Kalman Filter acceleration Integration

Hello, It appears that Kalman filter can yield the position given the acceleration data only (and initial position and velocity, of course). Are there any limitations for this in the real world? Thanks!

Answer 1

[John Petersen]

Accelerometers are notoriously noisy and also drift with temperature and age. The error in your estimate of position will grow over time if there are no position updates, because there will be no way for the filter to correct for the bias without that. With a position update occasionally you can include a bias state that tracks this drift. You will want to make the process noise for the bias very small so that it doesn't dominate the actual signal, but changes slowly with time. To get position, you have to integrate twice so that is good because it smooths out all that noise.