# Position Tracking using IMU

I am working on a robot tracking application, where our main tool (a camera) for locating the x & y position of the robot is working on a quite low frequency. Therefore, I am looking for ways to intermediately track the position of the robot without directly measuring it.

The robot has a built in accelerometer, gyroscope and magnetometer. I know that I can double-integrate the acceleration measurements to get a (very noisy) estimate of my current position. I also can use gyroscope and magnetometer to track the pose of the robot (which I am not primarily interested in).

Is there anything else I could do to improve the position estimate given those three sensors, besides double-integration of acceleration?

• I agree that the Kalman filter is the right way to go with this. That's precisely what a Kalman filter is for. If you find that the accelerometer is not giving you good enough results in between your camera measurements, you may want to consider adding wheel encoders to keep track of the robot position and orientation. Oct 24 '18 at 17:38

Basically the solution is to not try to reference where rob is. What you do is create a notional starting point,and then develop a sensor that measures x&e coupled to a time sensor. You can develop the notional starting points in time and “Tx&e” units. Thus you know where you are once you arrive at a known position.

So NSP X T X x&e = CURRENT POSITION & TIME. Any changes in position applied from known current position gives us a constant position and also a vector

Doesthathelp. Robin

You linked your question with a kalman filter. I guess that's the right way.

Basically it solves your problem: Estimation of a position, based on different sensor signals. It is usually used for sensor fusion with gps data and acceleration data. But you can use any kind of mesuarements of the robots dynamic. It is possible to implement a kalman filter with different sample times for the different signals. I can edit this post later, if I have my lecture notes with me for the equations. But maybe you find them with the help of google in the mean time.

If the quality of your available signals is not good enough, you could maybe use additional bluetooth bacons. But I never tested this by myself and read the accuracy is quite low. How about a laser distance sensors to allocate the distance to a wall? What kind of robot do you use?

Another interesting concept is the particle filter.