systems that induce falls in a safe controlled environment are essential to progress our understanding of the neuromuscular mechanisms underpinning falls events. Trips and slips are the focus of this work as they represents the majority of externally induced falls [2].
Several efforts have been made to create a system to induce an ecologically valid trip event. The most common types of fall induction systems generally rely on instrumented treadmill systems to perturb a participant during walking gait trials [3,4,5,6]. These treadmill systems allow for precise and accurate velocity control that conventional overground walking systems are not able to replicate. Obstacles and perturbations can also be rendered easily as many mechanisms can be hidden around and under the instrumented treadmill systems. Though there are obvious benefits for the use of instrumented treadmill systems within the falls literature, it is widely known that an individual's gait pattern changes when walking on a treadmill versus over ground. Differences in an individual's kinematics [7,8,9], joint moments and muscular activation [10,11,9] have been well documented. Additionally, control of the treadmill after fall onset is critical to replicate true fall dynamics. The treadmill must travel exactly to the speed of the recovery limb to prevent artificially widening or narrowing their Base of Support (BoS).
Another type of trip induction system uses overground walking to generate more realistic real-world type falls. However, the engineering problem with these types of systems are that they need to account for the changes in the subject's linear position during walking gait. Multiple hidden obstacles have been built to induce trips along a fixed pathway [12,13,14,15]. Since different gait phases induce different recovery strategies, these obstacles have to be densely packed to synchronise the simulated trip with the correct gait phase [12,16]. Although the varying gait patterns can be accounted for with these obstacles, only one obstacle can be engaged at every instance -rendering this fall induction system to be inefficient cost expensive to implement, making them a non-practical research solution for laboratories globally, particularly those located in economically developing countries. A more cost effective method to this engineering problem is to develop a brake and motor system, in place of the multiple obstacles, to induce falls over a distance [17]. Nonetheless, regardless of the fall induction mechanisms used, the system must utilise an overhanging railing harness system for subject's safety, limiting its use to a designated location. Moreover, this overhanging railing harness system, which can have high inertia, alters the gait mechanics of the subject/patient under investigation.
Slip experiments commonly depended on using a split-belt treadmill [18] or a motorised floor plate [19] to provide the sudden gain in acceleration during a slip. However, the limited actuation distance of these devices mean that slipping is only allowed over a short distance. Once the motorised plate or treadmill stops, the subject can generally regain stability immediately, unlike a real-world slip scenario where velocity is slowly decreased. Even if the deceleration of the motorised plate or treadmill are controlled, it is difficult to match the intended joint kinematics. Furthermore, as previously mentioned, treadmill-walking changes the gait mechanics of the subject, arguably preventing the recording of a true transition from walking to slipping. A sliding plate [20] performs better in replicating a true slip scenario, but they are typically constraint by their sliding distance and the synchronisation of the leg touchdown position. Another method of inducing slips relies on sliding over a slippery surface [21,22,23]. This method replicates the true slipping scenario, but similar to a trip, they are constrained to a high inertia overhanging harness which can prevent the recording of a true transition from walking to slipping.
To the best of our knowledge, there exists one fall inducing robot for overground walking that allows changing heading angle and does not constraint the walking path [24]. This robot induced fall-like imbalance by providing perturbation at the pelvis. However, this method of fall induction bypasses the lower limbs reactive responses that are present in real-world fall scenarios. In addition, the changes in the dynamics of the lower limbs caused by obstacles and slippery surfaces are disregarded. Therein, they can affect an individual's falls recovery strategy. For example, a leg that encountered an obstacle during a trip will experience sudden deceleration and hence, the user would likely require more time to widen their BoS. Instead of arresting the leg movement during a trip, the forward pelvis perturbation from the robot may actually assist the subject to widen their BoS, resulting in improved stability.
The purpose of this research is to develop a Fall Inducing Movable Platform (FIMP). It is innovated for realistic fall induction in efforts to induce more realistic real-world falls scenario (Fig. 1). FIMP has the following characteristics: • Allows change in heading angle, velocity and other gait patterns. • Provides usage in a variety of relatively level ground without space constraints. • Minimises mechanical inertia induced by the safety harness system applied to the subject. • Induces more ecologically valid falls via ankle perturbations. • Induces trip and slip with the same system. • Creates a random, unexpected perturbation. The design and methodology of FIMP is described in this paper, starting with the hardware, followed by the software. Additionally, experiments have been conducted to determine the transparency and kinematic effects of the trip and slip mechanisms of FIMP. Only trips and slips inducing capabilities were developed for this generation of FIMP system as they are the most common types of real-world fall scenarios [25,26,2].
It can also be argued that an additional drawback of other publications in this field of research was the analysis method previously used to analyse time varying perturbations. Therein, they do not analyse time varying or continuum human movement data as a time series, but multiple discrete or zero dimensional (0D) time points (e.g., maximum, minima, mean etc.). Here, a novel analysis tool called 1-Dimensional Statistical Parametric Mapping (SPM1D) is introduced for fall recovery analysis [27]. It was originally developed within the neuroscience field for the analysis of changes in brain activation differences in functional and Magnetic Resonance Imaging (MRI) scan [28]. SPM1D was recently validated in the journal of statistical software for the analysis of time normalised signals of human movement (i.e., kinematics and kinetics) [29], which makes it well suited for fall-related research.

Methods
Several engineering components were necessary to realise the development of the FIMP system: • Mobile Platform • Trip and Slip mechanism • Subject following and support algorithm • Gait phase detection algorithm • User Interface and System Control

Mobile Platform
The mobile platform of FIMP had a dimension of 175cm long by 115cm wide by 208cm tall (Fig. 1). This dimension was qualitatively tested to minimise the footprint necessary to prevent the lower limbs from hitting the platform while mitigating the sense of claustrophobia. Two motorised wheels were mounted along the central axis of the FIMP such that the wheels can rotate the FIMP with a zero turning radius. The 2 DC motors obtained from Motion Tech Motor had a power rating of 250W each, and came equipped with electromagnetic brakes. They were controlled via a Sabertooth 2x32 motor driver. The motor driver was configured in mixed mode, which accepted 2 inputs; one input signal that controlled the forward speed, and another that controlled the turning speed and radius via the wheels differential speed. Five caster wheels were placed around the platform to ensure stability during movement. A 24V, 35AH Lipo battery powered the entire system.

Trip mechanism
The trip scenario was induced via wire ropes attached to the ankle of the subject (green dotted circle in Fig.   1). The continuous movement of the legs during gait requires the wire ropes to be freely moving. However, these freely moving ropes were often slack and caused a delay between perturbation activation time to force output through the wires. This delay in force transfer can contribute to variability in the fall recovery motion, and creates the risk of uncontrolled tripping over the slack ropes. The solution to this problem was to mount a small power spring within the reel of the wire rope. This power spring constantly reels back with a constant torque, and thereby keeping the wire taut to the ankle.
Trips were induced via an electromagnetic brake connected to the ankle using a similar power spring attached wire rope. The brake was supplied from SIN-FONIA (ERS-260L) with 8Nm static friction torque (Fig. 2). When a trip scenario was required, the brake activates and generates up to 347.8N of force (Table  2). Otherwise, the brake remains powered off and the power spring pulls with a passive force of 3.0N and a dynamic force of 5.9N (Table 1).

Slip mechanism
The slip scenario was induced via a DC motor. This approach was chosen as the sudden acceleration for a slipping stance leg could only be achieved with a pull. The DC motor was obtained from XaJong Co. Ltd (M35SWG-2436), with continuous torque of 1Nm and speed of 3600rpm (Fig. 3). A wrap spring clutch from Tiny-Clutch -Helander Product, Inc was attached to the output of the DC motor to disengage the inertia of the motor from the power spring. The DC motor reels in the wire rope wound around the exterior shell of the power spring after engaging the clutch with 0.12N m of torque. The maximum pulling force exerted by the slip assembly was 110.3N (Table 2). When the motor was disengaged, the passive tension force caused by the power spring was 1.0N and 3.5N dynamically ( Table  1).
When conducting slip trials, the subject's walking pathway was covered with 2 layers of sliding sheet. The coefficient of friction between the 2 layers of sliding sheet was approximately 0.15 for both static and kinetic scenarios. This inclusion of sliding sheets replicated the true slip conditions, whereby the slip foot continues to slide forward even after the initial pull by the slip motor.
When both trip and slip assemblies were utilised together, subjects within FIMP will only feel an overall static tension of 2.0N and dynamic tension of 2.4N around their ankle as the opposing forces cancel each other. Hence, during either trip or slip experiments, both assemblies were attached to the ankle.

Subject Follower Algorithm
The subject follower algorithm allowed the user to walk around a variety of level ground while changing heading angle, velocity and gait pattern. It also Figure 3 The slip motor was mounted horizontally with its output shaft connected to a bi-directional clutch and finally to a constant force spring. To prevent over-reeling of the motor, a thick acrylic sheet with a hole prevented the cable endings from being reeled in. The hole also guided the wire rope into the reel getting tangled. A fuse was added in series with the motor to prevent excessive torque. reduced the perceivable inertia of safety harness as it was attached to the top of the following platform.
Manually pushing the platform was acceptable, but it causes varying degrees of fall severity as the moving platform will alter the pulling and braking force of the fall mechanisms if they were not synchronised. Hence, the subject follower algorithm is important to obtain a constant fall severity. FIMP was motorised with 2 wheels along the central axis to allow for seamless synchronisation of the fall mechanism(s) and the movement of the platform. An Intel RealSense Depth Camera D435 was used to estimate the position of the subject within FIMP. Usage of the depth camera allowed for subject tracking without any physical contact, eliminating any interference with normal walking gait. The camera was mounted at the back of FIMP, pointing directly forward at the centre of FIMP (Fig. 4). The estimated distance of the subject from the camera was obtained by averaging all the depth pixels of the camera. A Proportional Derivative (PD) controller controlled the speed of the motorised wheels to maintain a distance of 60 cm from the camera.
The angle of 2 vectors (subject's average pixel centre to camera, and FIMP's forward axis, Fig. 5) was used to re-centre the subject within FIMP when the subject turned. The angle was input into another PD controller to control the differential speed of two wheels via a motor driver. Gait Phase detection Algorithm Different types of falls call for different neuromuscular recovery strategies. For trips, an elevating strategy is used when an easy to overcome external perturbation occurs at the early to mid-swing gait phase [12,30]. When the obstacle or perturbation becomes large enough, a skipping strategy is utilised [31]. This is due to the perturbed leg being arrested from moving forward by the perturbation. Hence, the contralateral leg must skip forward to reinstate a suitable base of support and regain stability. A lowering strategy is used during late-swing trips, where the perturbed leg lowers immediately after perturbation and an additional step is taken to clear the obstacle [12,30].
Backwards falls are normally caused by slips that occur during the initial stance phase, as the contralateral stance leg is responsible for transferring the momentum of the body to the ground. When there is a lack of ground friction, the forward momentum pushes the stance leg forward, creating a slip. Slip's recovery strategy varies significantly with prior knowledge on the friction of the ground [16]. The initial slip recovery response will see the slipped foot immediately adopt a flat-footed configuration to the ground. Then, the contralateral leg is placed behind the Center of Mass (CoM) to provide a recovery moment. These responses require extending and flexing the perturbed and unperturbed leg respectively. Subsequent walking gait following the initial slip response will follow the surfing strategy which slides the swing foot forward rather than stepping off quickly during swing phase [16]. This is done to increase the contact area between the foot and the ground, thereby increasing frictional forces. Since the trip response differs based on tripped timing and slip must be triggered at initial stance phase, the correct identification of gait phase is important to induce the correct fall condition.
Two Inertial Measurement Unit (IMU)s were placed on the torso and left thigh of subjects to detect their body dynamics. Heading angle from the torso IMU's orientation vector were used to reorient the left thigh IMU's sagittal plane angular velocity and orientation output. Subsequently, these 2 axes (reoriented thigh IMU's angular velocity and orientation output) were dual-passed through a low-pass and high-pass Butterworth filter with cut-off frequency at 3Hz and 0.4Hz respectively. The Levenberg Marquardt Algorithm (LMA) was used to fit the filtered axes into a unit sine wave (sine wave with constant magnitude and offset). Finally, gait phase angle was calculated by doing an inverse tangent of the filtered and fitted sagittal plane's orientation over the angular velocity.
User Interface and System Control Integration of the system is as shown in Fig. 6. The subject follower algorithm ensured that the platform follows the subject around. Upon a randomised triggering of a button from the investigator, the microcontroller (Raspberry Pi 3B+) automatically started the gait phase identification process. The gait phase angle was separated into 20 equal segments from −π to π for easy segmentation of the gait phase. Once the gait phase has entered the pre-selected segment for the corresponding induced fall, the desired fall mechanism activates and FIMP stops. This induced a forward trip or a backwards slip depending on which fall inducing mechanism was used. Subjects were instructed to recover from the perturbation as fast as they could and stand straight after the fall. An emergency stop button was wired between the Lithium polymer battery power source and the wheels and fall inducing mechanism.
The automatic triggering of the fall mechanisms based on detected gait phase solved the problem of inducing different fall recovery strategies. Moreover, the stopping of FIMP and the fall mechanisms can now be synchronised.
The types of fall recovery strategies induced by FIMP are listed in Table 3. All the listed fall types were evaluated in this research except for the elevating strategy. This is due to the short time frame available to perturb and release the perturbed leg in order to reliably induce elevating strategy. More work will be conducted in the future to optimize the activation duration before the elevating strategy can be studied. Experimental Protocols All trials were approved by the Institutional Review Board of Nanyang Technological Universtiy (IRB-2018-08-006). The gait phase detection system relied only on IMUs placed anteriorly on the subject's left thigh and torso, although other IMUs on the right thigh and shanks were also present. The IMUs were obtained from Yost Labs, and they were connected to a central micro-controller (Raspberry Pi 3b+) using a USB hub. A mean sampling rate of 333Hz was obtained after using Yost Labs proprietary Q-Comp filter. Each sample for gait analysis consisted of the body's local acceleration, local angular velocity and global quaternion orientation output. The entire gait phase detection system was attached to a big Velcro belt, which was worn around the subject's waist (Fig.  7). The waistband was tightened to the level where it did not slide down and yet remained sufficiently loose to allow for comfortable hip and lumbar flexion. 16 Qualysis Miqus M3 mocap cameras and 2 video cameras were used to track and capture subject's motion. The cameras were configured to sample at 200Hz. 53 Markers were placed on the L/R (Left and Right) Forehead, L/R Back head, clavicle, sternum, C7, right back, T10, L/R shoulder, L/R upper arm, L/R elbow, L/R forearm, L/R dorsal tubercle of radius, L/R styloid process of ulna, L/R middle finger's metacarpophalangeal joint, L/R anterior superior iliac spine, L/R posterior superior iliac spine, L/R thigh (4 markers), L/R lateral knee epicondyle, L/R tibia (4 markers), L/R ankle medial malleolus, L/R ankle lateral malleolus, L/R heel and finally L/R middle toe metacarpophalangeal joint (Fig. 7).
A total of 7 subjects (2 females and 5 males) were recruited for this experiment. They have an average age of 25 ± 0.94 years, average height of 168 ± 2.36 cm, average weight of 58 ± 6.24 kg, and an average Body Mass Index (BMI) of 20 ± 1.69. All subjects had no history of locomotor impairment, neurological impairment or cardiovascular diseases.
For each subject, they were asked to perform 3 types of walking trials at their preferred walking gait velocity along a straight path: 1) NormalWalking (NW) : 5 trials of walking without ankle cable and without harness.  System control of FIMP. Subject starts walking in FIMP. RealSense camera detects the user's position and drives the wheels to follow the subject. As FIMP follows, it polls for investigator trigger via a button pressed for intention to trip the subject. When the button is pressed, the system automatically initiates the gait phase detection algorithm. Upon the correctly detected gait phase, the system immediately stops the wheels of FIMP and activates the trip brake. Subjects are instructed to recover and stand straight after the trip. the power spring, and its force may influence the way the subject walks. Section studies the effect of FIMP during normal walking (i.e. system transparency).
Due to the space constraints of the laboratory and calibrated motion capture volume (9m by 5m by 2m), all the falls were induced 3 meters within the middle of the subject's walking pathway (6th to 9th step of a 15 steps pathway). Perturbations randomness were introduced by walking multiple trials and activating the falls only during some of the trials. An investigator constantly followed behind FIMP to activate the fall via the control switches and intervene if necessary by depressing the emergency stop button. When conducting slip trials, the low-friction sliding sheet mentioned above were placed along the walking pathway of the subject. Trips (MS and TS) and normal walking trials (NW and SW) were conducted on the normal laboratory floor surface.
All captured data were filtered with a 2 nd order Butterworth low pass filter at 6Hz.

FIMP Transparency
This section investigates the effect of FIMP (Transparency) during normal walking gait. An ideal fall inducing system should not affect the normal walking gait of the subject under any conditions. FIMP requires users to be attached to its fall inducing mechanisms via ankle straps which may affect the normal walking gait. Hence, the normal walking (NW, no ankle straps and no safety harness) trials were compared to the strap walking (SW, walking with straps and safety harness) trials to verify what influence, if any, the FIMP may have on the subject's walking gait. For this analysis, time varying sagittal plane hip, knee and ankle joint kinematics were compared using SPM1D.
Conventional methods of comparing different trials typically used 0D or discrete time points of a time varying signal like maximums, minimums, standard deviations, which we can argue are deemed unsuitable for falls type research and this experiment specifically. This is because the entire waveform is not being statistically analysed, predisposing research's from making both type 1 and type 2 errors with this analysis [32]. Therefore, a topological method for detecting field changes in smooth n-dimensional continua called SPM1D [27] was employed. Specifically, SPM1D analysis of t-test statistics (SPM1D{t}) allows for the time-normalised analysis of sagittal plane flexion joint angles at each point in the time series, which forms a statistical parametric map. The temporal smoothness based on the average temporal gradient can then be estimated using random field theory [27] to generate the test statistic threshold. Significance is achieved only when the value of the test statistic exceeds the test statistic threshold. Hence, two datasets with high time-varying variance require a breach at a high threshold to be classified as statistically different.
The stride of each gait cycle of every trial were segmented by heel strike. For every stride, it was normalized for fixed stance and swing phase duration. The stance phase was normalised to fit the first 60% of the gait cycle and the remaining 40% for the swing phase. The toe-off of each leg was used to demarcate the end of stance phase. The left and right legs were compared separately as the ankle strap was only attached to the left ankle during SW trials. A two sample t-test was conducted with SPM1D analysis, with an alpha threshold of 0.05. The relative effect NW versus SW SPM1D{t} was also presented in the format of a colour map for both combined and individual trials ( Fig. 8 and Fig. 9). The colour map allows for readers to understand the relative effects of regions where statistical significance were observed. For t tests, the level of significance was separated into intervals of 1 standard deviation as t tests calculate the relative difference between 2 groups over their combined standard error of mean.

Left Leg
Results indicate that there was little to no differences between the hip and knee flexion kinematics of the left leg between NW and SW (Fig. 8). This implies that the ankle straps and safety harness did not affect the movement of the thigh and shank. On the contrary, there were significant differences in the ankle plantar/dorsi flexion kinematics for the left leg when the NW and SW conditions were compared. However, the magnitude of effect of the strap was not consistent among all the subjects, indicating that certain subjects were more influenced by the ankle straps heterogeneously. Even among those who were affected, this larger dorsiflexion is thought to have minimal effect on balance recovery during fall events. This is because the excess dorsiflexion happens during mid-swing, which does not generate any appreciable internal or external forces or torque experience on the subject. The only effect may be a slight increase in fatigue on the muscles responsible for dorsiflexion (e.g. tibialis anterior).
To ensure greater FIMP transparency and a greater resemblance to normal walking, it is still important to reduce the effect of the ankle straps. Several factors may contribute to the observed differences in ankle plantar/dorsiflexion. Firstly, the act of attaching the ankle straps requires interaction with the reflective markers on the ankle which may have affected their calibrated alignment. Secondly, the ankle strap can move during plantar/dorsiflexion of the swing phase, affecting the position of the shank malleolus markers, which were used to define the position of the ankle joint centre. Next, the tension force on the ankle strap may have require higher dorsiflexion to compensate for a lower toe clearance height. Lastly, the tightness of the ankle straps could have prevented the normal functioning of the ankle. Therefore, future research is recommended to determine the causal factors underpinning the observed differences. Guiding future research, it is speculated that the engineering solution to this problem is to improve the force characteristics of the power spring and improve methods to control the tightness of the ankle straps.

Right Leg
The right leg that did not have a strap attached to it shows no significant differences between NW and SW conditions for all lower limb joints (Fig. 9). This result also implies that the harness mounted on the movable FIMP did not affect the gait of the subject, and the method of using subject follower algorithm to reduce harness inertia is a success.

FIMP Terminal-Swing Tripping Effectiveness
Perturbations during the terminal-swing gait phase induced lowering recovery strategy [12], which corresponded to the lowering of the perturbed leg followed by the contralateral leg to overcome the obstacle ( Fig.  10 and Additional file 1). This section examines the effect of the trip on the ipsilateral leg (perturbed left leg) and the contralateral leg (right leg). One-way repeated measure Analysis of Variance (ANOVA) with SPM1D were employed to identify the time instance when the perturbed gait cycle deviated from NW and SW. Again, the chosen alpha level was 0.05. Similar to the t tests conducted in the gait transparency section Second row indicates the Two-sample t-test on the data in top row. The third and last rows show a colour map highlighting the t-test significant differences for the all subjects' combined and their individual NW vs SW trials. Significant differences are seen in the ankle joint during the mid-swing phase for the majority of the subjects. There are also differences found in the knee during the mid-swing phase, but it is less consistent among all the subjects.

Figure 9
Comparison of NormalWalking (NW) vs StrapWalking (SW) trials for the right leg using SPM1D. The right leg is not attached to any ankle straps. Top row indicates the mean hip, knee and ankle flexion angles for all subjects and standard deviation clouds; mean (± st.dev). The third and last rows show a colour map highlighting the t-test significant differences for the all subjects' combined and their individual NW vs SW trials. The last row shows the significant different effect size of individuals. No significant differences are found in any of the joints for the majority of the subjects above, the ANOVA relative effect colour maps were also presented. However, the intervals used to visualise statistical significance were Mean Square of Error (MSE). This was used as the F value of ANOVA was calculated as the mean of group effects over MSE.

Left Leg
The gait cycles of the left leg containing the instances of perturbation were averaged for all subjects' Termi-nalSwing (TS) trip trials as shown in Fig. 11. Results show that the ipsilateral hip joint did not exhibit significant differences between NW and SW conditions. However, the knee joint experienced less extension, while the ankle joint underwent higher plantarflexion at the end of the swing phase.
The lack of difference at the hip joint is in-line with the perturbation occurring at the terminal-swing phase. During terminal-swing phase, the hip flexion muscles are contracted near the maximum flexed position to allow momentum to swing the passive knees forward. As perturbation is applied, the disturbance did not affect the hip joint as significantly as the knee and ankle joints because it is thought that the hip joint has higher stiffness created by its activated muscles.
The effect of the tripping mechanism was most evident in the knee joint during the terminal-swing phase. This is because, the perturbation force vector acts on the knee joint at a significantly larger moment arm than the ankle joint. The low stiffness knee joint caused by its passive muscles during swing phase also resulted in the large deviation from the normal trajectory during disturbance.
The ankle joint was less affected by the perturbation force due to the perturbation force proximity to the ankle rotation axis. However, the foot still had higher plantar-flexion angle than the normal trajectory at the end of the gait phase to increase the air-time of the body for widening of the contralateral limb. This higher plantarflexion has also been observed in other trip studies [33,34].

Right Leg
The right leg gait cycles immediately after the left leg's perturbation were averaged for all subjects' TS trials (Fig. 12). The stance phase of the right leg happened concurrently with the swing phase of the left leg, which is where the perturbation was applied. Immediately after the perturbation, the contralateral right leg had to rapidly raise the thigh while the ipsilateral left leg extended the knee to cushion the trip impact, and plantarflexes the ankle to increase the time off the ground. This rise in hip flexion angle is observed in the second row of the SPM1D plot for the hip. Significant differences were found even before the swing phase has initiated (swing phase starts at 60% stride, demarcated by the vertical dotted line). Subsequently, the hip joint continued in its flexion trajectory to bring the thigh and shank forward, extending the body's BoS.
The right knee experiences higher than normal flexion as observed by the significant differences seen at the terminal-swing phase. This higher knee flexion was caused by shorter than normal swing phase, preventing normal knee extension. Consequently, the higher knee flexion angle allows more of the touchdown impact to be absorbed by the knee joint. The ankle had more pronounced plantarflexion at the initial swing phase in preparation for an earlier than usual heel strike, and also may be acting as a spring to absorb the higher touchdown impact.
Subject 7 walked with a pronounced upright fashion (similar to a ballroom dancer), leading to a walking posture with an elevated forward leaning angle as compared to other subjects. This caused the terminalswing perturbation to severely affect this subject's knee kinematics as his thigh needed to be raised much faster to compensate for the faster falling momentum.

Discussion
The terminal-swing tripping system successfully induced the lower strategy as described in the literature [12,30]. The induced trip kinematics and recovery strategies were replicated across subjects, suggesting that the FIMP system can induce ecologically valid terminal-swing trips. Future reliability based research is recommended

FIMP Mid-Swing Tripping Effectiveness
Perturbations during the mid-swing gait phase normally induced the elevating recovery strategy [12], corresponding to an immediate rise of the perturbed leg to overcome the obstacle ( Fig. 13 and Additional file 1). However, in this experiment, the brake's activation timing was lengthened to arrest the leg's kinematics for the majority of the swing phase. This mimics situations where the swing leg gets tangled on some rope or is in contact with some tall obstacles that the elevated toe clearance alone cannot overcome. Therefore, the skipping strategy that generally accompanies harder to overcome obstacles was observed. Similarly, as per previous analyses, ANOVA with SPM1D analysis were used for statistical analysis. An alpha of 0.05 was implemented.

Left Leg
The gait cycles of the left leg with perturbations were averaged for all subjects' MidSwing (MS) trip trials and shown in Fig. 14. The results indicate that there were significant differences in all 3 joints during the swing phase. Due to the tripping mechanism activating at the early to mid-swing phase, the hip had lower flexion throughout the majority of the swing phase. On the contrary, the knee joint experienced higher flexion angles during the mid to late swing phase. The ankle displayed significantly higher plantarflexion for the majority of the swing phase. The appearance of significant differences throughout the swing phase indicate that the tripping mechanism had successfully tripped the subject at the early to mid-swing phase.
The reduced hip flexion angle in the swing phase was caused by the tripping mechanism preventing the left leg from moving forward. This kinematic change was generally not observed in the terminal-swing trip because the hip joint has already reached the maximum flexion pose at the instance of trip. The hip flexion continues to increase slightly after the perturbation because of 1) the excessive forward momentum, 2) the forward leaning of the upper body and 3) the deactivation of the tripping brake only after 250 ms.
It was observed that the knee joint experienced a decrease in knee extension, similar to terminal-swing trip, but to a greater degree. This happened because the ankle was arrested much earlier than terminalswing, preventing majority of the knee extension from occurring. Significant ankle plantarflexion occurred throughout the whole swing phase due to the need to reach for stable ground, even as the brake continued to hold onto the ankle.
It is hypothesised that the hip flexion angle will increase instead of decreasing (as compared to NW and SW) when the brake activation timing is shortened. It is believed the shortened brake activation will induce the elevating strategy where the perturbed leg rise above and over the obstacle. However, the knee will remain in the current elevated flexed position due to the forward leaning of the body, shortening the time to heel strike.

Right Leg
The gait cycles of the right leg immediately after the MS perturbation on the left leg were averaged across all subjects and shown in Fig. 15. Undeniably, there are significant differences in hip flexion angle, happening as early as the start of the swing phase. The knee joint experienced lower flexion during the swing phase and the ankle had a relatively constant joint angles throughout the entire gait cycle.
The hip joint showed a significant difference in the early stance phase because the early mid-swing perturbation on the left leg corresponded to the beginning of the stance phase on the right leg. As the perturbation occurred, the whole body leaned forward, exerting its entire body weight on the right leg alone. This caused excessive hip joint flexion during the stance phase. As the left leg remains immobilized by the brake, all subjects reduced their trip-induced, unwanted angular momentum by skipping the right leg forward. This skip was observed to widen the BoS and allowed stability to be regained.
There was no difference in the knee flexion angle during stance phase as the whole body pivoted on the ankle and skipped forward rather than bending the knees to absorb the unwanted momentum. During the swing phase (during the skip), reduced knee flexion was observed as the legs were pushing off at the beginning and were trying to touchdown as fast as possible at the end. It is believed this motor control strategy was adopted to achieve a wider BoS along with the increased hip flexion, making the dynamic system more stable The ankle joint acted as the pivot for the whole body to lean forward during the stance phase. Additionally, the ankle joint is responsible for generating propulsion force during push off to allow for the skipping action to occur. This was observed by the elevated ankle plantarflexion during the end stance phase for push off. In the swing phase, all subjects over compensated for the larger plantarflexion during push off with a larger than normal dorsiflexion (seen by the valley during the ankle swing phase). However, most subjects quickly recover and revert back to normal swing kinematics. The last row shows the significant different effect size of individuals. Significant differences are observed in the knee and ankle joints near the terminal-swing phase.

Figure 12
One-way repeated measure ANOVA comparison of NormalWalking (NW) vs StrapWalking (SW) vs TerminalSwing (TS) trials for the right leg with SPM1D. The right leg is attached not attached to any ankle straps. Top row indicates the mean hip, knee and ankle flexion angles for all subjects and standard deviation clouds; mean (± st.dev). Second row indicates the ANOVA test on the data shown in top row. Third row shows a colour map highlighting significant differences of the ANOVA result in second row. The last row shows the significant different effect size of individuals. The hip joints have the most pronounced differences when comparing to NW and SW. The knee joint experiences lower extension angle at the terminal-swing phase, while the ankle joint has lower plantarflexion at the early swing phase.

Discussion
The induced mid-swing trip of the current experiment managed to immobilize the left ankle at the early to mid-swing phase. However, the recovery strategies employed did not follow the elevating strategy that is widely accepted as a mid-swing trip [35]. Instead, subjects skipped forward on their contralateral right leg to increase their BoS. This difference is likely attributed to the duration of the brake activation duration. This phenomenon has previously been reported but with the use of a different fall inducing method [31]. Since the current experiment arrested the ankle for 250 ms, the duration for fall recovery has already been expended by more than half. It is likely that certain proprioceptive reflex pathway senses the longer than normal perturbation and switch over to the skipping strategy. Nonetheless, these results showed that skipping strategy (a proven strategy of mid-swing trips) can be repeated, even among multiple subjects.

FIMP Slipping Effectiveness
Slip is induced at the beginning of early stance. Subjectively, subjects will only feel a loss of balance when sufficient body weight has been placed on the stance leg before pulling. Pulling at the late swing phase will induce a similar reaction to lowering strategy. Slip recovery likely requires subjects to stiffen their hip joint muscles [18], while adopting a flat-footed posture to reduce push-off's forward velocity [16] (Fig. 16 and Additional file 1).
Similarly, ANOVA with SPM1D analysis were used for statistical analysis.

Left Leg
The gait cycles of the left leg with perturbations were averaged for all subjects' slip trials (SL) and shown in Fig. 17. The results indicate that there were significant differences in all 3 joints during the swing phase. Since the slip motor mechanism pulled on the left leg at the beginning of the stance phase, significant differences were observed at the start of the gait cycle. The hip flexion angle remained relatively constant throughout the entire stance phase as hip muscles stiffened to prevent a backwards fall due to extension of the torso. Thereafter, during the swing phase, the left hip likely flexes to widen the BoS for stability while adopting a flat-footed surfing posture.
The knee joint began at a similar flexion angle as normal gait, but it flexes as slip was induced to lower the CoM for greater stability. The knee flexion was also caused by the upper body having a larger inertia than the lower limbs. As the upper body slid forward uncontrollably, the knees bent to absorb the unwanted forward velocity. The left knee maintained its flexion angle during the swing phase to adopt a surfing strategy which maximizes the ground to foot friction.
As the motor pulled on the ankle joint during the beginning of the stance phase, it increased the ankle plantarflexion. Thereafter, subjects immediately reacted to keep their foot flat on the ground to increase their friction with the ground, resulting in a similar trajectory as normal walking during the stance phase. However, significant differences were still observed in the swing phase because the subject has now adopted the surfing strategy, whereby the foot is continuously kept in constant contact with the ground. This caused the ankle to have a near zero flexion angle in the swing phase.

Right Leg
The gait cycles of the right leg that contained the instances of slip initiation on the left leg were averaged on all subjects and shown in Fig. 18. The most significant difference was observed near constant knee and ankle flexion angles during the swing phase. These changes were likely attributed to the adoption of the surfing strategy to decrease toe clearance and ankle push-off forward velocity. Reduced toe clearance allows subjects to touchdown sooner for stability and larger ground-foot frictional area. Minimizing the push-off forward velocity also reduces the chances of another slippage.   Anecdotally, subjects were observed to adopt a cautious walking gait, similar to the surfing strategy after the initial slip trial, as soon as they stepped onto the sliding sheet. This behaviour was also seen when individuals are walking on a slippery surface [16,36], enforcing the ecological validity of this slip inducing mechanism. This change in gait pattern also created larger variability in all of the hip, knee and ankle joints, even during the stance phase, when slip has not occurred. This change in gait pattern was most significant in the ankle as the cautious walking gait requires a flat-foot sliding posture to increase ground-foot friction.

Discussion
The slip inducing mechanism of FIMP relies on both the motor for pulling force and the sliding sheet to reduce the friction of coefficient. The sliding sheet creates a near-slip sensation while the motor allows for accurate control of the slip timing. No slip has occurred on the sliding sheet without the activation of the slip motor.
After the initiation of the slip, subjects likely stiffened their joints to reduce the posterior movement of their CoM. Thereafter, the near constant hip, knee and ankle kinematics of the perturbed leg implies that subjects have adopted a flat footed surfing strategy to maintain high friction contact with the ground. Flat-footed strategy in stance allows for greater floor foot friction, while surfing strategy reduces the swing phase's toe clearance and push-off velocity to decrease chances of another slip.
Increased variability in recovery kinematics following the slip initiation was also observed when compared to other falls. This is likely due to the recovery actions being performed on the low-friction sliding sheet, causing multi-directional slippage which increases kinematics variability. Slipping during recovery is likely to happen during an ecologically valid slip as slippery surfaces are likely to extend over a wide area.

Conclusion
A novel fall inducing movable platform (FIMP) system was developed and showed to effectively induce both slips and trips during over ground walking gait conditions. Added benefit(s) to the system is that it does not limit its subjects to a constant heading angle, walking velocity or gait patterns, a notable limitation to treadmill-walking. Additionally, the light weight moving overhand safety harness system, enabled by the subject follower algorithm, did not inhibit the motion of the subjects with respect to their individualised movement pattern.
It is feasible to employ FIMP for fall studies. Endswing trips which induce lowering strategy can be reliably created, and the characteristic leg lowering and rapid rise of the contralateral leg to regain balance were easily observed. Skipping strategies were induced via mid-swing trips, although elevating strategy may be created with a shorter fall inducing time duration. During skipping strategy, the perturbed leg tried to straighten the knee while its hip was arrested. Since the perturbed leg cannot reach the ground with sufficient time, the contralateral leg needs to swing forward to widen the BoS. Slips were induced with FIMP, using the help of sliding sheets to reduce the ground-foot friction. During slips, both legs stiffened, and quickly transit to knees straightening to force a flat-foot configuration. Slip trials have much higher kinematics variability as the low-friction ground surface causes multidirectional slip of the lower limbs. Therefore, FIMP has proven to induce ecologically valid overground walking gait and falls, similar to those reported previously in the literature [12,30,16,36,35,31].
The usage of SPM1D as an analysis tool provided researchers with the first ever quantifiable trip and slip reactive kinematics. SPM1D has also proven to be an invaluable statistical platform for visualising the changes in time-varying joint kinematics during various fall scenarios. Abnormalities with balance impaired subjects can now be targeted to the manage Figure 17 One-way repeated measure ANOVA comparison of NormalWalking (NW) vs StrapWalking (SW) vs Slip (SL) trials for the left leg with SPM1D. The left leg is attached to the ankle straps with a brake at the opposite end. Top row indicates the mean hip, knee and ankle flexion angles for all subjects and standard deviation clouds; mean (± st.dev). Second row indicates the ANOVA test on the data shown in top row. Third row shows a colour map highlighting significant differences of the ANOVA result in second row. The last row shows the significant different effect size of individuals. All the joint angles are relatively constant as compared to NormalWalking. This indicates that stiffening of joints occur during slips.

Figure 18
One-way repeated measure ANOVA comparison of NormalWalking (NW) vs StrapWalking (SW) vs Slip (SL) trials for the right leg with SPM1D. The right leg is not attached to any ankle straps. Top row indicates the mean hip, knee and ankle flexion angles for all subjects and standard deviation clouds; mean (± st.dev). Second row indicates the ANOVA test on the data shown in top row. Third row shows a colour map highlighting significant differences of the ANOVA result in second row. The last row shows the significant different effect size of individuals. The right knee and ankle has a more constant joint angle throughout the swing phase, indicating the adoption of the surfing strategy (keeping foot flat and close to ground).
the true cause of their balance deficiency, such as weak muscles or slow reaction time.
Abbreviations ANOVA: Analysis of variance; BoS: Base of Support; CoM: Center of mass; FIMP: Fall inducing movable platform; IMU: Inertial measurement unit; MSE: Mean square of error; PD: Proportional derivative; SPM: 1-dimensional statistical parametric mapping