Re: [Paparazzi-devel] (no subject)

[nisma]

> Quoting xx xx <address@hidden>:
> 
> 
> > > What kind of mouse are you talking about ? sloted disk and infrared
> fork ?
> > > How do you use this ?
> > Sloted disk, i have found a similar modification on a internet page, i
> can
> > seek it out or make some images from one of my modifications.
> > Basically, you can modify such a mouse to be a really accurate
> roll/pitch
> > sensor. The tilt sensor is used to calibrate the mouse and to map the
> > relative information of the sloted disk to a fixed 12 bit position.
> 
> That is very clever. I would appreciate to see a pic or plan.
I can made pics next week.
> 
> But I think that what you are measuring is acceleration (~forces). 
> It can indicate attitude (pitch + roll) only when the plane is flying a
> level
> straight line at constant speed (uniform mouvement?).
> For example, when the plane is flying a stabilized symetrical curve, the
> acceleration is exactly perpendicular to the plane, just as if it was
> level.
> 
No, i'm measuring real angles (degree). With the addition of a atmospheric
pressure sensor, i`m able to stabilize the plane and drive it on a higher 
level/ abstracter basis. This sensors gives no information of the real or 
relative position of the model. 
In reality, small inertia forces exists.
The model benefits from it. When a force is applied to the model,
the inertia reacts in the opposite direction of the force and stabilize
the model in this way. 
For the location information, i have experimented (hooked up the receiver)
to time the trasmission signal. On my equipment, the signal is repeated
constantly every x uS. Synchronizing the MCU on this frame, the cpu is able
to 
calculate the distance from trasmitting TX based on it's delay compared to
the calculated one.
If the MCU switches between
two receiving antennas immediatly after every frame where the antennas are
located between 1/4 and 1/2 lamda , it gots the direction of the
transmitting signal without losing any information for the receiver.
Two possible directions. One should be eliminate by software or tree
antennas should be used to got only one direction.

I have tested the two things separatly. I don't have written any code,
that allow the UAV to follow determinate waypoints using the above described
methods.

Its has to be noted that i use simpe RC transmitter/receiver without MCU's
in it. If the trasmitter use a MCU , the timing between the beginning of
one frame an the next frame is not necessary constant.
> 
> 
> > If i see this correctly, you can archive this cheaper using cheap CDS 
> > photo sensors with a polarisation filter.
> 
> I don't know. The Melexis chips are non contact temperature sensor. There
> are
> details on these components in this paper
> http://www.ctie.monash.edu.au/hargrave/horizon_sensing_autopilot.pdf 
> 
> If you try with cd sensor let me know :)
> 
Theoretically, the teory is similar to the follow paper.
When you have the capability to calibrate it (self learning algorithm)
or using a small self lerning neuronal network, or adapting your SW,
it should be possible to migrate from measuring the sky temperature to 
measuring the sky polarisation. (if there is a difference at all).

http://www.araa.asn.au/acra/acra2001/Papers/Usher.pdf
http://www.uni-tuebingen.de/cog/personpages/schmolke/gwalPaper.pdf

> 
> > Can you tell me more about the fine tuning and calibration .
> 
> We get the attitude of the plane using difference of temperature.
> 
> We have two pair of sensors, one "looking" front and back, the other one
> left
> and right.
> 
> T1 is the difference of temperature between right and left
> T2 is the difference of temperature between front and back
> 
> roll = K * T1
> pitch = K * T2
> 
> We get an initial K from a contrast measurement before the flight. Then,
> during
> the flight
> we compute a roll from the GPS heading variation rate, assuming we are
> flying
> symetrical curves.
> 
> We use a least square algorithm to fine tune K.
> 
> (this is in lls.c)
> 
> This is not very important for just stabilizing the plane, as the low
> level loop
> is very tolerant regarding gain setting. But the upper level navigation
> loop is
> far less tolerant and this really improves things. It also adjusts for
> fast
> changing contrast conditions, like sunrise.
> 
OK, now the operational principe is clear to me.
You measures the difference of temperature between the sky and the earth
using a simple pid algorithm. In addition, the integral temperature of the
sky and earth can be used to estimate the altitude of the model.
The termopile have probably a 30-40 degree sensitive angle.
More higher the model fly, the sensor register more sky and minor earth.
This all assuming no montains.

Using polarized photo sensors, we can measure the angle between the 
incoming light (anzimuth) and the polarized sensors. No altitude estimation
is possible. 
The pid controller can be the same as this you already used it. 
The virtual anzimut changes every 10-15 minutes and depend on the 
seasons and longitude/latitude.
For a short time (1-2h), it's predictable using a 15 minutes calibration
time. This can be reduced to seconds, if calibrated values exists from the
same latidute/longitude. 10 degree longitude/latitude difference don't
matter. It should be teoretically possible to calculate all major different
location and store it on a table. Storing only the differences from the
previous value, the table should shrink very well.

This works even at night, if a clear sky exist.



> 
> Hth
> 
> Antoine
> 
> 
> 
> 

-- 
NEU : GMX Internet.FreeDSL
Ab sofort DSL-Tarif ohne Grundgebühr: http://www.gmx.net/dsl