Wearable Robotic Devices Research Articles

A Completely Portable and Concealable, Lightweight Assistive Exosuit for Upper Limbs.

Exosuits are a relatively new trend in wearable robotics to answer the flaws of their exoskeleton counterparts, but they remain impractical as the lack of rigidity in their frames makes the integration of crucial components into a single unit a challenge. While some simple solutions exist, almost all current research focuses on the output performance of exosuits rather than the needs of potential beneficiaries of this technology. To address this, a novel mechanism of complete portability for exosuits was developed and tested to improve exosuit practicality and adoption. Designed for elbow flexion, the device produced 12.21-13.66Nm of assistive torque and could be mostly concealed by the wearer's clothing without impacting performance. The proof-of-concept design proved successful and demonstrated many advantages over current portability methods, particularly in size and convenience, weighing only 1.7kg. This device provides the sense of normalcy crucial for a technology to seamlessly integrate into the daily lives of its end users. It is extendable and upgradeable with access to advanced materials and manufacturing methods.Clinical Relevance- Exoskeletons are currently the only marketed wearable robotic device for full limb support. This research is the foundation for a new series of exosuits that could drastically enhance the adoptability, accessibility, and versatility of exosuits in physical rehabilitation and general physical enhancement, becoming a superior alternative or addition.

The Application of Soft Robotic Gloves in Stroke Patients: A Systematic Review and Meta-Analysis of Randomized Controlled Trials.

Wearable robotic devices have been strongly put into use in both the clinical and research fields of stroke rehabilitation over the past decades. This study aimed to explore the effectiveness of soft robotic gloves (SRGs) towards improving the motor recovery and functional abilities in patients with post-stroke hemiparesis. Five major bibliographic databases, PubMed, Embase, Cochrane Library, Web of Science, and the Physiotherapy Evidence Database, were all reviewed for enrollment regarding comparative trials prior to 7 March 2023. We included adults with stroke and compared their rehabilitation using SRGs to conventional rehabilitation (CR) on hand function in terms of the Fugl-Meyer Upper Extremity Motor Assessment (FMA-UE), Fugl-Meyer Distal Upper Extremity Motor Assessment (FMA-distal UE), box and blocks test score, grip strength test, and the Jebsen-Taylor hand function test (JTT). A total of 8 studies, comprising 309 participants, were included in the analysis. Compared to CR, rehabilitation involving SRGs achieved better FMA-UE (MD 6.52, 95% CI: 3.65~9.39), FMA-distal UE (MD 3.27, 95% CI: 1.50~5.04), and JJT (MD 13.34, CI: 5.16~21.53) results. Subgroup analysis showed that stroke latency of more than 6 months and training for more than 30 min offered a better effect as well. In conclusion, for patients with stroke, rehabilitation using SRGs is recommended to promote the functional abilities of the upper extremities.

Reducing the muscle activity of walking using a portable hip exoskeleton based on human-in-the-loop optimization.

Introduction: Human-in-the-loop optimization has made great progress to improve the performance of wearable robotic devices and become an effective customized assistance strategy. However, a lengthy period (several hours) of continuous walking for iterative optimization for each individual makes it less practical, especially for disabled people, who may not endure this process. Methods: In this paper, we provide a muscle-activity-based human-in-the-loop optimization strategy that can reduce the time spent on collecting biosignals during each iteration from around 120 s to 25 s. Both Bayesian and Covariance Matrix Adaptive Evolution Strategy (CMA-ES) optimization algorithms were adopted on a portable hip exoskeleton to generate optimal assist torque patterns, optimizing rectus femoris muscle activity. Four volunteers were recruited for exoskeleton-assisted walking trials. Results and Discussion: As a result, using human-in-the-loop optimization led to muscle activity reduction of 33.56% and 41.81% at most when compared to walking without and with the hip exoskeleton, respectively. Furthermore, the results of human-in-the-loop optimization indicate that three out of four participants achieved superior outcomes compared to the predefined assistance patterns. Interestingly, during the optimization stage, the order of the two typical optimizers, i.e., Bayesian and CMA-ES, did not affect the optimization results. The results of the experiment have confirmed that the assistance pattern generated by muscle-activity-based human-in-the-loop strategy is superior to predefined assistance patterns, and this strategy can be achieved more rapidly than the one based on metabolic cost.

In-Situ Measurement of Multi-Axis Torques Applied by Wearable Soft Robots for Shoulder Assistance

While a number of wearable soft robotic devices have been proposed to assist the shoulder, limited efforts have been made to quantify the amount of torque they apply to the body. Most work to-date has assessed soft actuator performance with simple benchtop experiments that may not be representative of the boundary conditions on the human body. We propose a new methodology to measure torques directly in-situ on the body and then use this technique to make a detailed comparison of two versions of a soft wearable shoulder assistance robot. The impact of a number of factors are considered, such as actuator design, garment anchoring, material hysteresis, arm pose, and inflation pressure. Many of these factors are not present on the benchtop and are found to significantly affect torque production. We compare results obtained on a simple benchtop test fixture with two on-body settings: an idealized mannequin and actual human subjects. The mannequin and human results were similar, but differed significantly from the benchtop, further motivating the need for on-body testing. Moving forward, we believe that the ability to directly quantify device performance in-situ will be critical to develop new design, modeling, and control strategies for wearable robots.

Combining transcutaneous interferential-current for nerve inhibition with a robotic assistant device for increasing ankle dorsiflexion in walking

BackgroundA kilohertz-frequency alternating current transcutaneously applied was introduced as a novel neuromodulation technology for nerve inhibition innervating antagonist muscles. Combining this electrical nerve inhibition with a robotic assistance device has been proposed but not investigated. Research questionThis study aimed to demonstrate the effect of combining electrical nerve inhibition with a wearable robotic device on increasing ankle dorsiflexion during walking. We hypothesized that the wearable robotic device would elicit a greater ankle dorsiflexion angle with the same force in walking by applying the transcutaneous interferential-current nerve inhibition (TINI) technique to the tibial nerve. MethodsEleven healthy young adults performed three experimental conditions. The ankle assistance (AA) condition was walking while wearing an ankle device with operating dorsiflexion assistance during pre-swing and swing phases. For the ankle assistance with electrical stimulation (AE) condition, TINI on the tibial nerve was additionally applied from the AA condition. In the ankle non-assistance (AN) condition, participants wore the device, but assistance was not provided. The joint angles during walking were measured and digitized through a motion analysis system. ResultsDuring a gait cycle, immediate changes in ankle joint motions were observed in the sagittal plane. In the pre-swing phase, ankle dorsiflexion angle was significantly greater in AE condition than AA and AN. There was no significant difference in joint angle between AA and AN. SignificanceThis study demonstrates the effectiveness of combining TINI with a wearable robotic ankle device in increasing dorsiflexion angle during the pre-swing phase. This finding provides the feasibility of using TINI as a neuromodulation technique for assisting functional movement in human walking.

Designing innovative digital platforms from both human and nonhuman perspectives

Digital platforms are becoming critical infrastructures for supporting a variety of innovative services that enhance our everyday lives. These platforms need to offer not only rational services but also ludic or slow services that focus on human pleasure. One important aspect of creating innovative digital platforms is that their concrete requirements and potential opportunities are vague before they are designed. Thus, designing, prototyping and evaluating digital platforms iteratively is essential for refining or customizing them, as knowledge is gradually gained throughout these iterations. However, it is costly to develop prototype platforms and evaluate them with traditional methods. A better tool that can be used to reveal these platforms’ potential opportunities by conceiving them in a simple and rapid way is needed. In this paper, we present our journey to develop nine digital platforms that share collective human sight and hearing with the Human-Material-Pleasure (HMP) annotation method, which is a tool that we use to describe the visually structured annotations of multiple digital platforms based on the annotated portfolio method. The most significant part of the paper presents annotated portfolios based on the HMP annotation method for the nine digital platforms that we develop and shows how these annotated portfolios play an essential role in revealing and exploring the potential opportunities of our platforms during the refinement process. We also discuss how the HMP annotation method is used in the context of exploring the potential opportunities of wearable shape-changing robotic devices; these devices have significantly different characteristics from our digital platforms, which allows for showing insights more objectively by extracting diverse insights from an alternative angle.

Developing of a wearable ankle rehabilitation robotic device

People with walking disorders caused by accidents or stroke can undergo treatment to restore their mobility. Traditional therapy is time-consuming and time-consuming. Therefore, a new trend was born – to facilitate rehabilitation and reduce the patient’s time. Rehabilitation robotics is an area that is constantly evolving, and new mechanisms have recently been developed to help people regain their mobility. This paper presents a 3-RPS parallel manipulator for the restoration of the ankle joint with three degrees of freedom. Parallel manipulator 3-RPS with three degrees of freedom, which was introduced by K. Hunt in 1983 as one of the sedentary parallel manipulators. Since then, 3-RPS has attracted a lot of attention from robotics and biomedical engineering engineers.

Synergistic effect of carbon nanotubes and carbon black as nanofillers of silicone rubber pressure sensors

Pressure sensors based on nitrogen-doped bamboo-shaped carbon nanotubes (N-BCNT) and carbon black (CB) as nanofillers, polyurethane foam (PU) as supporting substrate, and silicone rubber (SR) as a matrix were prepared. Dip coating was used to coat PU with 0.44 wt% nanofiller, including different mixing ratios of N-BCNT and CB (5:5; 6:4; 7:3; 8:2; 9:1). Then, the coated PU is impregnated in SR to fill the pores. Due to the higher aspect ratio of the N-BCNT, it contributes more to improving the electrical conductivity in the composites, while the CB fills the smaller gaps. The prepared sensors were tested in various applications, and it was found that the optimal mixing ratio of nanofillers was 7:3 N-BCNT:CB. Thus, a multifunctional pressure sensor has been developed successfully with excellent flexibility and good resilience, suitable for motion detection and finger touch applications. The pressure sensor showed high sensitivity, and the ability to detect a wide range of pressures. The sensor exhibited success in a range of applications, paving the way for its potential use in various fields in the future, such as wearable devices, prosthetics, robotic devices, and medical devices.

Design and validation of a novel online platform to support the usability evaluation of wearable robotic devices.

Wearable robotic devices (WRD) are still struggling to fulfill their vast potential. Inadequate daily life usability is one of the main hindrances to increased technology acceptance. Improving usability evaluation practices during the development of WRD could help address these limitations. In this work, we present the design and validation of a novel online platform aiming to fill this gap, the Interactive Usability Toolbox (IUT). This platform consists of a public website that offers an interactive, context-specific search within a database of 154 user research methods and educational information about usability. In a dedicated study, the effect of this platform to support usability evaluation was investigated. Twelve WRD experts were asked to complete the task of defining usability evaluation protocols for two specific use cases. The platform was provided to support one of the use cases. The quality and composition of the proposed protocols were assessed by (i) two blinded reviewers, (ii) the participants themselves, and (iii) the study coordinators. We showed that using the IUT significantly affected the proposed evaluation focus, shifting protocols from mainly effectiveness-oriented to more user-focused studies. The protocol quality, as rated by the external reviewers, remained equivalent to those designed with conventional strategies. A mixed-method usability evaluation of the platform yielded an overall positive image, with detailed suggestions for further improvements. The IUT is expected to positively affect the evaluation and development of WRD through its educational value, the context-specific recommendations supporting ongoing benchmarking endeavors, and highlighting the value of qualitative user research.

Shared Autonomy Locomotion Synthesis With a Virtual Powered Prosthetic Ankle.

Virtual environments provide a safe and accessible way to test innovative technologies for controlling wearable robotic devices. However, to simulate devices that support walking, such as powered prosthetic legs, it is not enough to model the hardware without its user. Predictive locomotion synthesizers can generate the movements of a virtual user, with whom the simulated device can be trained or evaluated. We implemented a Deep Reinforcement Learning based motion controller in the MuJoCo physics engine, where autonomy over the humanoid model was shared between the simulated user and the control policy of an active prosthesis. Despite not optimising the controller to match experimental dynamics, realistic torque profiles and ground reaction force curves were produced by the agent. A data-driven and continuous representation of user intent was used to simulate a Human Machine Interface that controlled a transtibial prosthesis in a non-steady state walking setting. The continuous intent representation was shown to mitigate the need for compensatory gait patterns from their virtual users and halve the rate of tripping. Co-adaptation was identified as a potential challenge for training human-in-the-loop prosthesis control policies. The proposed framework outlines a way to explore the complex design space of robot-assisted gait, promoting the transfer of the next generation of intent driven controllers from the lab to real-life scenarios.

Bowden-line Foot-controlled Finger Exoskeleton Device Driven by Pneumatic Muscles

With the world being increasingly reliant on computers, hand disorders have become prevalent amongst the general population. To help patients with hand disorders better recover while still allowing them to live their normal lives, glove-like wearable robotic de-vices were developed. Most of those robotic devices, however, generate motion through electric motors, which is heavy and inconvenient. Soft body robots are popular nowadays because they are versatile, safe, low in cost, and comfortable to wear. Thus, a pneumatic muscle hand device is presented in this paper to provide aid to those with impaired hands in their daily lives. This system features 10 pneumatic muscles connected with strings attached to the inner and outer sides of every finger powered by an air valve which allows the retraction of the patient’s fingers. The patient is introduced to 5 pressure sensors able to be controlled by foot. To enable the patient to use the device on their own, the system is designed in a back-pack-like style.

Active cooling of twisted coiled actuators via fabric air channels.

Twisted coiled actuators (TCAs) are promising artificial muscles for wearable soft robotic devices due to their biomimetic properties, inherent compliance, and slim profile. These artificial muscles are created by super-coiling nylon thread and are thermally actuated. Unfortunately, their slow natural cooling rate limits their feasibility when used in wearable devices for upper limb rehabilitation. Thus, a novel cooling apparatus for TCAs was specifically designed for implementation in soft robotic devices. The cooling apparatus consists of a flexible fabric channel made from nylon pack cloth. The fabric channel is lightweight and could be sewn onto other garments for assembly into a soft robotic device. The TCA is placed in the channel, and a miniature air pump is used to blow air through it to enable active cooling. The impact of channel size on TCA performance was assessed by testing nine fabric channel sizes-combinations of three widths (6, 8, and 10 mm) and three heights (4, 6, and 8 mm). Overall, the performance of the TCA improved as the channel dimensions increased, with the combination of a 10 mm width and an 8 mm height resulting in the best balance between cooling time, heating time, and stroke. This channel was utilized in a follow-up experiment to determine the impact of the cooling apparatus on TCA performance. In comparison to passive cooling without a channel, the channel and miniature air pump reduced the TCA cooling time by 42% ( s to s, ). Unfortunately, there was also a 9% increase in the heating time ( s to s, ) and a 28% decrease in the stroke ( mm to mm, ). This work demonstrates that fabric cooling channels are a viable option for cooling thermally actuated artificial muscles within a soft wearable device. Future work can continue to improve the channel design by experimenting with other configurations and materials.

VP.26 Potential of the MyoSuit, a lightweight wearable lower-limb cable-actuated exoskeleton in patients with neuromuscular disorders: preliminary findings

Neuromuscular disorders (NMD) are characterized by progressive muscle weakness leading to dramatic impairments in functional capacities and quality of life. A new class of promising lightweight lower-limb wearable powered exoskeletons is currently emerging. However, data regarding the potential of such devices in patients with NMD are scarce. Within this work, we investigated the potential of the MyoSuit (MyoSwiss AG, Zurich, Switzerland), a lightweight wearable robotic device providing assistance for hip and knee extension against gravity using actuated cables. Four patients with NMD (2 with muscular dystrophies, 2 with inclusion body myositis) were studied. Without using the MyoSuit, 2-min walking distance (2MWD) was 128±30 m (66±17% of predicted values), number of repetitions during a 30-s sit-to-stand test (30s-STS) was 8±3, and MRC scores for knee extension and hip flexion were 3±1.1 and 3.9±0.6, respectively. Changes (n patients improved/impaired) in performance while using the MyoSuit during the following tests were as follows: 2MWD (1/3), 30-STS (1/3), TUG (0/4), 4-stairs climbing test (1/3). Regarding perceptions during the 2MWD as assessed using visual analogue scales, changes were: stability (0/4), exertion (2/2), and dyspnea (3/1). One fall occurred in one patient while wearing the device without consequences. Usability of the Myosuit as assessed using the system usability scale (/100) was 77±20, corresponding to good to excellent usability. Discomfort as assessed using the modified Nordic questionnaire (/10) was 1.0±0.5, corresponding to a low level of discomfort. The limited sample size limits the use of insightful statistics so far. Within this small group of patients, the use of the MyoSuit was associated with variable changes in performance and perceptions depending on the functional test. Further research in a larger sample is warranted to further investigate the potential of this device in patients with NMD.

Design and single-parameter adaptive fuzzy control of pneumatic lower limb exoskeleton with full state constraints

AbstractWith the excellent characteristic of intrinsic compliance, pneumatic artificial muscle can improve the interaction comfort of wearable robotic devices. This paper resolves the safety tracking control problem of a pneumatically actuated lower limb exoskeleton system. A single-parameter adaptive fuzzy control strategy is proposed with high control precision and full state constraints for the safe gait training tasks. Based on the barrier Lyapunov function, all signals in the closed-loop system can be bounded in finite time, which guarantees the deviation of the exoskeleton’s moving trajectory within a bounded range. Furthermore, with the proposed single-parameter adaptive law, the computational burden and the complexity of the control are reduced significantly. Finally, numerical simulations, no-load tracking experiments, and passive and active gait training experiments with healthy subjects validate the effectiveness of the proposed method.

Gait-Symmetry-Based Human-in-the-Loop Optimization for Unilateral Transtibial Amputees With Robotic Prostheses

Gait asymmetry due to unilateral limb loss increases the risk of injury or progressive joint degeneration. The devel- opment of wearable robotic devices paves the way for improving the gait symmetry of unilateral amputees. However, the state-of-the-art studies on human-in-the-loop optimization strategies with an optimization task of reducing the metabolic cost face several challenges, e.g., an excessively long optimization period and the infeasibility of optimization for unilateral amputees who have a deficit of gait symmetry. Herein, we propose a gait-symmetry-based human-in-the-loop optimization method to reduce the risk of injury or progressive joint degeneration for unilateral transtibial amputees. Experimental results ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${N} =3$ </tex-math></inline-formula> ) indicated that convergence took the range of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$388{s}$ </tex-math></inline-formula> to 821 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${s}$ </tex-math></inline-formula> . After optimization, compared with using passive prostheses, the gait-symmetry indicator of subjects wearing the robotic prostheses was improved by the range of 6.0% to 52.0%, and the net metabolic energy consumption was reduced by the range of 3.0% to 13.4%. Additionally, the rationality of gait indicators based on kinematics rather than kinetics was assessed. The results indicated that the human-in-the-loop strategy can improve the gait symmetry by reducing the metabolic cost and thus reduce the risk of joint injury for unilateral amputees.

A Study on Artificial Kinesthesia Generation by Simultaneous Stimulation of Mechanical Vibration and Mechanical Skin Stretch.

Artificially controlled kinesthesia can be applied to many situations because kinesthesia is essential to recognizing body movements. It could be used to generate artificial kinesthesia in rehabilitation or daily motion assist to improve self-efficacy of the robot user. Moreover, the controlled artificial kinesthesia could make people feel as if they are performing the actions of the robotic limbs with their own limbs. Mechanical vibration stimulation is one of the candidates to artificially control kinesthesia. It is known that mechanical vibration stimulation on human muscles or tendons from skin surface evokes an illusion of movement as if the stimulated muscles are extended. That effect of artificial kinesthesia is called Kinesthetic Illusion (KI). In this paper, a method to increase the amount of KI without changing the frequency of the vibration stimulation is investigated by applying mechanical skin stretch stimulation at the same time with the mechanical vibration stimulation. The experiment was conducted by generating KI for flexion motion of the elbow joint on a horizontal plane to evaluate the proposed approach. In the experiments, three out of five subjects showed obvious increase in the amount of KI when skin stretch stimulation was applied at the same time with the mechanical vibration stimulation. The results of this study provide a first step toward artificial kinesthesia control using a wearable robotic device using the mechanical vibration stimulation.

Improving Ankle Muscle Recruitment via Plantar Pressure Biofeedback during Robot Resisted Gait Training in Cerebral Palsy.

Neurological impairment from stroke or cerebral palsy often presents with diminished ankle plantar flexor function during the propulsive phase of gait. This deficit often results in slow, energy-expensive walking patterns that limit community mobility. Robotic gait training interventions may prove effective in improving functional outcomes, including exoskeleton resistance used to provide targeted neuromuscular recruitment. However, these interventions to date have required regular verbal cues and coaching for proper plantar flexor engagement with resistance, particularly for pediatric applications. In this validation study, we sought to address the need for automating and improving the effectiveness of facilitating user engagement with robotic resistance. Specifically, our main goal was to compare changes in plantar flexor activity between walking with plantar flexor resistance alone vs plantar flexor resistance combined with plantar pressure biofeedback in individuals with cerebral palsy. We recruited 8 ambulatory adolescents with cerebral palsy between the ages of 11-18 years old to participate in this cross-sectional feasibility study. Supporting our hypothesis, we observed a 36 ± 36% and 46 ± 39% increase in mean and peak soleus activity, respectively, between resistance plus biofeedback vs resistance alone (both p < 0.05). Compared to other biofeedback sensing modalities like assessment of muscle activity via surface electrodes, integrating the plantar pressure-based system within the wearable robotic devices minimizes barriers to clinical implementation by reducing cost, complexity, and setup time. With these positive feasibility results, our future work will explore longer-term training effects of ankle resistance combined with plantar pressure biofeedback.

Comparison of Machine Learning Techniques for Activities of Daily Living Classification with Electromyographic Data.

Advances in data science and wearable robotic devices present an opportunity to improve rehabilitation outcomes. Some of these devices incorporate electromyography (EMG) electrodes that sense physiological patient activity, making it possible to develop rehabilitation systems able to assess the patient's progress when performing activities of daily living (ADLs). However, additional research is needed to improve the ability to interpret EMG signals. To address this issue, an off-line classification approach for the 26 upper-limb ADLs included in the KIN-MUS UJI dataset is presented in this paper. The ADLs were performed by 22 subjects, while seven EMG signals were recorded from their forearms. From variable-length EMG time windows, 18 features were computed, and 13 features more were extracted from frequency domain windows. The classification performance of five different machine learning techniques, including Support Vector Machines (SVM), Convolutional Neural Networks (CNN), Gated Recurrent Unit (GRU) network, XGBoost, and Random Forests, were compared. CNN performed best amongst individual models, with an accuracy above 80%, compared to SVM with 77%, GRU with 73.9%, and the tree-based models below 64%. Ensemble learning with four CNN models achieved an even higher accuracy of 86%. These results suggest that the CNN ensemble model is capable of classifying EMG signals for most ADLs, which could be used in off-line quantitative assessment of robotic rehabilitation outcomes.

Cost Function Determination for Human Lifting Motion via the Bilevel Optimization Technology.

Investigating the optimal control strategy involved in human lifting motion can provide meritorious insights on designing and controlling wearable robotic devices to release human low-back pain and fatigue. However, determining the latent cost function regarding this motion remains challenging due to the complexities of the human central nervous system. Recently, it has been discovered that the underlying cost function of a biological motion can be identified from an inverse optimization control (IOC) issue, which can be handled via the bilevel optimization technology. Inspired by this discovery, this work is dedicated to studying the underlying cost function of human lifting tasks through the bilevel optimization technology. To this end, a nested bilevel optimization approach is developed by integrating particle swarm optimization (PSO) with the direction collocation (DC) method. The upper level optimizer leverages particle swarm optimization to optimize weighting parameters among different predefined performance criteria in the cost function while minimizing the kinematic error between the experimental data and the result predicted by the lower level optimizer. The lower level optimizer implements the direction collocation method to predict human kinematic and dynamic information based on the human musculoskeletal model inserted into OpenSim. Following after a benchmark study, the developed method is evaluated by experimental tests on different subjects. The experimental results reveal that the proposed method is effective at finding the cost function of human lifting tasks. Thus, the proposed method could be regarded as a paramount alternative in the predictive simulation of human lifting motion.

Real-Time Hierarchical Classification of Time Series Data for Locomotion Mode Detection.

Accurate real-time estimation of motion intent is critical for rendering useful assistance using wearable robotic prosthetic and exoskeleton devices during user-initiated motions. We aim to evaluate hierarchical classification as a strategy for real-time locomotion mode recognition for the control of wearable robotic prostheses and exoskeletons during user-initiated motions. We collect motion data from 8 subjects using a set of 7 inertial sensors for 16 lower limb locomotion modes of different specificities. A CNN based hierarchical classifier is trained to classify the modes into a specified label hierarchy. We measure the accuracy, stability, behaviour during mode transitions and suitability for real-time inference of the classifier. The method achieves stable classification of locomotion modes using [Formula: see text] of time history data. It achieves average classification accuracy of 94.34% and an average AU(PRC) of 0.773 - comparable to similar classifiers. The method produces more informative classifications at transitions between modes. Less specific classes are classified earlier than more specific classes in the hierarchy. The inference step of the classifier can be executed in less than 2 ms on embedded hardware, indicating suitability for real-time operation. Hierarchical classification can achieve accurate detection of locomotion modes and can break up mode transitions into multiple transitions between modes of different specificity. Multi-specific hierarchical classification of locomotion modes could lead to smoother, more fine grained control adaptation of wearable robots during locomotion mode transitions.