Machine Learning Approaches for Activity Recognition and/or Activity Prediction in Locomotion Assistive Devices—A Systematic Review

Abstract

Locomotion assistive devices equipped with a microprocessor can potentially automatically adapt their behavior when the user is transitioning from one locomotion mode to another. Many developments in the field have come from machine learning driven controllers on locomotion assistive devices that recognize/predict the current locomotion mode or the upcoming one. This review synthesizes the machine learning algorithms designed to recognize or to predict a locomotion mode in order to automatically adapt the behavior of a locomotion assistive device. A systematic review was conducted on the Web of Science and MEDLINE databases (as well as in the retrieved papers) to identify articles published between 1 January 2000 to 31 July 2020. This systematic review is reported in accordance with the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines and is registered on Prospero (CRD42020149352). Study characteristics, sensors and algorithms used, accuracy and robustness were also summarized. In total, 1343 records were identified and 58 studies were included in this review. The experimental condition which was most often investigated was level ground walking along with stair and ramp ascent/descent activities. The machine learning algorithms implemented in the included studies reached global mean accuracies of around 90%. However, the robustness of those algorithms seems to be more broadly evaluated, notably, in everyday life. We also propose some guidelines for homogenizing future reports.

Understanding LSTM Network Behaviour of IMU-Based Locomotion Mode Recognition for Applications in Prostheses and Wearables

“…The following activities were selected, Walking (W), Stair Ascent (SA), Stair Descent (SD), Ramp Ascent (RA), Ramp Descent (RD) and Stopped (S). Labarrière et al identified these as the most commonly investigated and they require no equipment or skill to perform [ 10 ]. The study received ethical approval from the University of Bath Research Ethics Approval Committee for Health (REACH), reference EP 19/20 003 .…”

“…The tuning of these controllers is time-consuming and requires a highly skilled prosthetist. In the current state of the art for LMR techniques, the focus has been on the use of ML techniques to automate the process of feature selection, output classification, and personalization [ 10 ].…”

“…Many different machine-learning techniques have been investigated including, Support Vector Machines, Hidden Markov Models and Convolution Neural Networks (CNN) with success [ 10 ]. As sensor data from human gait is temporal, the best architecture for solving this will be one that can take into account the sequential nature of the input data.…”

“…In HAR tasks, LSTMs have been demonstrated to provide exceptional performance [ 11 ] although very little work has been done investigating this in the context of prostheses. Labarrière et al conducted a systematic review of the ML methods used in activity recognition; for assistive device LSTM networks were only used once [ 10 ].…”

“…Machine Learning (ML) offers the ability to significantly increase the sophistication of such systems, through understanding of a wide range of activities and personalization to individual characteristics, without specialist intervention [ 10 ]. As classifying activities is temporal is nature, sequential ML networks, such as Long Short-Term Memory (LSTM), are a good fit.…”