Publicação
Towards a human-in-the-loop control for a smart orthotic system
| Resumo: | Stroke is the main cause of paralysis. This pathology has provoked a considerable increase of persons with motor impairments. With a therapy focused on each clinical case, the total or partial recovery can be achieved. Powered orthoses have been developed to promote an effective recover, based on repetitive gait training and user’s active participation. Many control approaches have been developed to control these devices, but none of them promotes an user-oriented strategy focused to the user’s needs. In an attempt of solving this issue, a new approach named Human-in-the-loop is emerging. This strategy allows the adaptation of some assistive parameters based on the user’s energetic cost, promoting a therapy tailored to each end-user needs. However, to estimate the energy expenditure, the use of non-ergonomic sensors, not suitable for clinical context, is required. Thus, it is necessary to find new ways of estimating energy expenditure using wearable and comfortable sensors. In this dissertation, the first steps to introduce the Human-in-the-loop strategy into a powered orthosis are presented. For this purpose, two strategies were developed: a strategy that allows the angular trajectory adaptation in real-time and other that promotes a stiffness adaptation all over the gait cycle. Both strategies were validated with healthy subjects. In the first strategy, the orthosis was able to modify its assistance in a fraction of microseconds, and the end-users were able to follow her with a median error below 10%. Regarding the second strategy, the results show that the orthosis allowed an effective change in the systems’ interaction stiffness, promoting an active participation of each user during its assistance. The energetic impact of using the robotic assistive device is also presented. As it promotes an energy expenditure augmentation in more than 30% in comparison to walk without the device, the necessity of implementing the Human-in-the-loop strategy was highlighted. In an attempt of finding an ergonomic technique to estimate the energetic cost, the use of machine learning algorithms was tested. The results, obtained with a MLP and a LSTM, prove that it is possible to estimate the energy expenditure with a mean error close to 11%. Future work consists in the implementation of the model in real-time and the collection of more data with the aforementioned control approaches, in a way of constructing a more robust model. |
|---|---|
| Autores principais: | Lopes, João Pedro Mendes |
| Assunto: | Adaptive trajectory control Assisted-as-needed impedance control Human-in-the-loop control Powered orthoses Stroke rehabilitation Controlo de trajetória adaptativa Controlo de impedância assisted-as-needed Controlo human-in-the-loop Ortóteses ativas Reabilitação de AVC |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Stroke is the main cause of paralysis. This pathology has provoked a considerable increase of persons with motor impairments. With a therapy focused on each clinical case, the total or partial recovery can be achieved. Powered orthoses have been developed to promote an effective recover, based on repetitive gait training and user’s active participation. Many control approaches have been developed to control these devices, but none of them promotes an user-oriented strategy focused to the user’s needs. In an attempt of solving this issue, a new approach named Human-in-the-loop is emerging. This strategy allows the adaptation of some assistive parameters based on the user’s energetic cost, promoting a therapy tailored to each end-user needs. However, to estimate the energy expenditure, the use of non-ergonomic sensors, not suitable for clinical context, is required. Thus, it is necessary to find new ways of estimating energy expenditure using wearable and comfortable sensors. In this dissertation, the first steps to introduce the Human-in-the-loop strategy into a powered orthosis are presented. For this purpose, two strategies were developed: a strategy that allows the angular trajectory adaptation in real-time and other that promotes a stiffness adaptation all over the gait cycle. Both strategies were validated with healthy subjects. In the first strategy, the orthosis was able to modify its assistance in a fraction of microseconds, and the end-users were able to follow her with a median error below 10%. Regarding the second strategy, the results show that the orthosis allowed an effective change in the systems’ interaction stiffness, promoting an active participation of each user during its assistance. The energetic impact of using the robotic assistive device is also presented. As it promotes an energy expenditure augmentation in more than 30% in comparison to walk without the device, the necessity of implementing the Human-in-the-loop strategy was highlighted. In an attempt of finding an ergonomic technique to estimate the energetic cost, the use of machine learning algorithms was tested. The results, obtained with a MLP and a LSTM, prove that it is possible to estimate the energy expenditure with a mean error close to 11%. Future work consists in the implementation of the model in real-time and the collection of more data with the aforementioned control approaches, in a way of constructing a more robust model. |
|---|