Revision history [back]

Yes, this would be expected behavior.

You are asking the filter to estimate your position through double-integration of linear acceleration data (acceleration -> velocity -> position). If there's even a small bias in the acceleration data (and I see, your velocity estimate will grow without bound, which will make your position estimate take off. Similarly, the covariance for both linear velocity and position will also explode, which will cause huge instability in the filter.

TL;DR, you need a velocity or position reference.

Yes, this would be expected behavior.

You are asking the filter to estimate your position through double-integration of linear acceleration data (acceleration -> velocity -> position). If there's even a small bias in the acceleration data (and I see, see your acceleration has non-zero values in X), the velocity estimate will grow without bound, which will make your position estimate take off. Similarly, the covariance for both linear velocity and position will also explode, which will cause huge instability in the filter.

TL;DR, you need a velocity or position reference.

Yes, this would be expected behavior.

You are asking the filter to estimate your position through double-integration of linear acceleration data (acceleration -> velocity -> position). If there's even a small bias in the acceleration data (and I see your acceleration has non-zero values in X), X and Y), the velocity estimate will grow without bound, which will make your position estimate take off. Similarly, the covariance for both linear velocity and position will also explode, which will cause huge instability in the filter.

TL;DR, you need a velocity or position reference.