Learning-based feedforward augmentation for steady state rejection of residual dynamics on a nanometer accurate planar actuator system