A Hierarchical-Based Learning Approach for Multi-Action Intent Recognition DOI Creative Commons
David Hollinger, Ryan S. Pollard, Mark C. Schall

и другие.

Sensors, Год журнала: 2024, Номер 24(23), С. 7857 - 7857

Опубликована: Дек. 9, 2024

Recent applications of wearable inertial measurement units (IMUs) for predicting human movement have often entailed estimating action-level (e.g., walking, running, jumping) and joint-level ankle plantarflexion angle) motion. Although or information is frequently the focus intent prediction, contextual necessary a more thorough approach to recognition. Therefore, combination may offer comprehensive intent. In this study, we devised novel hierarchical-based method combining classification subsequent regression predict joint angles 100 ms into future. K-nearest neighbors (KNN), bidirectional long short-term memory (BiLSTM), temporal convolutional network (TCN) models were employed classification, random forest model trained on action-specific IMU data was used prediction. A action-generic multiple actions backward kneeling down, up, walking) also angle. Compared with approach, had lower prediction error up. TCN BiLSTM classifiers achieved accuracies 89.87% 89.30%, respectively, they did not surpass performance when in an model. This been because from actions. study demonstrates advantage leveraging large, disparate sources over Moreover, it efficacy IMU-driven, task-agnostic future across

Язык: Английский

Evaluating the Performance of Joint Angle Estimation Algorithms on an Exoskeleton Mock-Up via a Modular Testing Approach DOI Creative Commons
Ryan S. Pollard,

Sarah Bass,

Mark C. Schall

и другие.

Sensors, Год журнала: 2024, Номер 24(17), С. 5673 - 5673

Опубликована: Авг. 31, 2024

A common challenge for exoskeleton control is discerning operator intent to provide seamless actuation of the device with operator. One way accomplish this joint angle estimation algorithms and multiple sensors on human–machine system. However, question remains what can be accomplished just one sensor. The objective study was deploy a modular testing approach test performance two models—a kinematic extrapolation algorithm Random Forest machine learning algorithm—when each informed solely gait data from single potentiometer an ankle mock-up. This demonstrates (i) feasibility implementing mock-up evaluation promote continuity between configurations (ii) that yielded lower realized errors estimated angles decreased time than model when deployed physical device.

Язык: Английский

Процитировано

0
A Hierarchical-Based Learning Approach for Multi-Action Intent Recognition DOI Creative Commons
David Hollinger, Ryan S. Pollard, Mark C. Schall

и другие.

Sensors, Год журнала: 2024, Номер 24(23), С. 7857 - 7857

Опубликована: Дек. 9, 2024

Recent applications of wearable inertial measurement units (IMUs) for predicting human movement have often entailed estimating action-level (e.g., walking, running, jumping) and joint-level ankle plantarflexion angle) motion. Although or information is frequently the focus intent prediction, contextual necessary a more thorough approach to recognition. Therefore, combination may offer comprehensive intent. In this study, we devised novel hierarchical-based method combining classification subsequent regression predict joint angles 100 ms into future. K-nearest neighbors (KNN), bidirectional long short-term memory (BiLSTM), temporal convolutional network (TCN) models were employed classification, random forest model trained on action-specific IMU data was used prediction. A action-generic multiple actions backward kneeling down, up, walking) also angle. Compared with approach, had lower prediction error up. TCN BiLSTM classifiers achieved accuracies 89.87% 89.30%, respectively, they did not surpass performance when in an model. This been because from actions. study demonstrates advantage leveraging large, disparate sources over Moreover, it efficacy IMU-driven, task-agnostic future across

Язык: Английский

Процитировано

0