Development of Wearable Gait Assistive Device Using Recurrent Neural Network

• Published in: 2019 IEEE/SICE International Symposium on System Integration (SII) pp. 626 - 631
• Main Authors: Tang, Shi Yuan, Hoang, Ngoc Son, Chui, Chee Kong, Lim, Jeong Hoon, Chua, Matthew C. H.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.01.2019
• Subjects: Accelerometers; Assistive devices; Data models; Foot; Gait Assistive Device; Gait Cycle Recognition; Legged locomotion; Prototypes; Recurrent Neural Network; Solid modeling
• ISSN: 2474-2325
• DOI: 10.1109/SII.2019.8700415

In elderly population, gait disorders are common where majority of these disorders are associated as symptoms of neurodegenerative diseases including Parkinson's Disease (PD), Huntingtons Disease (HD), and Amyotrophic Lateral Sclerosis (ALS). In addition to affected mobility, the patients are also susceptible to greater risk of falls, hence increasing the demand for caretakers. With the trend of aging population, personal assistive device could be deployed to assist patients to regain independence and improve their quality of life. This paper proposes an end-to-end solution architecture for real-time standalone wearable gait assistive device to automate the rehabilitation activity. A key aspect of this study is to incorporate recurrent neural network (RNN) model that provides accurate pattern recognition and output actuation cue to the patients. Prototype and simulation data was used to show the feasibility of the proposed architecture and machine learning model. Preliminary results indicate favorable accuracy gait cycle detection for implementation. However, further optimizations are required to lower the computational costs and shorten the time lag between cycles to ensure low cost feasibility of the device.