Flight plan

Aerospace engineers at Imperial College London hope to develop a control technique that will enable unmanned aircraft to fly in formation autonomously.

It will increase the speed of an advanced control technique known as model predictive control (MPC), in which the controller repeatedly updates in real-time its planned operation and chooses the best options, accounting for variables.

The technology has previously been used in chemical refineries where, for example, a controller can be programmed to avoid exceeding a certain temperature and can adjust its operations if it is in danger of reaching the set limit.

‘Predictive control is a new control technology because you are doing real-time optimisation as part of the control,’ said Dr Paul Goulart, head of the project.

‘You decide right now what you are going to do for, say, the next 10 seconds and come up with a little plan. You carry out the first part of the plan, then just repeat it over and over again, which allows you to take into account constraints.’

The limitations that would need setting for aircraft to fly in autonomous formation would include the safe distance one aircraft would need to maintain from another — usually less than one aircraft distance apart.

To make MPC work in an aircraft, the researchers will need to develop algorithms that are sufficiently robust to ensure safe, reliable operation, and fast enough to keep within the dynamics of a plane.

Currently, a slow rate controller, such as one in a chemical plant, would provide a new decision every 60 seconds. Depending on the number of variables in a system, Goulart said it is possible to run the MPC at a few hundred hertz.

‘Dr Adrian Wills in Newcastle, Australia, has got one to run at 25,000 terahertz [providing 25,000 decisions a second] for controlling the vibration of a metal beam. That is very encouraging and way faster than we need for a plane. A speed in the order of 50hz would be sufficient,’ he said.

The final speed of Goulart’s system will depend on the number of variables it would need to take into consideration and the length of time the system would need to plan for. While the researchers have not yet determined the length of look-ahead time, they want a system that can plan as far ahead as possible.

‘If you need your controller to run at 50Hz, then you just figure out how far ahead in time you can look in one 50th of a second and that is how far you look and you hope that works,’ said Goulart.

‘You can, however, look a little less far ahead, which means your controller can update faster — no-one knows what the best combination is.’

He said there would be no more than 12 variables in the aerospace application (there were three or four in the metal beam application) and include the position of a plane relative to the others, velocity, heading and pitch angle.

‘That is a relatively big number for this MPC technology but you could reduce it by making various approximations and so forth,’ said Goulart.

‘If you want your plan to extend out 15 seconds of flight ahead, you are already talking about quite a big optimisation problem to solve.

Goulart hopes to be able to demonstrate the technology in the autonomous flight formation application in Imperial College’s full-motion simulator by the end of the project, scheduled for 2011.

The researchers are also investigating the possibility of using MPC in another project to develop anti-hijacking technology that would provide a reactive level of autonomy.

This means that the controller would allow a pilot to fly his desired path if it is safe, but not if it is dangerous; that is, it would veer him away from a marked circumference around potential targets predefined in the aircraft’s electronic database, such as government buildings.

‘Ultimately we would like to put a real pilot in our simulator and say “see that building there? Fly into it”, and see if we can keep him from doing it while giving him as much control of the plane as possible,’ said Goulart.