Australia and Singapore have conducted a trial of collision-avoidance technologies for swarming UAVs with a 97% success rate claimed for the suite after intensive testing at the Woomera experimental range in central Australia in December.
Those trials saw six AAI-Aerosonde Aerosonde UAVs flown in a networked environment alongside six simulated UAVs.
The cooperative development effort, known as Project Careful, is being jointly conducted by the Australian Defence Science and Technology Organisation (DSTO) and Singapore’s Defence Science Organisation.
Dr Anthony Finn, head of DSTO’s automation of the battle space initiative, says that the trials explored multiple collisions scenarios, commencing with a single real UAV against a simulated aircraft, progressively building up to 12 UAV formations. “We ran a 103 serials of which we have calculated that there were greater than 750 deconfliction events.”
Speaking earlier this month at the Association for Unmanned Vehicle Systems International’s (AUVSI) Unmanned Systems Asia Pacific 2007 conference in Melbourne, Australia, Finn said: “We started with real and simulated, as you might imagine, in a one on one, and debugged the system and ironed out the wrinkles.
“As we became more confident we went from simulated to real one-on-one, and then real-to-real one-on-one, then from real-to-real one-on-a-few, and to real-to-real one-on-many. We got up to six real UAVs flying and deconflicting, and six simulated flying and deconflicting, all in the same airspace.”
The underlying deconfliction technique is based on the UAV swarm sharing their respective positional data in real time during a mission.
Algorithms developed by the project calculate each UAV’s own position relative to the known position of neighbours over a 180º forward-facing hemisphere. Those calculations are made on a cyclic basis by each UAV’s own autopilot system, with positional and flight path projection updates across the hemisphere monitored in one second intervals.
Each projection looks forward some 10min in terms of the UAV’s anticipated flight path, assessing “the risk associated with each individual path that this, the own UAV if you like, could possibly have bearing in mind the information that it has on the presence of the other UAVs”.
Where potential conflicts are identified, each individual UAV responds in a three layered approach based on distance and time. On initial identification of a possible collision at a distance the UAV performs a vertical deconfliction manoeuvre. “It determines a trajectory that minimises the risk profile.”
If this fails to achieve the desired result, the UAV then reacts with a horizontal course correction. In the final inner layer the UAV undertakes an abrupt horizontal course change if collision remains imminent.
Finn said that the underlying algorithm objective emphasises cooperative and non-aggressive manoeuvre solutions in the outer two layers. No details on the miss distances associated with each layer have been released.
He told the conference that failure in the trials “was defined not a collisions but by an incursion into that final collision avoidance zone”.
Development of algorithms was achieved via a series of increasingly sophisticated simulations, ending with these being fed by a simulator into an Aerosonde hardware in the loop test bed ahead of the trials.
Finn said the volume of data needing to be shared between individual swarm aircraft using the deconfliction system remained relatively low and was unlikely to place pressures on available network bandwidth.
The algorithm also has potential to incorporate other airspace deconfliction data provided by non-swarm sensor assets Finn said, such as a single UAV fitted with an air search radar or an airborne early warning asset. This could provide a swarm with the ability to respond to non-cooperative or hostile aircraft in the same airspace.