Systems and Processes for Gait Phase Estimation of Wearable Robots

Abstract

A wearable robot and a method for controlling a wearable robot having at least one leg unit include: (a) obtaining at least one input signal from at least one encoder for tracking the hip joint angle relative to time, the encoder being connected to the wearable robot and corresponding to at least one leg unit; (b) windowing the at least one input signal within a window size according to the hip joint angle relative to time; (c) decomposing the at least one input signal using a discrete wavelet transform (DWT); (d) identifying at least one gait event in the gait cycle by using the DWT; (e) calculating temporal gait parameters according to the at least one gait event; (f) generating an assistive force in the at least one leg unit in response to the temporal gait parameters.

Description

Technical Field

[0001] Systems and processes for estimating gait phase for wearable robots are provided. The systems and processes employ an adaptive oscillator and a gait phase estimator and are based on discrete wavelet transform (DWT) applied to angular measurements for gait assistance of a portable hip exoskeleton. The systems and processes can continuously track gait phase and identify associated biomechanical gait events to segment the gait cycle in real time. Background Art

[0002] Gait disorders and lower limb injuries caused by aging and pathological conditions such as neurological and musculoskeletal diseases are reasons for limiting life and losing personal independence. Wearable robots (WRs), such as robotic exoskeletons (REs), can be used to assist movement, restore functional gait patterns, and demonstrate technological assistance to provide support in daily life scenarios. In addition, wearable robots can facilitate the rehabilitation programs of people with limb injuries and movement disorders.

[0003] Portable WRs are commercially available, and some have recently entered the market, indicating that the WR field is rapidly "moving out of the laboratory" to meet real-life applications.

[0004] There are still great challenges in designing WRs to achieve more natural cooperation between human users and robots. Ideally, the so-called physical human-robot interface (pHRI) will ensure a comfortable mechanical power exchange from the robot to the human user, preferably based on appropriate kinematic coupling, as well as stable and effective weight distribution. The cognitive human-robot interface (cHRI) should ensure intuitive decoding and adaptive and reliable assistance strategies in different motion tasks. In the case of providing assistive devices for people with mild lower limb injuries, monitoring the wearer's motion signals is crucial for implementing a voluntary control strategy.

[0005] Some sensing systems include reliable gait phase estimation and assistive controllers, such as foot switches, pressure-sensitive insoles, electromyography sensors, encoders, inertial measurement units (IMUs), etc. In particular, foot sensors have been proven to be able to reliably detect gait events related to foot-ground interaction; however, their use in daily scenarios is severely limited due to the inherent need for repeated calibration and limited sensor durability. In the WR field, although not clearly demonstrated through dedicated research, the choice to connect the exoskeleton to an external sensing device is expected to limit the usability of the entire system, especially in applications outside the laboratory. A self-contained sensing system in the exoskeleton, on the other hand, will improve the portability, reliability, and usability of the device.

[0006] Allowing the user to move freely and enhancing comfort during daily use of the WR requires minimizing any burden imposed by the sensory system. Thus, an important challenge for the WR is to accomplish reliable gait phase estimation and assistive control by using an integrated sensor that minimizes the invasiveness of pHRI to the user.

[0007] Continuous gait phase estimation enables the WR to provide phase-based assistive actions that are specifically tailored to the user's unique gait pattern and synchronized with specific biomechanical events or gait phases, while allowing the user to move freely when no assistive action is required. One method for real-time gait phase estimation is based on an adaptive oscillator (AO). The adaptive oscillator can extract high-level features such as phase, frequency, amplitude, and offset from a quasi-periodic input signal. In some WRs, the adaptive oscillator has been combined with a gait event detection (GED) process to smoothly and stably reset the phase when a specific preset biomechanical event occurs (e.g., heel strike or other events).

[0008] In controlling an active pelvic orthosis (APO), a past method was to extract the maximum flexion angle (MFA) from the hip flexion angle to reset the AO-based phase estimator. The hip angle was obtained from on-board sensors to avoid the need for other sensing systems. This method has been tested in controlling the APO to provide zero assistance and has shown good gait phase estimation performance during treadmill walking.

[0009] Although this method is simple, when applied to sensors (such as encoders) that measure joint angles configured on the drive unit shaft, this method has inaccuracies in angle measurement due to the human-robot interaction dynamics. Specifically, the soft tissue compression of the user's limb caused by the assistive action limits the reliability of peak detection in the MFA-based method. Additionally, although easy to detect, the MFA is not a standard biomechanical event for segmenting the gait cycle because the phase transition / occurrence in the stride may be related to the subject, speed, and task.

[0010] Identifying biomechanically relevant events such as heel strike (HS) and toe off (TO) from the input signal obtained from the integrated sensor will enable easier definition of the phase-locked assist profile. Additionally, such identification can be conveniently adjusted by clinicians based on normative data and estimate temporal gait parameters to provide quantitative metrics related to gait quality, while enhancing the portability, reliability, and usability of the WR device. Summary of the Invention

[0011] Systems and processes for estimating gait phase of a wearable robot (WR) or robotic exoskeleton (RE) according to the present disclosure provide a novel gait phase estimator that is capable of tracking gait phase in real time and identifying associated biomechanical gait events, such as resetting phase and segmenting the gait cycle. Additionally, gait event detection is based on discrete wavelet transform (DWT), which identifies at least one gait event (such as heel strike (HS) and toe off (TO) events) based on hip angle signals rather than acceleration / velocity, thereby avoiding the use of additional sensors other than those already integrated in the RE structure.

[0012] An exemplary robotic exoskeleton (RE) includes an active pelvic orthosis (APO) and a drive system to assist flexion / extension movement of bilateral hips transmitted by first and second leg units. The active pelvic orthosis (APO) may include a torso from which the drive system extends to at least one hip joint on which at least one encoder is disposed, the at least one hip joint corresponding to at least one leg unit. The at least one encoder may be arranged to measure the flexion / extension angle of the hip of at least one leg unit relative to the torso at the at least one hip joint. The active pelvic orthosis (APO) preferably includes a power unit and a computing unit. Preferably, the power unit is arranged to provide assistive power to the drive system to drive at least one leg unit at the hip joint according to the segmentation of the user's gait cycle determined by the computing unit from the input signals of the at least one encoder.

[0013] The active pelvic orthosis (APO) includes a gait phase estimator combined with the computing unit and adapted to process the input signals of the at least one encoder. The gait phase estimator includes a primary phase estimator based on an adaptive oscillator (AO), a gait event detector, and a phase error compensator. The adaptive oscillator (AO) of the primary phase estimator decomposes the periodic hip angle signal into different harmonics. Then, the phase, frequency, amplitude, and offset of each harmonic are estimated, and the primary phase estimator reconstructs the corrected gait phase.

[0014] The gait event detector can identify gait events of the gait cycle based on the DWT of the input signal. For example, the gait event detector may be arranged to identify HS and / or TO events of the gait cycle by processing the hip angle signal measured by the at least one encoder using DWT.

[0015] The main part of the gait event detection (GED) process is based on using DWT to decompose the signal into low-level and high-level frequency components, which are then processed using a threshold-based routine to identify specific events.

[0016] The gait event detection (GED) process can compute discrete wavelet coefficients on a vector of hip angle samples obtained within a time window. The process of identifying HS and TO from the input signal preferably has five main steps: (i) computing discrete wavelet coefficients; (ii) threshold-based event detection; (iii) knowledge-based verification; (iv) computing temporal gait parameters, and (v) adaptively updating the window length.

[0017] When HS DWT is detected, the phase error learning block of the phase error compensator can determine the phase reset error and dynamically compensate for the phase error. The estimated phase can be a continuous variable from 0 to 2π in each stride. The zero phase value should match the desired event identified by the event detection block. The phase error learning block computes the mismatch between zero and the estimated phase during the detected event. The error between the reconstructed signal and the original signal is fed back to the input to guide the evolution of the dynamic system. Thus, the adaptive oscillator (AO) in the presently disclosed embodiments can adaptively and rapidly learn in real time and synchronize with the input periodic signal.

[0018] According to the systems and processes of the present disclosure, a new gait phase estimator for a wearable robot (WR) or a robotic exoskeleton (RE), such as an APO, can be implemented based on employing DWT and AO. By synchronizing the zero phase with a desired biomechanical event (such as HS) and designing a desired assistive torque that depends on the gait phase, the systems and processes improve the high-level APO control system. More specifically, the correct and precise detection of specific HS and TO events enables the APO to: (i) adapt the assistive torque profile of the actuator system in the APO to the user's gait pattern, thus having the ability to distinguish between the swing and stance phases; (ii) adapt the assistive torque profile to different gait patterns of the user, whether due to physiological or non-physiological characteristics (e.g., patients with neurological disorders, etc.); (iii) adapt the assistive torque profile to varying walking speeds; and (iv) improve clinician usability. The process uses the AO to adaptively learn the walking frequency and identifies the start and end of each step through a novel gait event detector based on DWT.

[0019] One advantage of this process is that it can accurately identify gait events and estimate the gait cycle in real time without the need for additional sensors located outside the wearable robot (WR) (such as shoe sensors or similar sensors used in prior art systems and processes). The system and process according to this embodiment can be configured to extract distal foot events (i.e., HS and TO) in real time by only using the signals of the encoders located at the hip joints of the wearable robot (WR), enhancing the potential of cognitive cHRI while keeping the complexity of physical pHRI at the lowest level. In addition, the DWT-based process can enhance the distinctive signal features for identifying relevant biomechanical events and gait cycle phases.

[0020] By identifying HS and TO events in the gait cycle in real time, the energy cost - benefit and human - robot cooperation between the user and the WR can be maximized. Similarly, the real - time identification of gait events based on the DWT of the input signal allows the synchronization of the zero phase of the estimated gait with the HS event and allows the real - time provision of information about the TO event. These advantages are crucial for designing a phase - locked assistive profile that can be easily attributed to the physiological hip torque profile, making the adjustment procedure easy to use for both clinicians and users.

[0021] The identification of two related gait events (i.e., HS and TO) also configures the APO with additional capabilities to measure quantitative gait parameters during walking activities, including the stance and swing phases, single - support and double - support phases. For assistive lower - limb devices designed for rehabilitation in a clinical environment and gait - assist devices used outside the clinic, the additional ability to evaluate the user's gait quality (in terms of time parameters, symmetry index, rhythm, etc.) is an important feature because it can evaluate the user's gait pattern without the need for additional instrumentation and outside a dedicated gait analysis laboratory, enhancing the patient's self - assessment and paving the way for new monitoring technologies, whereby clinicians can remotely monitor patients walking outdoors.

[0022] These and other features of the present disclosure will be better understood by referring to the following description, the appended claims, and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1A A perspective view of an exemplary wearable robot device arranged as an active pelvic orthosis is shown.

[0024] Figure 1B A perspective view of an exemplary wearable robot device arranged as an active pelvic orthosis with an open housing is shown.

[0025] Figure 2 An exemplary schematic diagram and corresponding graph illustrating a gait cycle segmented at the heel - strike and toe - off phases are shown.

[0026] Figure 3 is another exemplary schematic diagram showing a repetitive gait cycle.

[0027] Figure 4 is a flowchart of a discrete wavelet transform process for gait phase estimation for heel strike and toe off.

[0028] Figure 5 is a flowchart showing a gait phase estimator.

[0029] Figure 6 is an exemplary schematic diagram describing an experimental setup for evaluating a DWT process for gait phase estimation for heel strike and toe off.

[0030] Figure 7 is a graphical example diagram showing a real-time discrete wavelet transform process for gait phase estimation.

[0031] Figure 8A and 8B respectively show perspective views of an exemplary wearable robotic device of an active pelvic orthosis when arranged to be fixed to a user.

[0032] The drawings are not necessarily to scale, but are drawn in a manner that provides a better understanding of the components and are not intended to limit the scope but to provide an exemplary illustration. DETAILED DESCRIPTION

[0033] Systems and processes for estimating the gait phase of a wearable robot (WR) or a robotic exoskeleton (RE) according to various embodiments provide a new gait phase estimator that is capable of tracking the gait phase in real time and identifying relevant biomechanical gait events to reset the phase and segment the gait cycle. Gait event detection is based on discrete wavelet transform (DWT), which identifies at least one gait event (e.g., heel strike (HS) and toe off (TO) events) based on the hip angle signal, thus avoiding the need to use additional sensors other than those already integrated into the exoskeleton structure.

[0034] Although the exemplary embodiments of the present disclosure show and describe gait phase estimation based on hip angle signals, the embodiments of the present disclosure can also be adapted to different applications of WRs. Thus, as will be understood by those of ordinary skill in the art from the present disclosure, the embodiments of the present disclosure are not limited to hip exoskeletons, but can be applied to any robotic joint where it is necessary to detect repetitive events containing specific frequency contributions, such as in robotic joints associated with assisting lower limb physiological joints.

[0035] As Figure 1A and 1BAs shown, an exemplary WR or RE includes an Active Pelvic Orthosis (APO) 100 having a drive system arranged to assist the flexion / extension movement of the bilateral hips transmitted by the first and second leg units 140. An exemplary APO can be found in U.S. Patent Application Publication 2017 / 0367919, published on December 28, 2017, which is incorporated herein by reference. The APO may include a torso 110, from which drive units 120 of the drive system extend to first and second hip joints 130, having first and second encoders 132 thereon and corresponding to the first and second leg units 140. The first and second encoders 132 may be arranged to measure the flexion / extension angles of the first and second leg units 140 relative to the hips of the torso 110 at the first and second hip joints 130. The APO preferably includes a power unit 112 and a computing unit 114, which have a battery and a control board of a processor or system-on-module, respectively. According to one embodiment, the power unit 112 and the computing unit 114 may be arranged in the torso 110 of the APO. The power unit 112 is preferably arranged to provide auxiliary power to the drive units 120 to drive the transmission units 142 of the first and second leg units 140 at the first and second hip joints 130 according to the segmentation of the user's gait cycle determined from the input signals provided by at least one of the first and second encoders 132 by the computing unit 114.

[0036] According to different embodiments, the torso 110 may be fixed to a belt to fit the user, and the first and second leg units 140 may each be fixed to limb bands or linkages to transfer mechanical power to the user's limbs.

[0037] The APO 100 includes a gait phase estimator combined with the computing unit 114 and adapted to process the input signals. As Figure 5 shown, the gait phase estimator 500 includes an adaptive oscillator-based phase estimator 550, a gait event detector 560, and a phase error compensator 570. The gait segmentation of the gait cycle is determined based only on the input signals 532 from the first and second encoders 132 at the first and second hip joints 13. The gait event detector may identify the gait events of the gait cycle based on the discrete wavelet transform (DWT) of the input signals. For example, the gait event detector may be arranged to identify the heel strike of the gait cycle by processing the hip joint angle signals measured by at least one of the first and second encoders using the DWT.

[0038] The proposed gait event detection process aims to identify HS and TO events in real time by using the hip joint angle as the input signal in the WR. The main part of the GED process is based on using DWT to decompose the signal into low-level and high-level frequency components, which are then processed through a value-threshold-based routine to identify specific events.

[0039] From this data, the WR can be used to assist the user through a drive system based on the user's gait phase estimation, according to the data obtained and processed by the gait phase estimator 500. To this end, the gait phase estimator can communicate the determined gait pattern of the user to the computing unit and / or the drive system, so that the torque profile of the drive system can be adapted to the user's gait pattern.

[0040] More specifically, the correct and precise detection of specific HS and TO events enables the APO to: (i) adapt the assistive torque profile of the drive system in the APO to the user's gait pattern, thus having the ability to distinguish between the swing and stance phases; (ii) adapt the assistive torque profile to different gait patterns of the user, whether due to physiological or non-physiological characteristics (e.g., patients with neurological disorders, etc.); (iii) adapt the assistive torque profile to the changing walking speed; and (iv) improve the usability for clinicians.

[0041] For example, based on the detection of HS and TO events, the APO can be used to generate a force by providing assistive flexion / extension torque at the hip joint at the correct time and amplitude through a thigh linkage at the user's thigh to assist the user's unique movement in dynamic tasks. Therefore, due to using only DWT of the hip joint angle signal input for gait phase estimation, the APO provides an assistive torque with amplitude and time that automatically depends on the user and responds to the user's needs in real time.

[0042] Brief description of the gait cycle

[0043] Figure 2 Illustrated is a single gait cycle 200 and the gait cycles of the left and right legs segmented by the extracted time parameters at HS and TO. The double support period (both feet in contact with the ground) in the stance phase of the gait cycle gives way to two double floating periods (neither foot in contact with the ground) at the start and end of the gait swing phase. Single support refers to the period when only one foot is in contact with the ground. When walking, this corresponds to the swing phase of the other limb.

[0044] The stance phase refers to the period when the foot is in contact with the ground. The swing phase is the period when the foot is not in contact with the ground. In cases where the foot never leaves the ground (dragging the foot), it can be defined as the phase when all parts of the foot are moving forward.

[0045] Figure 2 and Figure 3 shows the HS (or initial contact). HS includes a short period starting from when the foot touches the ground and is also the first phase of double support. HS includes 30° of hip flexion with the knee fully extended, the ankle moving from dorsiflexion to the neutral position and then into plantar flexion. After that, the knee starts to bend (5°) and increases as the plantar flexion of the heel increases. The plantar flexion is caused by the eccentric contraction of the anterior tibialis, and the extension of the knee joint is caused by the contraction of the quadriceps. In addition, the contraction of the hamstrings causes flexion, and the contraction of the rectus femoris causes hip flexion.

[0046] TO occurs when terminal contact of the toes occurs. The hip becomes less extended, the knee is usually bent 35 - 40°, and the toes leave the ground.

[0047] Discrete Wavelet Transform (DWT) and Gait Event Detection (GED)

[0048] The GED process identifies HS and TO events in real time by using the hip joint angle obtained from the encoder of the APO as the input signal. The main part of the GED process focuses on the DWT, which decomposes the signal from the encoder into low-level and high-level frequency components and is processed by a threshold-based process to identify specific events.

[0049] Theoretical Background

[0050] The wavelet transform decomposes an arbitrary signal into elementary signals by convolving the input signal with shifted and scaled variants of the mother wavelet. The DWT uses a binary grid of scale (s) and position (τ) to calculate the wavelet coefficients, i.e., s = 2 j and τ = k·2 j . The DWT of the signal x(t) is defined as:

[0051]

[0052] where, ψ j，k (t) is the mother wavelet, ψ * is the complex conjugate of ψ, and k, j ∈ Z, which are the scaling parameters of time and frequency of ψ respectively.

[0053] In 1989, Mallat developed an efficient computational method to implement DWT, which uses a digital filter bank related to the mother wavelet and a classical two-channel subband encoder scheme. Mallat's algorithm uses iterative high-pass (HP) and low-pass (LP) filters to split the signal spectrum, decomposing the original signal into high-frequency components called detail coefficients (cD) and low-frequency components called approximation coefficients (cA). According to Mallat's method (which is also called the decomposition pyramid algorithm), decomposing the signal requires four steps: (1) symmetric extension of the input signal to avoid boundary effects; (2) filtering through two complementary filters, namely, a high-pass filter (HP) and a low-pass filter (LP); (3) cropping the boundary samples; (4) downsampling by a factor of 2. These four steps will be iterated until the decomposition level is equal to the depth of the decomposition set used for analysis.

[0054] Process Architecture

[0055] As Figure 4 shown, the GED process relies on the discrete wavelet coefficients on the hip angle sample vector obtained within a time window. Therefore, while calculating the DWT coefficients, the DWT-based GED process is used to identify heel strike (HS DWT ) and toe off (TO DWT ) in real time, including the following five different steps:

[0056] Step 1: Calculation of DWT coefficients. Using two mother wavelets, namely fk22 and Haar, two DWTs are calculated for the input vector, which are used to identify HS and TO events respectively. The fourth-level and fifth-level detail coefficients for the Haar (cD4-Haar) and fk22 (cD3-fk22) mother wavelets are calculated respectively.

[0057] Step 2: Detection of threshold-based events. The event detection process operates on the DWT coefficients to detect the valley value of the cD3-fk22 coefficient profile (event 1) and the zero crossing value of cD4-Haar (event 2). Event 1 and event 2 have been selected because they occur synchronously with HS and TO respectively. To avoid false detection of event 1, a threshold equal to zero is applied to the cD3-fk22 coefficient before detecting the valley value.

[0058] Step 3: Knowledge-based verification. The verification step aims to avoid false event detection based on prior knowledge of gait biomechanical constraints. Specifically, gait event detection will be rejected unless a pre-defined percentage of the gait cycle has passed since the previous HS and TO events. Specifically, only when these conditions (2)-(4) and (3)-(5) are met respectively, will event 1 and event 2 be accepted as HS and TO.

[0059] HS(k) - HS(k - 1) > ρ1T * (2)

[0060] TO(k) - TO(k - 1) > ρ1T * (3)

[0061] HS(k) - TO(k - 1) > ρ2T * (4)

[0062] TO(k) - HS(k - 1) > ρ3T * (5)

[0063] where T * represents the average duration of the last five gait cycles, and ρ1, ρ2, and ρ3 are constants (∈[0, 1]), equal to 50%, 10%, and 20% respectively.

[0064] Step 4: Calculation of time - domain gait parameters. When a new HS event is detected, the following gait time parameters are calculated on the last gait: swing duration, stance duration, single - support duration, initial / end double - support duration, and cadence.

[0065] Step 5: Window length update. At each HS detection, the window length of the input signal is updated according to the real - time estimated gait cadence. A linear function describes the relationship between the optimal window size and the gait cadence.

[0066] Gait Phase Estimator

[0067] Figure 5 shows the architecture of the real - time gait phase estimator 500 based on three subsystems (AO - based phase estimator 550, phase reset module or gait event detector 560, and phase error learning block 570). The AO - based phase estimator 550 is set to estimate the phase, frequency, amplitude, and offset of the harmonics of the hip - angle signal. The HS detection event is detected according to the DWT - based method. When an HS event is detected, phase error learning is performed. The phase reset error P e (t k ) can be defined as:

[0068]

[0069] The state variable learns how to compensate for the phase reset error. The following equation models the dynamics of the phase error:

[0070]

[0071] where is The desired variant captures the error in a single gait; k p is a constant, and ω(t) is the input signal frequency tracked by AO.

[0072] Experimental Validation

[0073] The following embodiments provided illustrate specific embodiments of the systems and processes of the present disclosure and demonstrate the features and advantages of these embodiments, but are not intended to limit their scope. On the contrary, these embodiments guide those of ordinary skill in the art to understand and apply the inventive content of the present disclosure. The DWT-based GED process and gait phase estimation have been implemented and tested on a powered APO. The purpose of the experimental tests was to quantify the performance of the process to reliably identify HS and TO events and estimate the gait phase at different walking speeds. The foregoing process was benchmarked against an MFA-based GED algorithm and equipped with shoes fitted with pressure-sensitive insoles as a reference signal for detecting events.

[0074] A. Experimental Setup

[0075] Figure 6 Generally shows the three systems included in the experimental setup: WR is defined as APO600, two pressure-sensitive insoles 680, and a treadmill 690.

[0076] The Figure 1A -B is associated with Figure 8A -B, and the APO is a bilateral powered hip exoskeleton 800 designed to assist people with mild motor disabilities in hip flexion / extension movements. Since the control electronics and battery are installed in a backpack or torso 810, the APO is portable and compact, including a power supply and a computing unit. The mechanical structure includes: a main frame 812 connected to an orthopedic belt 814 that wraps around the user's torso; and a leg unit 840 connected to two linkages 844 that transmit mechanical power to the user's limbs. For each side, the active degrees of freedom (DOFs) provide hip flexion / extension torque. The passive degrees of freedom of hip abduction / adduction allow the user to move in the frontal plane. The drive unit 820 is based on a series elastic actuator structure (SEA) to ensure applicability and safe interaction in the so-called "transparent" (i.e., zero torque command to the drive unit) and "assist" modes. Two absolute encoders are embedded on each hip joint 830, a 17-bit encoder for measuring the deformation of the series spring and a 13-bit encoder for measuring the hip flexion / extension angle.

[0077] The control system has a hierarchical structure, including a low layer, a middle layer, and a high layer. The low layer is responsible for closed-loop torque control and is implemented based on a proportional-derivative regulator. The middle layer implements an adaptive assistance strategy: (i) estimating the gait phase in real time and (ii) calculating the desired assistance torque. The high layer identifies the user's motion intention.

[0078] The instrumented shoe communicates with the APO through a UWB transceiver module (data rate of 6.8 Mbps) integrated on the APO electronic board. The data acquisition rate is 100 Hz. The shoe is equipped with a piezoresistive insole 680 composed of 16 optoelectronic sensors. The sensing insole 680 records the plantar pressure distribution, the vertical ground reaction force (vGRF), and the center of pressure (CoP) in real time. The biomechanical variables are saved by the processing system 682 and processed offline to identify the reference HS and TO events, which are used as benchmarks for evaluating the phase estimation process based on DWT and MFA. A threshold-based process is applied to the vGRF to identify the HS event (HS insole ) and the TO event (TO insole ).

[0079] Figure 7 Shows the hip angle measured by the encoder, the DWT coefficients for identifying HS and TO, the ground reaction force and the center of pressure estimated by the instrumented shoe, and the gait phase estimated in real time using the new wavelet-based method. The valley value of the fk22 wavelet coefficient corresponds to the estimated HS event (solid circle), and the zero-crossing value of the Haar wavelet coefficient corresponds to the estimated TO event (filled slashed circle). The reference events from the insole are identified as the initial point HS (hollow solid circle) of the CoP and the final point TO (hollow dashed circle) of the CoP, respectively. The vertical dashed line marks the time error between the estimated event and the reference event.

[0080] B. Experimental protocol

[0081] Eight healthy subjects (7 males) participated in this study (32 ± 3 years old, 75 ± 12 kg, 175 ± 9 cm). The shoe sizes of the subjects were between 41.5 and 43 European sizes.

[0082] Throughout the experiment, the subjects were required to wear a pair of instrumented shoes 680 and APO 600. When wearing shoes 680 and APO 600, the subjects participated in two experimental activities, namely DWT mode activity and MFA mode activity. Within each activity, the subjects were required to conduct four walking tests at different speeds. Three tests were completed at a constant speed (i.e., 2 km / h; 3 km / h; 4 km / h), and the other one was completed with increasing speed (i.e., from 2 to 4 km / h, with the speed increasing by 0.5 km / h every 30 seconds). Between two consecutive phases, the subjects rested for at least 5 minutes. The order of the activities was randomized among different subjects to avoid order effect bias.

[0083] Each walking test included two 2-minute walking tests (2MWT). The first one was performed in the transparent mode (TM), and the second one was performed in the assisted mode (AM). The TM state was always implemented before the AM state.

[0084] Under the AM condition, the subjects received phase-locked hip flexion and extension torques according to real-time phase estimation (based on the DWT or MFA-based segmentation method). The adjustment of the amplitude and time of the assistive torque was done manually. The assistive torque profile consisted of two cosine curves, with a peak of +6.1 ± 1.2 Nm at flexion and a peak of -4.3 ± 0.5 Nm at extension, at 61.9 ± 2.1 and 15.6 ± 3.8% of the gait cycle (at 0% of the gait cycle of HS insole . Taking the hip torque profile reported in the Winter dataset when walking at normal speed as a reference, the assistance provided at flexion was approximately equivalent to 20% of the biological hip flexion moment, and the assistance provided at extension was approximately equivalent to 10% of the biological hip extension moment (Wimer D.A., Biomechanics and Motor Control of Human Movement. 2009).

[0085] C. Data Analysis

[0086] Comparing the MFA-based process with the DWT-based process, including quantifying the accuracy and precision of gait phase estimation and GED blocks. To evaluate the performance of each gait phase estimation, two key performance indicators (KPIs) were calculated.

[0087] The RMS of the phase reset error (RMS_P e ) was calculated as:

[0088]

[0089] where P e(i) is the phase reset error for the i-th gait, and N is the number of gaits in a specific walking test. Additionally, the linearity of the estimated phase is calculated as the standard deviation of the first derivative of the estimated phase. For each walking test, the median of the phase linearity error is calculated.

[0090] The performance of the GED block relative to the insole reference events is evaluated by calculating other parameters. The MFA, and the time distance between HS DwT and HS insole are calculated to study the phase shift and the invariance of the phase reset event compared to the reference heel strike event. To evaluate the TO identification performance of the DWT-based GED, the time distance between TO DWT and TO insole is also calculated. For each walking test, the median of the performance metrics is extracted.

[0091] The error detection of MFA, HS DWT and TO DWT is checked through the gait cycle duration calculated based on these events. The recognition accuracy of these gait events is calculated as the ratio of the number of correctly recognized events to the total number of gaits. Finally, using Bland - Altman and regression analysis for gait, the ability of the DWT-based GED process to extract temporal gait parameters is studied by calculating the mean absolute error (MEA) of the standing, swing, single, and double support durations estimated from the reference signal and the DWT-based process.

[0092] D. Statistical Analysis

[0093] For each of the above performance metrics, the median and interquartile range (IQR) of different subjects are extracted to quantify the accuracy and precision of gait phase estimation and GED of the two methods (i.e., MFA and DWT-based).

[0094] After verifying the normality of the data using the Lilliefors test, inferential statistical tests (one-tailed Wilcoxon signed-rank) are implemented to infer the statistical differences in accuracy and precision between the two methods (MFA and DWT-based). The statistical significance of all analyses is set at 5%.

[0095] Discussion

[0096] Real-time inference of reliable and accurate information about gait phase is a fundamental requirement for WR to achieve human-robot synchronization and provide smooth and safe assistive actions consistent with human gait biomechanics.

[0097] The gait phase estimator of the portable hip exoskeleton of the present disclosure is implemented based on DWT and AO. The gait phase estimator adaptively identifies the start and end of each step through a novel wavelet-based gait event detector and learns the walking frequency through AO. According to the present disclosure, the gait phase estimator process unexpectedly shows that distal events related to foot-ground interaction (i.e., HS and TO) can be reliably extracted by only using the signal of the hip joint angle of the integrated sensors in the exoskeleton, and the extracted information can reliably control the robotic hip exoskeleton.

[0098] Based on HS DWT The gait phase estimator process based on HS is reliable and robust enough to deliver actual torques and shows several advantages compared to previous solutions of the prior art, such as using pressure-sensitive insole. The proposed method shows remarkable performance in real-time estimating the gait phase, and most of the performance metrics are not affected by the assistance condition and gait speed. As recognized in the relevant literature, this result is crucial for maximizing human-robot cooperation and potential energy cost-effectiveness. In addition, in the proposed method, the zero phase of the estimated gait phase occurs synchronously with the HS event and provides information about the TO event in real time. These advantages are crucial for designing a phase-locked assistance profile that can be easily attributed to the physiological hip torque profile reported in most biomechanical literature, thus making the adjustment process easy to use for clinicians. Similarly, the possibility of detecting HS and TO events according to the present disclosure provides the possibility of using the exoskeleton to estimate temporal gait parameters (stance, swing, and double support duration) and symmetry metrics. For assistive devices, the additional ability to evaluate the user's gait quality is a relevant feature for evaluating the outcome of any home-based rehabilitation program and will enrich the range of available functions of the WR.

[0099] A. Performance invariance with respect to gait speed and assistance mode

[0100] The large variability of gait patterns of healthy and impaired users makes the development of a gait segmentation process independent of the user challenging. There are several factors that affect joint kinematics and dynamics, such as the presence of any gait pathologies, as well as factors such as gender, age, weight, height, walking speed, rhythm, surface, slope, etc. Gait variations are also more pronounced at the hip compared to the ankle, which makes the development of a hip-based segmentation process more complex. Probably due to these obstacles, few studies have considered the gait segmentation process of hip exoskeletons, and no known study has considered the gait segmentation process of hip exoskeletons having a continuous assistance role during the gait cycle (i.e., flexion and extension) and considering torque amplitudes sufficient for gait assistance scenarios.

[0101] The results obtained according to the presently disclosed embodiments show that based on HS DWTThe gait phase estimator is robust to different walking speeds, and the robot's assistive actions do not degrade performance. The results show that, under all test conditions, this process performs better than the MFA-based process because, under all conditions, the phase reset error of the DWT-based phase estimator is approximately 64% lower. Under different speed conditions, compared to the MFA-based method, HS DWT 's phase performance metric shows significantly less inter-subject variability. Although the sensing technologies employed by both methods may be similar, i.e., hip encoders integrated into the mechanical structure of the exoskeleton, the performance of the gait phase estimator based on HS DWT has a substantial improvement over prior art methods, including the MFA-based method.

[0102] B. Comparison with AO-based gait phase estimators using distal sensing devices

[0103] Although several recent studies have proposed new methods for continuously and real-time estimating gait phase using AO, these existing methods require the use of sensory signals from different locations (primarily distal locations), such as instrumented shoes, wearable depth cameras, and non-contact capacitive sensors. Additionally, according to the present disclosure, the wavelet-based phase estimator unexpectedly performs better than existing methods in the literature while advantageously only requiring the use of proximal hip encoders. In the comparative example, the average value of the RMS of the phase reset error of the present gait phase error estimator based on HS DWT is 2.1% of the gait cycle in TM and 1.4% of the gait cycle in AM, relative to the average value of the prior art, which is 2.5% of the gait cycle in TM achieved using an instrumented shoe, 4% of the gait cycle in TM and 3.6% of the gait cycle in AM achieved using a capacitive sensor worn on the calf, and 3.8% of the gait cycle achieved using a wearable depth camera.

[0104] Particularly advantageous is that the improved results of the currently disclosed gait phase estimator based on HS DWT extend to a wider range of experimental conditions and assistive scenarios. This is particularly important because the variability in phase estimation can affect the repeatability of the assistive profile, which has been shown to reduce the effectiveness of assistive actions. For example, assistive torque provided at the wrong time during the gait cycle can reduce the assistive effect and even conflict with the remaining motor ability of the target end user.

[0105] C. Comparison with gait phase estimators using hip exoskeleton sensing devices

[0106] While a self - contained sensory system within an exoskeleton structure has been considered a convenient design choice for developing usable and acceptable exoskeletons by end - users, differences in joint kinematics and dynamics between users (especially when measured from the hip) have hindered the successful development of such self - contained systems. In particular, inertial measurement units have been investigated for developing control strategies for hip exoskeletons. However, these investigations have shown that specific training of users is required before successful use and that they fail to achieve levels comparable to those of the phase reset errors currently disclosed. Thus, compared with known embodiments, the systems and processes of the present disclosure show the additional advantages of not requiring specific model training of subjects and seamless adaptation to sudden changes in gait speed.

[0107] D. Calculation of Temporal Gait Parameters

[0108] For an assistive lower - limb device designed for assistance outside of clinical rehabilitation and laboratory environments, the additional ability to assess the gait quality of users is an important feature. The possibility of assessing user gait patterns based on temporal gait parameters and symmetry - related metrics without the need for additional instrumentation and outside of a dedicated gait analysis laboratory paves the way for new remote - monitoring technologies, such that clinicians can remotely evaluate patients while they are walking outdoors.

[0109] In the present disclosure, the described embodiments are capable of extracting HS and TO events using only a hip exoskeleton, without the need for additional instrumentation and outside of a dedicated gait analysis laboratory. Although the similar possibility of using a hip exoskeleton to extract HS and TO events has not been considered in the art, a DWT - based phase estimator can reliably detect HS and TO events, with average errors of 0.4% and - 1.1% respectively. Unexpectedly, this favorable performance level is consistent with the methods reported for wearable - sensor - based arrangements in the art.

[0110] Conclusion

[0111] It should be understood that not all objectives or advantages need to be achieved under any embodiment or method of the present disclosure. Those skilled in the art will recognize that the systems and methods of the present disclosure can be embodied or implemented such that it achieves or optimizes one or a group of the advantages taught herein, without the need to achieve other objectives or advantages taught or suggested herein.

[0112] Those skilled in the art will recognize the substitutability of the various features and methods disclosed. In addition to the variations described, those skilled in the art can mix and match other known equivalents of each feature and method to construct an HRI in accordance with the principles of the present disclosure. Those skilled in the art can understand that the described features can be applied to other types of orthotics, prosthetics, or medical devices.

Claims

1. A method for controlling a wearable robot having at least one leg unit, the method comprising: (a) obtaining at least one input signal from at least one encoder for tracking hip joint angle relative to time, the at least one encoder being connected to the wearable robot and corresponding to the at least one leg unit; (b) windowing the at least one input signal within a window size according to the hip joint angle relative to time; (c) decomposing the at least one input signal using discrete wavelet transform (DWT); (d) identifying at least one gait event in a gait cycle by using the DWT; (e) calculating temporal gait parameters according to the at least one gait event; (f) generating an assistive force in the at least one leg unit in response to the temporal gait parameters.

2. The method according to claim 1, characterized in that, The method further comprises the following steps: (g) updating the window size as a function of an estimated gait rhythm.

3. The method according to claim 2, wherein The method further comprises the step of repeating steps (a)-(g) when the at least one gait event is detected and for each subsequent gait cycle.

4. The method according to claim 2, wherein The at least one encoder is only arranged to correspond to the hip joint.

5. The method according to claim 1, wherein The method further comprises: using a gait phase estimator configured to track at least one gait phase in real time and identify at least one biomechanical gait event to reset the gait phase and segment the gait cycle.

6. The method according to claim 5, characterized in that, The gait phase estimator comprises an adaptive oscillator phase estimator, a phase reset module, and a device for phase error learning.

7. The method according to claim 6, wherein The adaptive oscillator phase estimator is configured to estimate at least one of phase, frequency, amplitude, and harmonic offset of the at least one input signal corresponding to the hip joint angle.

8. The method according to claim 6, characterized in that, The device for phase error learning comprises calculating a phase reset error when a gait event is detected.

9. The method according to claim 1, characterized in that, The step of using the DWT to identify a gait detection event in a gait cycle comprises the following sub-steps: calculating discrete wavelet coefficients based on at least one input signal; determining the gait event to identify at least one of heel strike and toe off; detecting at least one of heel strike and toe off for the at least one leg unit.

10. The method according to claim 9, wherein The step of calculating discrete wavelet coefficients based on at least one input signal comprises obtaining an input signal at the hip joint angle.

11. The method according to claim 9, wherein, The step of using the DWT to identify at least one of heel strike and toe off in a gait cycle further comprises the following sub-steps: (v) calculating temporal gait parameters according to the detection of at least one of heel strike and toe off.

12. The method according to claim 9, characterized in that, The calculation of temporal gait parameters comprises the following durations: (i) gait time; (ii) swing phase; (iii) stance phase; and (iv) at least one parameter selected from the group consisting of single support, initial double support, final double support, double support, and rhythm.

13. The method according to claim 1, wherein At least one adaptive oscillator is configured to decompose a periodic input signal into different harmonics including at least one of phase, frequency, amplitude, and offset.

14. The method according to claim 13, characterized in that, The method further comprises: determining an error between a reconstructed periodic input signal and the periodic input signal and inputting the error to correct the estimation of the gait phase.

15. The method according to claim 1, wherein The wearable robot includes an active pelvic orthosis having a drive system configured to assist bilateral hip flexion / extension movements transmitted by at least one of a first and a second leg unit. The active pelvic orthosis includes a torso from which hip joints extend, and first and second encoders of the at least one encoder are arranged and respectively correspond to the first and second leg units.

16. The method according to claim 15, characterized in that, The first and second encoders are arranged to measure hip flexion / extension angles of the first and second leg units relative to the torso.

17. The method according to claim 15, wherein The method further includes driving the first and second leg units according to an estimated gait cycle and the at least one input signal transmitted by the first and second encoders. The active pelvic orthosis includes a power supply unit and a computing unit. The power supply unit provides auxiliary power to the first and second leg units at the hip joints in response to a segmentation of the gait cycle determined by the computing unit.

18. The method according to claim 17, characterized in that, The determination of the segmentation of the gait cycle is achieved based only on at least one input signal from the first and second encoders at the hip joints.

19. A wearable robot, comprising: An active pelvic orthosis having a drive system configured to assist hip flexion / extension movements transmitted by at least one leg unit. The active pelvic orthosis includes a torso from which hip joints protrude, and at least one encoder is arranged. The at least one encoder corresponds to at least one leg unit to respectively transmit at least one input signal. The at least one encoder is configured to measure hip flexion / extension angles of the at least one leg unit relative to the torso; A power supply unit and a computing unit. The power supply unit provides auxiliary power to the at least one leg unit at the hip joints in response to a gait segmentation of the gait cycle determined by the computing unit from at least one encoder; A gait phase estimator configured to track at least one gait phase in real time and identify at least one biomechanical gait event to reset the gait phase and segment the gait cycle; A gait event detector configured to identify a flexion angle from an input signal based on a discrete wavelet transform (DWT), the discrete wavelet transform being used to identify at least one relevant phase-invariant gait event of the gait cycle.

20. The wearable robot according to claim 19, wherein, The determination of the gait segmentation of the gait cycle is achieved based only on the at least one input signal obtained from the at least one encoder.

Images