Simulated prosthetic vision

A methodology for developing and evaluating encoders without a patient: render, on a head-mounted display for a sighted subject, the phosphene image a given array and encoding would produce, then measure how well the subject reads, navigates, or recognizes. Computational virtual-patient models (notably the open-source pulse2percept) go further, predicting each electrode's distorted percept, its axon streaks, fading, and spread, from array geometry and stimulation.

Simulated prosthetic vision is the field's main fast-iteration loop and, when differentiable, lets encoders be optimized end to end. Its validity is limited by how faithfully the phosphene model matches real evoked percepts and by the fact that a sighted brain is not an adapted, long-blind one.