A Novel Sparse-IMU Convolutional Neural Network for Real-Time Hip Kinematics in Exoskeleton-Augmented Gait

Lauren Low

Wearable robotic exoskeletons have the potential to help people with mobility impairments walk more independently, but they require accurate real-time information about how a user’s joints are moving. Traditional motion capture systems and full-body inertial measurement unit arrays can be accurate, but they are often expensive, uncomfortable, and difficult to set up. In this project, I investigated whether a reduced sensor system using only thigh-mounted IMUs could predict hip motion with accuracy comparable to a full-body configuration. Using convolutional neural networks trained on gait data from both unassisted and exoskeleton-assisted walking, this study shows that sparse sensing can provide a practical path toward more comfortable and accessible rehabilitation robotics.

Engineering
Artificial Intelligence