The mems accelerometer is mounted on a quadcopter. Because the quadcopter is very small and flying indoors, it is basically impossible to use GPS to do sensor fusion. The position information is mainly used as feedback signal for autopilot mode.
I haven't tried Kalman filter or other filter technologies yet. And because of the thermal mechanical error I understand the accuracy couldn't last for a long time. Currently my target is to maintain a relatively accurate position signal in one minute.
Thanks.
Well, put it to the test. Fly the quadcopter manually in hovering condition, register the accelerometer data and then double integrate it offline and see what it gives for a time window of one minute.
This would not work practically, as normal MEMS accelerometers have a noisy output signal. In spite of some assumptions this is not a white noise. In consequence you integrate the mean value twice and you may "measure" a change in position, which is not there. In addition to that, a drift of the quadcopter is mainly at low speeds and low accelerations. This results in a bad signal-to-noise-ratio and the detection of a starting motion is very difficult. You have to check the datasheet of the used accelerometer and maybe exchange it with a better one to achieve a quite constant position for one minute. But you can't really ensure, that it is repeatable.
I agree with the previous comments. Try it. It is alway interesting for academic purpose and to definitively understand (hands-on experience) that double integration of the signal will lead to an acceleration of the drift; the platform will end up moving quite rapidly from the set point. If a short duration flight stabilisation is needed, you could consider (with the platform standing on the ground and after termal stabilisation of the MEMS chip) evaluate the per-axis (or 3D) drift of the position. However, only relying on accelerometer will not work. This is why almost all MEMS IMU pack together 3 axis accelerometer, gyro and compass and yes, all bound together with filters as the Kalman. All this said, go foward and experiment, it is instructive. Oh! if you still consider only the accelerometer, try to find the 3D center of rotation of the platform to minimise level arm effect. Few words about your choices or experiences here will be appreciated!
Regard.
Thanks for everyone's answers. I have tested the accelerometer and integrated the signal twice when I put the accelerometer at rest. Theoretically the position and velocity should be zero, but because of the existence of drift and noise the position and velocity keep increasing with time, which is predicable. Because the standard deviation of noise is proportional to time, let alone the constant drift. I think magnetometer and accelerometer are very useful to compensate the gyro, but I am not sure ifI can use magnetometer and gyro to compensate the accelerometer?
It is very good to carry out experiments and see how things go in reality. Magnetometer and gyro do not provide information about translational position or velocity and therefore they are not useful. Are you limited to accelerometers? There are other options for indoor navigation out there. If you give more information about the size of the quadcopter, its payload and calculation capabilities, someone may be able to better help you.
Dear Ying,
this approach works only theoretically. In fact, you integrate among others time variant nonzero meanvalue bias of the accelerometer output. This will produce incipient position error.
Good morning Ying,
Depends on your application, you may use an optical flow technique with a chip built for optical mouse. There is numerous example even with Arduino board (and yes someone have wrote all the code :-) ) This technique will provide you impressive precision in the x and y axis and less in the z axis. There is good communities of interest (read serious hobbyers) that do good work out there; have a look to DIYDrones.
Regards.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Process