IMU Brick 2.0 get position

[JPannetier]

Hello everybody,

 

I'm trying to implement a code (Python) to get position x y z from IMU Brick 2.0 accelerometer, in live mode.

In the first step, I implemented the rotate accelerometer by quaternion to remove the gravity and rotation of my sensor.

After i integrate directly to have velocity and position :

example :

Vx(k) = Vx(k-1)+Ax(k)*dt

X(k) = X(k-1) + Vx(k)*dt 

 

Nevertheless, into my accel. data I have an offset.

 

Somebody would have already implemented this type of code to remove the offset (drift) ? to get the position

 

I don't know if i must a kalman filter or other....

 

Thank you

[pwab]

Hello JPannetier,

 

I wanted to do the same a while back and did some research and I want to help not wasting time on this. To sum things up: you can't get good position information over time by only using an IMU. This topic is called Dead reckoning (https://en.wikipedia.org/wiki/Dead_reckoning) and is way harder to get running as it seems to be at first sight.

 

A lot of people had this idea and ended up using something different. See here for reference:

 

• https://stackoverflow.com/questions/26476467/calculating-displacement-using-accelerometer-and-gyroscope-mpu6050
  
• https://electronics.stackexchange.com/questions/4983/how-to-determine-position-from-gyroscope-and-accelerometer-input#4986
  
• https://electronics.stackexchange.com/questions/96722/tracking-2d-positioning-with-imu-sensor
  

 

The problem is noise and precision of measurement when using acceleration or velocity data over time for estimating a position. To get a grasp of the topic I can recommend watching this video (the starting position marks the explanation of why your double integration leads to a drift):

 

In fact VR-headsets use sensor fusion by combining their IMU data (gyroscope and accelerometer) with their optical IR-light-based marker positions (also called pose estimation). I can recommend a very good talk by Philipp Zabel about HMD tracking systems (german language but images and animations should give you some impressions) and some related videos on youtube:

 

• https://chemnitzer.linux-tage.de/2018/de/programm/beitrag/288
  
• 
  
• 
  
• 
  
• 
  

 

So what you could do is to:

 

• Add internal sensors which track the current position with markers, by scanning a known environment with markerless tracking algorithms or by using systems like GPS
  
• Add external sensors which track the position and orientation of your system and use it to recalibrate your IMU data with at least more than 30 Hz
  
• Use position sensors which will inevitably affect the shape and size of your system (https://en.wikipedia.org/wiki/Position_sensor)
  

 

Hope this helps

 

Regards,

Philipp