Future rover missions will be enhanced through the addition of science to the planetary traverse phase. Scientific targets are selected through a random search and salient gradient tracking in the visual field, which requires both a search algorithm and a reactive pan-tilt camera controller. This thesis presents a cerebellar-like reactive pan-tilt controller to track salient targets in the visual field as the rover moves based off the cerebellar models and the human vestibulo-ocular reflex. An online neural network using an EKF training law is used as a feed forward controller and it's
performance is compared to standard batch and online neural network training techniques. The controller was then applied to the Barrett WAM to control the manipulator wrist.
The online EKF trained network is able to adequately model the internal dynamics of a pan-tilt, while remaining stable due to the continuous learning. This is shown in both simulation and practice.