Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-02-0156
Joint-Based Model Using Centre of Pressure and Muscle Movement for Motion Attempt Detection of Lower-Limb Exoskeleton
School of Mechanical Engineering, The University of Adelaide, Adelaide, 5005, Australia
ABSTRACT
In the last few decades, the exoskeletons have been developed to assist patients in their rehabilitation process or provide extra power for industry workers. Regardless of the applications, user motion attempt recognition is of its paramount importance to provide timely accurate assistance as well as more flexible and comfortable experience. Past studies had limited success in estimating the wearer motion intentions, but the setup is often complex and dependent on specific scenarios to predict effectively. Our study focuses on three main types of sensors and develops an algorithm to detect the user intention for a single knee joint exoskeleton. Three force-sensitive resistors are used to calculate the user's changes in centre of pressure (COP) and a novel buffer sensor joint is designed to detect any immediate motion. The data collected was processed under a Finite State Machine (FSM) model to classify the exoskeleton into states. Additionally, the user's thigh muscle activation is also monitor using Mechanomyography (MMG) signal. MMG is a new technique that utilises the mechanical manifestation of muscle movement to capture the vibration from the muscle fibre when placed at core muscle groups responsible for knee motion. The signal is collected by placing six accelerometers at the key positions. We then use a Support Vector Machine (SVM) model to train the MMG signal retrieve from different movement scenarios (including sit-to-stand, walking, stand-to-sit and stair-ascending, descending) in order to build a predictive model that works out the knee movement attempt. The key usage of the trained SVM model in conjunction to the buffer sensor joint is to determine whether the wearer require support in certain motions, such as during walking or transition from sitting to standing. The SVM model prediction is used as switching conditions in the FSM model into different states, such as following and supporting state. For each SVM prediction, a score is given to determine the level of confidence of the model on the prediction itself. The FSM model then take the prediction score into account along with user's changes in centre of pressure to decide whether joint support is needed. The proposed system and models can follow the user’s movement while monitoring its motion states and provides assistance. The study aims to lay the groundwork for smoother exoskeleton motor control with lighter weight and comfortable design.
Keywords: Exoskeleton, Mechanomyography, Centre of Pressure, Intention detection, Support vector machine, Finite state machine.
School of Mechanical Engineering, The University of Adelaide, Adelaide, 5005, Australia
ABSTRACT
In the last few decades, the exoskeletons have been developed to assist patients in their rehabilitation process or provide extra power for industry workers. Regardless of the applications, user motion attempt recognition is of its paramount importance to provide timely accurate assistance as well as more flexible and comfortable experience. Past studies had limited success in estimating the wearer motion intentions, but the setup is often complex and dependent on specific scenarios to predict effectively. Our study focuses on three main types of sensors and develops an algorithm to detect the user intention for a single knee joint exoskeleton. Three force-sensitive resistors are used to calculate the user's changes in centre of pressure (COP) and a novel buffer sensor joint is designed to detect any immediate motion. The data collected was processed under a Finite State Machine (FSM) model to classify the exoskeleton into states. Additionally, the user's thigh muscle activation is also monitor using Mechanomyography (MMG) signal. MMG is a new technique that utilises the mechanical manifestation of muscle movement to capture the vibration from the muscle fibre when placed at core muscle groups responsible for knee motion. The signal is collected by placing six accelerometers at the key positions. We then use a Support Vector Machine (SVM) model to train the MMG signal retrieve from different movement scenarios (including sit-to-stand, walking, stand-to-sit and stair-ascending, descending) in order to build a predictive model that works out the knee movement attempt. The key usage of the trained SVM model in conjunction to the buffer sensor joint is to determine whether the wearer require support in certain motions, such as during walking or transition from sitting to standing. The SVM model prediction is used as switching conditions in the FSM model into different states, such as following and supporting state. For each SVM prediction, a score is given to determine the level of confidence of the model on the prediction itself. The FSM model then take the prediction score into account along with user's changes in centre of pressure to decide whether joint support is needed. The proposed system and models can follow the user’s movement while monitoring its motion states and provides assistance. The study aims to lay the groundwork for smoother exoskeleton motor control with lighter weight and comfortable design.
Keywords: Exoskeleton, Mechanomyography, Centre of Pressure, Intention detection, Support vector machine, Finite state machine.
