The small autonomous vehicles of the future will have to navigate close to obstacles in highly unpredictable environments. Risky tasks of this kind may require novel sensors and control methods that differ from conventional approaches. Recent ethological findings have shown that complex navigation tasks such as obstacle avoidance and speed control are performed by flying insects on the basis of optic flow (OF) cues, although insects' compound eyes have a very poor spatial resolution. The present paper deals with the implementation of an optic flow-based autopilot on a fully autonomous hovercraft. Tests were performed on this small (878-gram) innovative robotic platform in straight and tapered corridors lined with natural panoramas. A bilateral OF regulator controls the robot's forward speed (up to 0.8m/s), while a unilateral OF regulator controls the robot's clearance from the two walls. A micro-gyrometer and a tiny magnetic compass ensure that the hovercraft travels forward in the corridor without yawing. The lateral OFs are measured by two minimalist eyes mounted sideways opposite to each other. For the first time, the hovercraft was found to be capable of adjusting both its forward speed and its clearance from the walls, in both straight and tapered corridors, without requiring any distance or speed measurements, that is, without any need for on-board rangefinders or tachometers.