Abstract
HeartLander is a miniature mobile robot which adheres to and crawls over the surface of the beating heart to provide therapies in a minimally invasive manner. Although HeartLander inherently provides a stable operating platform, the motion of the surface of the heart remains an important factor in the operation of the robot. The quasi-periodic motion of the heart due to physiological cycles, respiration and the heartbeat, affects the ability of the robot to move, as well as localize accurately. In order to improve locomotion efficiency, as well as register different locations on the heart in physiological phase, two methods of identifying physiological phases are presented: sliding-window-based and model-based. In the sliding-window-based approach a vector of previous measurements is compared to previously learned motion templates to determine the current physiological phases, while the model-based approach learns a Fourier series model of the motion, and uses this model to estimate the current physiological phases using an Extended Kalman Filter (EKF). The two methods, while differing in approach, produce similarly accurate results on data recorded from animal experiments in vivo.