Avoiding collisions with other aircraft is an absolutely fundamental capability for semi-autonomous UAVs. However, an aircraft avoiding moving obstacles requires an evasive tactic that is simultaneously very quick to compute, compatible with the platform’s flight dynamics, and deals with the subtle spatio-temporal features of the threat.
We will give an overview of a novel prototype method of rapidly generating smooth flight-paths constrained to avoid moving obstacles, using an efficient trajectory-optimisation technique. Obstacles are described in terms of simple geometrical shapes, such as ellipsoids, whose centres and shapes can vary with time. The technique generates a spatio-temporal trajectory which offers a high likelihood of avoiding the volume in space-time excluded by the predicted motion of each of the known obstacles. Such a flight-path could then be passed to the aircraft’s flight-control systems to negotiate the threat posed by the obstacles. Results from a demonstration implementation of the collision-avoidance technique will be discussed, including non-trivial scenarios handled well within 100ms on a 300MHz processor.