Researchers from the Universidad Politécnica de Madrid in Spain have developed an automatic air navigation and collision avoidance model using an automatic learning system. Based on how human beings learn to perceive motion, the model builds software using a simulator-based learning process. Built into the aircraft avionics, the resulting software aligns aircraft flight paths with other nearby aircraft in a dense airspace like an airport, avoiding collisions.

If an aircraft is heading towards a building, it automatically avoids the obstacle. Likewise, if several aircraft are to land on the same runway, the system prioritizes landing maneuvers without human intervention. The model has experienced all the major combinations of in-flight incidents in the simulator, learning how to solve from the simplest to the most complex contingencies thanks to its bio-inspired algorithm.

The model can be tailored to each aircraft according its flight characteristics. The software built by the model in simulation is capable of automatically navigating an aircraft and steering it to a target or consecutive targets. During navigation, the aircraft is capable of automatically avoiding collision with basically any physical obstacle in its path: other aircraft, land, geographical features, and buildings. Additionally, it can avoid and skirt areas with adverse weather conditions, military operations areas, national airspace, etc.

The model has proved capable of avoiding all possible collisions with other aircraft, even if more than two aircraft are heading for the same spatial coordinates. Unlike current collision avoidance maneuvers, the maneuvers are not merely vertical (ascend/descend). Additionally, it is the pilot that activates collision avoidance maneuvers today.

The model implements totally automatic 3D maneuvers, where aircraft perform turns with ascents and descents. The model controls all the maneuvers from start to finish, without any pilot or air traffic control involvement, or remote steering. To date, the model has been used in a simulator with notable success, but its designers aim to implement it physically in real air navigation systems as an additional safety measure against human error.

The toughest problem in the implementation of this model was to automatically apply all pilot and air traffic controller experience to avoid collisions not only with other aircraft, mountains, land, or buildings, but also in very dense air traffic environments, as in the vicinity of civil aviation airports where an aircraft may be surrounded by other planes, and perhaps also be only a short distance from the ground.

Source