Generating Arm-Swing Trajectories in Real-Time Using a Data-Driven Model for Gait Rehabilitation With Self-Selected Speed.

Gait rehabilitation is often focused on the legs and overlooks the role of the upper limbs. However, a variety of studies have demonstrated the importance of proper arm swing both during healthy walking and during rehabilitation. In this paper, we describe a method for generating proper arm-swing trajectories in real time using only measurements of the angular velocity of a person's thighs, to be used during gait rehabilitation with self-selected walking speed. A data-driven linear time-invariant transfer function is developed, using frequency-response methods, which captures the frequency-dependent magnitude and phase relationship between the thighs' angular velocities and the arm angles (measured at the shoulder, in the sagittal plane), using a data set of 30 healthy adult subjects. We show that the proposed method generates smooth trajectories for both healthy individuals and patients with mild to moderate Parkinson disease. The proposed method can be used in future robotic devices that integrate arm swing in gait rehabilitation of patients with walking impairments to improve the efficacy of their rehabilitation.