10 Human Participants Gait Data collected outdoors from two smartwatches placed on each wrist

Assistive robotic systems are crucial in healthcare; however, their safe adoption is hindered by traditional authentication methods that present significant usability challenges. Assistive robots must authenticate users continuously without burdening older adults or people with disabilities. To address this gap, we present an intelligent, continuous user authentication system that leverages human gait data captured by smartwatch sensors to authenticate users. Our proposed system securely transmits and analyses biometric gait data using a suite of machine learning models, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Feedforward Neural Networks (FNN), and Support Vector Machines (SVM). Performance evaluations revealed that the CNN model significantly outperformed its counterparts, achieving the highest test accuracy of 99.06% (with corresponding precision, recall, and F1 scores), particularly when analysing combined data from sensors worn on both wrists. Optuna based tuning further approached near-perfect validation accuracy. The encrypted data transmission added less than 20 milliseconds of end-to-end latency, supporting continuous streaming. This novel approach enhances security by preventing unauthorised access through continuous authentication. This ensures minimal disruption to a user\u2019s daily activities, thereby fostering the trust and acceptance necessary for real-world adoption of assistive robotic systems in healthcare environments.