|
From: Jean-Christophe Z. <Jea...@ep...> - 2001-08-29 11:18:05
|
Hello Curt, and everybody, Thanks a lot for taking time to give me good answers. > > My sensors will probably be phototransistors placed on the leading edge > > of the wing, for example. Each of these will sense the light from a cone > > of about 30=B0. Let's assume an elliptic wing with a 80cm span, if = we put > > one of these sensors every 2cm, this lead to a total of 30 > > phototransistors. >=20 > So each sensor will return the "average" brightness of the scene > within it's cone? =20 Yes. > How would you then leverage this information to > control the aircraft? Would the floor be a different color than the > walls? How different would it have to be? OK. I have only preliminary ideas about this but for the first step I will probably try to set up a room with specifically painted walls. My first thought is to have b/w sinusoidal patterns both on the 4 walls and the floor. Then the controller (between the sensors and the actuators of the plane) will be a neural network whit "genetically evolved" characteristics. I don't know if you are familiar with evolutionary algorithms but what I can say is that they require a lot of individual trials with randomly chosen values. That's why we need a simulator: we can let the real plane crash several times into the walls just because the currently tested individual is one of the early generations and is simply not able to fly and/or avoid obstacles. > > Could you be more precise? I have no experience with 3D modeling program > > except OpenGL. Could you also give me some clues to replace the standard > > "real world" environment by my simple "box", within FlightGear? >=20 > This is perhaps more involved than I can deal with in one message. > But for each scenery tile, there is a .stg file which you can add > objects to. If you have the box model saved in one of the many > standard 3d file formats that plib supports, then you can just add it > in directly. So what you think is that I should keep the whole FG Earth structure, build a sort of box on its surface, and fly in it? OK. Then I red some of your documents about the tile structure of the Earth. But could tell me more about how to edit one of the *.stg file and merge my "box" to it? What kind of tools would you use for that purpose? > > If you want to see my preliminary (for the moment R/C) aircrafts, take a > > look at http://dmtwww.epfl.ch/~jzuffere/ISFI/ (this page is written in > > French, except the pictures ;-). I'll soon prepare an English version > > with some more "scientific" overviews. >=20 > Wow, it seems rather ambitious to be able to mount that much > intelligence on a small plane requiring such small weight. =20 Yes it is ;) But it is exactly what we would like to test: whether genetic algorithms can lead to lightweight, simple, and fast image processing. See http://dmtwww.epfl.ch/isr/east/team/zufferey/project.html. Cheers, jcz |