�rea Cient�fica: Robótica
Feedback-Error-Learning: a low-level control strategy for powered robotic systems
Publicada a 2018-01-23
Aluno: Pedro Nuno Lopes Fernandes       N�mero: a68385       Email: pnlfernandes@gmail.com
Data in�cio: 01/10/2017   

Orientador(es):
Nome: Cristina Manuela Peixoto dos Santos
Email: cristina@dei.uminho.pt   

Descri��o:

Modern societies are privileged with high quality of life and high average life expectancy, mainly due to the improvement of living conditions, development of medicine and improvement of medical care. Consequently, research for new, improved and faster rehabilitation therapies is a requirement for these populations and for researchers themselves. A target of these rehabilitation therapies is mobility, since it is one of the common and relevant human physical activities. Without it, daily life activities such as getting up from bed in the morning, walking to the bathroom, going for a walk or even working can be drastically affected. Furthermore, lack of mobility can lead to a deprived social life and a decrease in communication and mental health. Our body health is also affected, since our muscles can deteriorate without mobility. As mentioned in [12], human motor impairment can result in social and work (10% in working age) exclusion, early retirement, costly medical treatments and social assistance. Therefore, rehabilitation therapies regarding the improvement of people’s mobility can be a goal to robotics rehabilitation, once it can reinforce and/or restore the human performance concerning mobility.
Therefore, lower-limb robotic devices are an excellent research opportunity in the field of assistive technologies to motor rehabilitation. Consequently, therapies assisted by these wearable devices for treating lower limb impairments have been increasingly used in addition to conventional therapeutic approaches, in the fields of rehabilitation and assistance [1][13]. Particularly, powered exoskeletons and orthoses have been highlighted, as they increase human motor function, and rehabilitate through task-oriented training.
The main goal of these robotic devices is to have active technologies, ultimately controlled by the user, to both increase human motor function, and rehabilitate through task oriented and repetitive gait training. One key feature of these technologies, is the ability to act according to patients’ intention by applying assist as need (AAN) strategies.
In the backbone of these technologies, low-level assistive strategies are responsible for controlling the actuators that allow the execution of the AAN strategies and, consequently, of the rehabilitation therapies. Hence, these devices should be embedded with efficient, robust, and adaptive low-level controllers, capable of performing an effective and reliable actuation so the reestablish of the abnormal gait pattern can be successful.


Objectivos:

The main goal proposed in this work addresses the development of a low-level Feedback-Error Learning (FEL) control scheme, aiming to be validated in the active lower limb BioMot Exoskeleton [10] and Powered Knee Orthosis [11]. The system should overcome real-time constraints to provide assistance in walking at different conditions. Thus, to accomplish this main goal, the project shall achieve the following goals:
• Goal 1: The first goal covers the development of a low-level feedback controller, robust to disturbances and variant frequencies. The efficiency and stability of the controller for different operating conditions is key to the success of the developed strategy, so a variety of walking motions can be performed by the robotic device. The validation of the controller will be performed with healthy volunteers, in gait treadmill walking, and performance metrics should be obtained to test the reliability and effectiveness of real-time execution.
• Goal 2: As a second goal, a low-level feedforward controller with an artificial neural network (ANN), capable of online and real-time learning, will be performed. This feedforward controller should be able to learn the inverse dynamic of the robotic system to control, so it can anticipate the control command to perform. As the learning process of the ANN should be done in online and real-time conditions, so the considered design for the ANN should be carefully planned to accommodate these constraints. The validation of the controller will be performed analyzing its output and comparing it to the one achieved by the feedback controller. The consumed time to learn its dynamic is also going to be a constraint to be validated.
• Goal 3: The third goal addresses the validation of the projected FEL control strategy on the active lower-limb BioMot Exoskeleton [10] and Powered Knee Orthosis [11]. These robotic systems will be used by healthy subjects and a protocol to validate the whole system will be performed. Different walking conditions (speeds and slopes) and users with different anatomic characteristics should be taken into account, so the system is validated with de most different scenarios as possible. The considered validation of the system, should cover a comparison study with different low-level motion strategies, so the limitations and advantages of the current strategy could be identified.


Palavras chave:
gait; rehabilitation; low-level controller, FEL, ANN, PID, lower-limb, exoskeletons; orthosis

Copyright © DEI Universidade do Minho - Todos os direitos reservados
Powered byNetmove