Brain-machine interface systems (BMIs) allow people to control external devices, such as powered wheelchairs, telepresence robots, robotic arms, exoskeletons, mouse and keyboards, through brain signals. A BMIs is defined as a closed-loop composed of acquisition, processing, feature extraction, classification and feedback. To interact with the devices there are two different control strategies to convert the BMI decoder's output into an appropriate signal for the robotic device. The first one is called discrete control which allows the user to send discrete commands, therefore there are some seconds between two consecutive commands. In the case of BMIs based on motor imagery, for instance, a command is sent to the robot when a predetermined threshold is reached, or, in other words, when the control framework is certain of the user's intention. The other is continuous control, which maps each brain signal to a control signal for the external device. Hence, continuous control increases precision and performance compared to discrete control, however, is more challenging to implement. A framework based on dynamical systems is a possible solution to implement such a continuous control. However, state-of-art control frameworks based on dynamical systems rely on several user-dependent hyper-parameters that make an optimal fine-tuning difficult for each user. In addition, the continuous control based on a dynamic system already proposed in the literature is designed to work symmetrically namely it implements the same behaviours for all the BMI classes. This constraint may be too strict, especially for a beginner user who has not yet mastered how to balance different classes in BMIs. Therefore, the aim of the thesis is twofold: first, to reduce the number of parameters required for continuous control based on a dynamical system, and second, to allow the dynamical system to work with an asymmetrical behaviour, so each class can behave differently. To this end, we propose a new metric in order to find the most effective combination of parameters for each user. In the first phase, we examine, a dataset available at IAS-Lab to evaluate the metric and find the proper correlation among parameters via a posterior analysis. Then, we recruited 12 subjects to perform a 2-class motor imagery (MI) task by virtually controlling a steering wheel. The preliminary results confirm that there is a relation between the parameters used for the symmetrical and the asymmetrical dynamic system. Consequently, we validate a possible implementation of the asymmetrical dynamic system based on the relation found in the previous step with an experiment divided into three sessions. In these sessions, we compare the dynamic control system to the exponential one. The results of the experiment confirm that for almost all the users the asymmetric dynamical system provides better performance and less workload than the asymmetric exponential system.

Brain-machine interface systems (BMIs) allow people to control external devices, such as powered wheelchairs, telepresence robots, robotic arms, exoskeletons, mouse and keyboards, through brain signals. A BMIs is defined as a closed-loop composed of acquisition, processing, feature extraction, classification and feedback. To interact with the devices there are two different control strategies to convert the BMI decoder's output into an appropriate signal for the robotic device. The first one is called discrete control which allows the user to send discrete commands, therefore there are some seconds between two consecutive commands. In the case of BMIs based on motor imagery, for instance, a command is sent to the robot when a predetermined threshold is reached, or, in other words, when the control framework is certain of the user's intention. The other is continuous control, which maps each brain signal to a control signal for the external device. Hence, continuous control increases precision and performance compared to discrete control, however, is more challenging to implement. A framework based on dynamical systems is a possible solution to implement such a continuous control. However, state-of-art control frameworks based on dynamical systems rely on several user-dependent hyper-parameters that make an optimal fine-tuning difficult for each user. In addition, the continuous control based on a dynamic system already proposed in the literature is designed to work symmetrically namely it implements the same behaviours for all the BMI classes. This constraint may be too strict, especially for a beginner user who has not yet mastered how to balance different classes in BMIs. Therefore, the aim of the thesis is twofold: first, to reduce the number of parameters required for continuous control based on a dynamical system, and second, to allow the dynamical system to work with an asymmetrical behaviour, so each class can behave differently. To this end, we propose a new metric in order to find the most effective combination of parameters for each user. In the first phase, we examine, a dataset available at IAS-Lab to evaluate the metric and find the proper correlation among parameters via a posterior analysis. Then, we recruited 12 subjects to perform a 2-class motor imagery (MI) task by virtually controlling a steering wheel. The preliminary results confirm that there is a relation between the parameters used for the symmetrical and the asymmetrical dynamic system. Consequently, we validate a possible implementation of the asymmetrical dynamic system based on the relation found in the previous step with an experiment divided into three sessions. In these sessions, we compare the dynamic control system to the exponential one. The results of the experiment confirm that for almost all the users the asymmetric dynamical system provides better performance and less workload than the asymmetric exponential system.

Discrete and continuous control for brain-actuated robotic devices

FORIN, PAOLO
2022/2023

Abstract

Brain-machine interface systems (BMIs) allow people to control external devices, such as powered wheelchairs, telepresence robots, robotic arms, exoskeletons, mouse and keyboards, through brain signals. A BMIs is defined as a closed-loop composed of acquisition, processing, feature extraction, classification and feedback. To interact with the devices there are two different control strategies to convert the BMI decoder's output into an appropriate signal for the robotic device. The first one is called discrete control which allows the user to send discrete commands, therefore there are some seconds between two consecutive commands. In the case of BMIs based on motor imagery, for instance, a command is sent to the robot when a predetermined threshold is reached, or, in other words, when the control framework is certain of the user's intention. The other is continuous control, which maps each brain signal to a control signal for the external device. Hence, continuous control increases precision and performance compared to discrete control, however, is more challenging to implement. A framework based on dynamical systems is a possible solution to implement such a continuous control. However, state-of-art control frameworks based on dynamical systems rely on several user-dependent hyper-parameters that make an optimal fine-tuning difficult for each user. In addition, the continuous control based on a dynamic system already proposed in the literature is designed to work symmetrically namely it implements the same behaviours for all the BMI classes. This constraint may be too strict, especially for a beginner user who has not yet mastered how to balance different classes in BMIs. Therefore, the aim of the thesis is twofold: first, to reduce the number of parameters required for continuous control based on a dynamical system, and second, to allow the dynamical system to work with an asymmetrical behaviour, so each class can behave differently. To this end, we propose a new metric in order to find the most effective combination of parameters for each user. In the first phase, we examine, a dataset available at IAS-Lab to evaluate the metric and find the proper correlation among parameters via a posterior analysis. Then, we recruited 12 subjects to perform a 2-class motor imagery (MI) task by virtually controlling a steering wheel. The preliminary results confirm that there is a relation between the parameters used for the symmetrical and the asymmetrical dynamic system. Consequently, we validate a possible implementation of the asymmetrical dynamic system based on the relation found in the previous step with an experiment divided into three sessions. In these sessions, we compare the dynamic control system to the exponential one. The results of the experiment confirm that for almost all the users the asymmetric dynamical system provides better performance and less workload than the asymmetric exponential system.
2022
Discrete and continuous control for brain-actuated robotic devices
Brain-machine interface systems (BMIs) allow people to control external devices, such as powered wheelchairs, telepresence robots, robotic arms, exoskeletons, mouse and keyboards, through brain signals. A BMIs is defined as a closed-loop composed of acquisition, processing, feature extraction, classification and feedback. To interact with the devices there are two different control strategies to convert the BMI decoder's output into an appropriate signal for the robotic device. The first one is called discrete control which allows the user to send discrete commands, therefore there are some seconds between two consecutive commands. In the case of BMIs based on motor imagery, for instance, a command is sent to the robot when a predetermined threshold is reached, or, in other words, when the control framework is certain of the user's intention. The other is continuous control, which maps each brain signal to a control signal for the external device. Hence, continuous control increases precision and performance compared to discrete control, however, is more challenging to implement. A framework based on dynamical systems is a possible solution to implement such a continuous control. However, state-of-art control frameworks based on dynamical systems rely on several user-dependent hyper-parameters that make an optimal fine-tuning difficult for each user. In addition, the continuous control based on a dynamic system already proposed in the literature is designed to work symmetrically namely it implements the same behaviours for all the BMI classes. This constraint may be too strict, especially for a beginner user who has not yet mastered how to balance different classes in BMIs. Therefore, the aim of the thesis is twofold: first, to reduce the number of parameters required for continuous control based on a dynamical system, and second, to allow the dynamical system to work with an asymmetrical behaviour, so each class can behave differently. To this end, we propose a new metric in order to find the most effective combination of parameters for each user. In the first phase, we examine, a dataset available at IAS-Lab to evaluate the metric and find the proper correlation among parameters via a posterior analysis. Then, we recruited 12 subjects to perform a 2-class motor imagery (MI) task by virtually controlling a steering wheel. The preliminary results confirm that there is a relation between the parameters used for the symmetrical and the asymmetrical dynamic system. Consequently, we validate a possible implementation of the asymmetrical dynamic system based on the relation found in the previous step with an experiment divided into three sessions. In these sessions, we compare the dynamic control system to the exponential one. The results of the experiment confirm that for almost all the users the asymmetric dynamical system provides better performance and less workload than the asymmetric exponential system.
BMI
Continuous control
Discrete control
Dynamical system
File in questo prodotto:
File Dimensione Formato  
Forin_Paolo.pdf

accesso aperto

Dimensione 4.16 MB
Formato Adobe PDF
4.16 MB Adobe PDF Visualizza/Apri

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/45846