Drone autonomy software is the software that flies an aircraft, finds its way and works with other aircraft without GPS and without a link to an operator. It is also called an AI pilot. GPS is the satellite signal that most drones use to know where they are. It can be blocked, or faked, in a conflict. A radio link to an operator can be blocked too.
An ordinary autopilot follows a list of points on a map and uses GPS to know where it is. It cannot handle a surprise, such as a hill that was not on the map. Autonomy software replaces the person in three ways. It estimates the drone's position without GPS, by watching the ground below with a camera and comparing it with a stored map. It plans and changes the route, for example to go around a hill. And it avoids other aircraft, using ideas like those in satellite collision avoidance.
For instance, an inspection drone enters a long tunnel at a hydropower plant in Himachal Pradesh. Inside, there is no GPS and no radio. The software watches the tunnel walls with its cameras, keeps the drone in the middle and records the cracks it sees. It flies out when the battery falls to 30%, which leaves enough charge for the way back.
It uses computer vision to see. Engineers first test it on a computer with autonomous vehicle simulation, because a mistake in the air destroys a drone. Related software drives cars (autonomous driving software) and steers boats (unmanned systems). The same software flies each drone in a swarm. These steps are the levels of autonomy. Software that moves a real machine without a person is a case of an AI agent that moves real things.