the auto pilot for a Revolution« kite is a box in the grass controlling all four lines of the kite. It uses a video camera and pattern recognition algorithms to receive information about the state (position and attitude) of the kite and utilizes the mathematical model of RevSim for a model following control algorithm in the control computer.
Fig. 1 AutoRev overview
Every now and then there is a discussion (download and use Netscape Mail) on rec.kites and via private email about old questions and new ideas regarding AutoRev. This discussion significantly supports me to find a reasonable, suitable, and affordable approach towards an automatic Revolution« pilot. I will therefore try to summarize all our thoughts on these pages and take this opportunity to thank you all for your participation in this fascinating project.
The first questions a researcher has to ask himself when developing a new unusual idea are: "What's it all good for? Who's gonna benefit from it? And who's gonna pay for it?". If you want to end up with something that is really pushing farther outwards the frontiers of mankind's knowledge, you need money! I therefore invited the worldwide kite flyers community of rec.kites to find possible AutoRev applications that might persuade the right people. It was incredible how many extraordinary ideas I received (thank you Dave) and I will try to list most of them:
|This list is updated periodically. Please feel free to contribute.|
As you can see in Fig. 1 AutoRev divides into the three main devices of a general control loop:
The actuators emulate the hands and arms of the human pilot manipulating the handles. To a certain extend also the feet/legs of the pilot are emulated, because if we use winches to wind up the lines, AutoRev can also produce his own "wind" (relative airspeed); something a human can only do by running. If we utilize a fifth motor to eccentrically rotate the whole actuator/camera block about a vertical axis, AutoRev can even do true indoor (no wind) 3D flying.
There has been quite a discussion on the best actuator choice and arrangement. Ideas range from: "Give it human touch and try to get two used industrial robot arms" to the good old KISS principle: "Four lines - four motors".
Fig. 2 RoboKite
While the robot version would definitely be the most professional and spectacular one, it has three main drawbacks:
The third argument directly leads to a simplified "body harness" actuator version:
Fig. 3 Three motors, no static forces
Inspired by the way human pilots use harnesses to handle line forces too big for their hands, wrists, and arms, Fig. 3 shows an actuator arrangement with only three motors. The left and right motors operate on two handle substitutes and both motors together can be rotated about the vertical axis by the central motor for sideslip maneuvers.
The main advantage of this version is the elimination of static line forces in the motors. If the handles are fixed to the motor axis at the right proportions (note the asymmetric cross point of handle and motor axis in Fig. 3, since the upper line forces are about four times greater than the lower ones), the motor torque only has to take care of the very much smaller control inputs for steering the kite. The main drag forces are absorbed by the linkage and the bearings. This motor arrangement has to be the choice if a big, strong kite is used e.g. for propulsion. Once again, the disadvantage of this version is the inability to wind up the lines.
And yes, for a practical realization of this concept you would most probably not let the central motor accelerate heavy left and right motors, but construct a lightweight linkage using unaccelarated hydraulic cylinders, Bowden wire, etc.
If we want to scan 3D-volumes for measurement purposes we need to alter the line length from zero to maximum during flight. This can only be done by any kind of winches. The most simple one is a motor winding up the line on its axis (see Fig. 1). You have four lines, you need four motors. Main disadvantage: Every motor has to cover the maximum line forces with its torque.
This left me with the decision whether I wanted to concentrate on high forces propulsion or on 3D-measurement applications in this first demonstration step. I chose the 4 lines 4 motors version in combination with a Revolution« II SUL (small kite, small line forces, small motors).
Taking into account the demand for fast, accurate motors that do not have to be too strong there is just one reasonable choice today: EC motors (brushless DC or synchronous permanent magnet servomotors. These motors (like the 8C Series of ABB) come with an adapted digital converter (controller) that is powered directly by a 230 V or 400 V single- or three-phase mains power supply and controls all states (current, acceleration, speed, position, ...) of the motor.
Sensors basically emulate the eyes of the human pilot. The visual feedback works as the outermost and most important control loop for the pilot to control elevation (up/down), azimuth (right/left) and roll angle (around the lines) of the kite.
On the other hand it is possible to at least stabilize a Revolution« for a certain period with your eyes closed. If you actually try this out and give your adaptive human control algorithms some time to reconfigure, you will be astonished how much information about the state of the kite you can gain from the four different line forces and the corresponding handle moments measured by your hand/wrist tactile measurement system. The line forces could therefore be used in an inner control loop of a cascade control system to quickly take care of inner disturbances.
In addition to the line forces, the line angles measured at the ground can give an indication on the position (elevation and azimuth) of the kite. Even the roll angle can be estimated by the differential angles between the four lines. Naturally this can only work for uncrossed lines, i.e. roll angle much less than 180 degrees. Unfortunately the correlation between kite state and line angle on ground is not sufficiently high due to the low signal to noise ratio. Very small angle changes and wind turbulence acting on the lines make it very difficult to estimate the kite state via line angles. Furthermore the lines (mass and damping) work as low pass filters: It always takes some time until a change in the kite state is transferred via the lines down to the ground.
Many different ways have been discussed on how to directly measure the three state variables (elevation, azimuth and roll angle) of the kite. Like in air traffic control we can distinguish between active and passive measurements. On the one hand the measurement is done onboard the aircraft/kite via gyros, IRU (Inertial Reference Unit), GPS (Global Positioning System), etc. and the information is down linked via radio communication to the controller on the ground; on the other hand the controller himself "looks" at the aircraft/kite with all his sensors (eyes, radar, microwave, sonar, laser, etc.). This passive (as seen from the kite point of view) measurement variant can be enhanced by equipping the kite with any kind of (unique) reflector (transponder) for the controller's "beam" (bicycle reflectors, color/light patches, bar code, etc.).
The first AutoRev approach does not feature line force or angle measurements nor any kind of active measurement onboard the kite. It completely relies on the visual feedback information via a video camera (Fig. 1). This concept enables the auto pilot to learn to fly any kite without having to prepare the kite with special equipment. Additionally, visual information becomes essential if the controlled kite has to be synchronized with other kites, e.g. in a kite ballet.
The prize we have to pay for exclusively exploiting visual information is called "pattern recognition".
Fig. 4 Pattern Recognition
To be defined ...
AutoRev is a research project of J÷rg J. Buchholz at Hochschule Bremen.