Improving the imaging speed of atomic force microscopy (AFM) requires accurate nanopositioning at high speeds. However, high speed operation excites resonances in the AFM's mechanical scanner that can distort the image, and therefore typical users of commercial AFMs elect to operate microscopes at speeds below which scanner resonances are observed. Although traditional robust feedforward controllers and input shaping have proven effective at minimizing the influence of scanner distortions, the lack of direct measurement and use of model-based controllers have required disassembling the microscope to access lateral scanner motion with external sensors in order to perform a full system identification experiment, which places excessive demands on routine microscope operators. Further, since the lightly damped instrument dynamics often change from experiment to experiment, model-based controllers designed from offline system identification experiments must trade off high speed performance for robustness to modeling errors.