Yet another visual auto pilot system looking beyond the limitations of accelerometers and gyroscopes.
From
The Engineer:
The technology of unmanned aerial vehicles (UAVs) has advanced so far that they’re now mainstream technology, particularly in the military arena. Despite this, they are still developing, and a the major gaps in the use of one of the most versatile varieties of UAV may soon be closed.
Helicopter UAVs (HUAVs) have all the advantages of manned helicopters, such as hovering flight and the vertical take-off and landing which allows them to be operated from virtually any area. However, landing an unmanned helicopter isn’t easy, and it’s the one part of their operation which cannot be carried out autonomously. A trained operator has to land the craft by remote control, and remote landing of a helicopter is particularly tricky. A cushion of air builds up underneath the craft as it descends, and this has to be spilled, by slowing down the rotors at a controlled rate, before it can settle on the ground. Any mistakes tend to lead to the helicopter tipping over, which is catastrophic, as the large amount of energy stored in the spinning blades will then be dissipated by smashing the whole craft (and probably any equipment on board) to pieces.
Engineering electronics developer Roke Manor has been developing a system to automate HUAV landing, and it’s based on a technology which is familiar to millions — but not from where they might expect. Who could think that the system which tells tennis umpires whether a ball is in or out, or shows TV viewers whether a delivery is LBW, might guide a drone helicopter in to land?
The Hawk-Eye system was developed by Roke in the 1990s, and works by analysing the data from fixed cameras to extrapolate information about how a ball is moving in three dimensions. The HUAV landing system works in exactly the same way, but in reverse — the camera is mounted on the moving helicopter, and its on-board data processes use its images to work out the motion of the helicopter in relation to its landing position, which is generally (but not always) fixed.
In fact, as Roke’s Future Technology Manager Peter Lockhart explained, the system was developed for use on any UAVs, as they all need to be remote-piloted for landing, unless they are to be recovered by parachute. ‘But we were testing this on our own site at Roke, and it just isn’t big enough to fly fixed-wing UAVs. As it happens, helicopters are the hardest to land anyway, so that suited both our purposes — we could control the experimental and testing phase without going offsite and tackle the most challenging aspect of the problem.’
Ed Sparks, consultant engineer at Roke and one of the original developers of Hawk-Eye, said that the relationship between the two systems is direct: ‘Hawk-Eye tells you where the ball is, we look at the landing pad and work out where we are in relation to it.’
The visual processing system works out the helicopter’s roll, pitch and yaw in relation to the ground. There are distinct advantages to using this system rather than accelerometers and gyroscopes, which give an absolute measurement of orientation, Sparks explained. ‘With accelerometers, gravity is a very large acceleration which is applied constantly while the craft is flying, so to prevent you from confusing gravity with your own motion, you need an extremely accurate accelerometer,’ he said. ‘Also, the accelerometer tells you your attitude relative to where you started, so if it’s running throughout an hour’s flight, that’s an hour’s worth of errors it could have accumulated.’
The visual system, on the other hand, is unaffected by these sorts of errors. ‘You turn it on when you’re starting to make your landing approach, and you see exactly what it sees on the ground,’ Sparks said. ‘The landing system measures your position relative to the specified landing spot, from exactly where you are to exactly where you want to be, so it’s minimising the errors from the word go.’
One of the most important criteria for developing the system was that it had to be entirely self-contained on board the HUAV. ‘We don’t want any reliance at all from information passed up from the ground,’ Sparks said. This meant that all the image processing hardware had to be on board as well as the camera itself. The camera and the UAV itself were off-the-shelf products, and Roke brought in SME UAV manufacturer Blue Bear Systems, which developed a new variant of a lightweight computing platform with bespoke video units to house Roke’s image processing software. The team also worked with the aeronautics department of Bristol University, which has a long history of working with autonomous systems, to work on the control theory for the system, in particular the algorithms which take the visual measurements and turn those into guidance commands for the helicopter.
Another partner in the collaboration was MBDA, a large aerospace manufacturer, which brought its expertise on flight control algorithms to bear on solving the problem of what happens when the landing platform is moving as well as the UAV —if it has to land on a flat-bed truck, for example. ‘They do a lot of work on controlling two platforms to optimum use,’ Sparks said. Roke acted as the system integrator as well as providing the UAV itself and the image processing know-how.
The result is a system which allows the UAV to land in any conditions where the ground is visible. ‘Basically, we can operate in any conditions in which a piloted helicopter can operate,’ said Lockhart. ‘Landing at night isn’t a problem. Thick fog would be, but you wouldn’t be flying in those conditions anyway.’
The system requires no human intervention at all to land, and in many cases the UAV will have a camera trained on the ground in any case, as many UAV flights are for reconnaisance purposes. However, among the possible applications for this system is the unmanned, autonomous resupply of troops in difficult-to-reach locations, reducing the risk to helicopter pilots and other personnel.
The next phase of the research is aimed at making the system even more user-friendly, with features such as point-and-click navigation. ‘And unskilled operator could just click on an area he or she wanted to investigate, or wanted to designate as a landing area,’ Lockhart said.
The Roke team was particularly pleased with the speed of the project. ‘We’ve brought a lot together in a very short time,’ Sparks said. ‘We started in the spring of 2009, and we were landing the helicopter by the summer. In the autumn we did our first landing on a moving target. We’re now in a position to start selling the system, and we have a number of leading UAV vendors who have it in trials to decide whether they want to install it on their platforms. These UAVs are multi-use, so the first application depends on who buys it, but it’s likely to be defence or police."
