I am thinking about the possibility to use the accelerometer to calculate the position, the orientation, the speed and the path.

If we give to a program the initial conditions (orientation, speed, position), it could be possible, by integrating accelerations, to know always the position, the speed and the orientation.

Such a program could replace the gps programs in some cases (for exemple, store the plan of a cave) or be used as a magetometer-like device.

I know planes use such a method (Inertial measurement unit) to estimate their position.

The question is to know if N900's accelerometers are precise enough to get some useful result after a while.

the phone can't detect if it's being turned around the vertical axis (not the phones but the one in relation to earth) for example. The Acceleration data doesn't change.
So your logic is somewhat flawed. If you assume that the phone's position stays fixed in relation to the car for example then it might work, since you know a car doesn't move sideways for example.
But you can't do 6DOF positioning with only linear acceleration data. You'd need rotational rate as well.

That is clear.

Thanks !

what do you mean it can't?
www.youtube.com/watch?v=RvXUAM8bnkc

here it can also know if you gonna go upside down

a) you don't have rotate along axis detection if axis is aligned with earth. (hold it with one axis aligned with gravity ( | )and rotate on the axis parallel to earth) (---). Axis should be aligned with accelerometer axis, not phone center. That's why several controllers, as well as phones like Neo Freeruner have two, misaligned.

b) you can never navigate using accelerometer, it's a joke in precision terms. I have already tried integration of data from the accelerometer, using a data logger and a PC. I accumulated errors of up to 50% of actual speed over 30 seconds. At 50 Km/h, that translates into hundreds of meters.

c) just so we're clear, one CAN navigate using accelerometers, planes have inertial navigation systems as backups for decades. Such a device can detect drifts of a few meters in many, many kilometers. Nokia's, after integration, has about 255 steps peak to peak, meaning that one can get drift of up to (worst case scenario) of a meter or so every 10-20 meters meaning that by a block you could be on the next street.

d) I'm still trying. Still logging and still hoping to get some data out of it. One can still calculate 0-100 KPH with it, because in a fast car one does 5-7 seconds, not enough to drift. But in a 30-second journey, errors compound.

You can review the initial problems here. I might get a better result if someone can and will write a better accelerometer logger. Nothing fancy, just time of day and each axis in a decent format, best if CSV. I can't compile for N900, but I can do actual research on the PC side.

ETA: My fast car is in the shop, I'll try logging with the slower car, see if I get a result now that the white stuff is gone from the streets. Accelerometer redivivus.

Also, I now have GPS loggers and a few new ideas. I'm thinking also filming the trip to have actual time index, unless of course, the compression chokes, which it kind of does.

slightly odd question. can the accelerometer data from the sixaxis be read on the N900 with bluetooth? something like this video:

http://www.youtube.com/watch?v=a2rM5Zy5T5Y

It's all relative...

i think as you move faster phone - than you get errors in data. anyway i think it's possible to do like in video with ps3 controller

You're only half wrong here Planes use rate sensors on all six degrees of freedom (acceleration and rotation around all axes). The n900 only covers 3 with bad accuracy.

Which half, actually?