Initial-contact-timing Estimation Based on Supervised Learning in a Hip-assistive Wearable Robot

Hip-assistive wearable robots have been widely adopted for gait rehabilitation due to their lightweight design and ease of use. However, the lack of distal sensors, such as shank inertial measurement units (IMUs) or foot pressure insoles, limits precise gait event detection. This study proposes an initial contact (IC) estimation method using only thigh angle measurements and two IMU signals embedded in a hip-assistive robot. The algorithm was validated against IC timings obtained from foot pressure sensors. Using a k-nearest neighbors model trained on kinematic features around IC, the method accurately detected IC events and demonstrated high agreement with the reference data. Moreover, the model achieved a root mean squared error of 2.20% ± 1.15%. These results demonstrate that reliable IC estimation and spatiotemporal gait analysis can be achieved using only hip-embedded sensors, without additional distal instrumentation.