Wayfinding is an important activity that can be performed with limited visual resources, and thus may be an important application of early visual prostheses. In a pair of experiments we explored minimal visual resolution requirements of a simulated retinal electrode array for mobility in real and virtual environments, experienced by normally sighted subjects in video headsets. In experiment 1, inexperienced and experienced subjects traveled similar routes around a suite of offices with simulated implants of 4 x 4, 6 x 10 and 16 x 16 dots. In experiment 2, the effects of adding dynamic noise and removing a subset of 'phosphenes' from a 6 x 10 dot array on the mobility of experienced subjects through a series of different virtual 10-room buildings were determined. Performance was quantified in terms of time and navigation errors in both experiments, and wall contacts in the real environment; a compound score was also computed for trials in experiment 1. In experiment 1, inexperienced subjects required 16 x 16 dots for adequate performance, while experienced subjects reached similar levels with 6 x 10 dots. In experiment 2, dot removal up to 30% led to modest yet significant performance deterioration, and noise addition to slight but non-significant improvement, while practice led to a reduction in travel time by 50% over the 28-trial experiment. Error counts in experiment 2 were fairly high, but largely randomly distributed, and attributable to the high risk of becoming disoriented in the sparse visual environment. Substantial performance level differences were found between subjects, spanning a threefold range even after practice. The findings suggest that a retinal implant with as few as 60 electrodes may provide independent wayfinding abilities to the adventitiously blind, but that substantial practice and supervision will be required in learning this task.