||For those who have suffered stroke or spinal cord injury, rehabilitation is often the answer for improving gait function. Rehabilitative exercises, which often focus on the legs and deemphasize the role of the upper limbs, are done to help stimulate muscles and exploit neuroplasticity for the diminished functions. However, it has been shown that upper limb muscle activity can induce lower limb muscle activity. It has also been shown that proper arm swing is necessary during gait for balance. This thesis presents the design concept and fabricated prototype of a device that swings the arms during gait rehabilitation. The device is low-powered, lightweight, wearable, and capable of assisting the user's arm swing in the sagittal plane and has unhindered kinematics in the remaining unactuated degrees of freedom. The design comprises three key subassemblies: a backpack frame, an underactuated arm-swing mechanism, and a power train to transfer and amplify motor torques to the arm-swing mechanism. Tests are performed to validate the shoulder-angle prediction equations based on the noncollocated motor-angle sensor measurements, to validate the device's ability to provide adequate torque to generate arm-swing in a passive user, and to investigate whether or not the user's active involvement can be observed by examining motor torque or shoulder angles. The results show that the device does provide sufficient torque to move the arms with a factor of safety, but that the model-based shoulder-angle estimates obtained from the motor measurements have nonnegligible error with the current prototype. It is recommended that a Proportional-Derivative (PD) controller with high PD gains be used with the device because of its low root mean square (RMS) tracking error, shoulder-angle amplitude creation, and ability to diagnose user-assistance level (i.e., is the user passive or actively assisting arm swing) online by observing shoulder-angle amplitudes and peak motor torques.