Real-time Reactive Trip Avoidance for Powered Transfemoral Prostheses

This paper presents a real-time reactive controller for a powered prosthesis that addresses the problem of trip avoidance. The control estimates the pose of the leg during swing with an extended Kalman filter, predicts future hip angles and hip heights using sparse Gaussian Processes, and reactively plans updated ankle and knee trajectories with a fast quadratic program solver to avoid trips. In preliminary experiments with an able-bodied user who purposefully lowered the hip to elicit trips on each swing, the proposed control reduced the rate of tripping by 68% when compared to a swing control that follows standard minimum-jerk trajectories. In addition, the proposed control also reduced the severity of toe-scuffing. To the best of our knowledge, this controller is the first to incorporate visual feedback in the real-time planning and control of a lower limb prosthesis during gait. The results demonstrate the potential of reactive and environment-aware controls to improve amputee gait robustness and encourage future development of leg prosthesis controls that can react in real-time to the environment and user state.