Mobile clients for telepresence robots are cluttered with interactive elements that either leave a little room for the camera feeds or occlude them. Many do not provide meaningful feedback on the robot's state and most require the use of both hands. These make maneuvering telepresence robots difficult with mobile devices. TiltWalker enables controlling a telepresence robot with one hand using tilt gestures with a smartphone. In a series of studies, we first justify the use of a Web platform, determine how far and fast users can tilt without compromising the comfort and the legibility of the display content, and identify a velocity-based function well-suited for control-display mapping. We refine TiltWalker based on the findings of these studies, then compare it with a default method in the final study. Results revealed that TiltWalker is significantly faster and more accurate than the default method. Besides, participants preferred TiltWalker's interaction methods and graphical feedback significantly more than those of the default method.