MPU6050 accelerometer double integration doesn't work

[mrimskiy]

Hi everyone!

Is there anyone here who has experience in integrating the MPU6050 twice to get the displacement? I have worked on it for months, but without any result. After reading in different forums that a very small iteration time was needed when integrating an accelerometer twice, I decided to work on my iteration time and managed to get an iteration time of 3 ms. However, even this did not give me any satisfying result. The way I integrate twice is as following:

acceleration reading => low-pass filter => high-pass filter => integration => high-pass filter => integration

I have also tried all sorts of combinations and cut-off frequencies, with bias compensations, or with only one high-pass filter, without low-pass filter, you name it, but the results I get are far from being satisfying...

I hope someone has the answer!

Thanks in advance!

MrImskiy

[jremington]

@mrimskiy

I have worked on it for months

Sadly, you have wasted a lot of time. Double integration with any consumer grade IMU (let alone the long obsolete MPU-6050) will always fail, due to sensor noise and inability to accurately estimate 3D orientation.

This article explains just one of the major problems: Using Accelerometers to Estimate Position and Velocity | CH Robotics

[mrimskiy]

Thanks for your reply @jremington! It is indeed sad...

[jremington]

Double integration of IMU data is certainly possible, but the IMUs accurate enough for that purpose cost on the order of USD 10K-50K.