Don't ad-block us - support your favorite websites. We have safe, unobstrusive, robotics related ads that you actually want to see - see here for more.

0 Members and 1 Guest are viewing this topic.

and filled with enough mathematical hieroglyphics to make your head spin

I am finally implementing my own kalman filter (on two separate projects) and I realized that my last post was pretty bad.I am using a kalman filter for sensor fusion in order to take wheel odometry data and merge it with my IMU data to get an accurate heading, and also I am trying to use a kalman filter to "correct" my velocity (I am integrating acceleration to get velocity so it is incredibly noisy IMU data. The main steps are: 1. Build a mathematical model of your system (kinematics / dynamics) and of its error.2. Estimate your current state by running your prior state through your mathematical model.3. KalmanGain=Error * (Weight + Error) (you can use the weighted error to say that you trust one sensor more than another).4. CurrentState= CurrentState + (WhatYouMeasuredHappened-(KalmanGain * Weight * CurrentState))5. NewError= (1-KalmanGain)* Error6. goto step 2The actual math is not to bad, the hard part is accuratly building the system model, the error model, and tweaking the Weights to get accurate results.

I can imagine that confirming your results as accurate could also be very difficult...

The data we use is taken at 10-20 thousand herts, so as you can imagine it's very noisy.

The nice thing is that data taken at that rate is pretty darn accurate once it's integrated for velocity and then position.

I am using a kalman filter for sensor fusion in order to take wheel odometry data and merge it with my IMU data to get an accurate heading, and also I am trying to use a kalman filter to "correct" my velocity (I am integrating acceleration to get velocity so it is incredibly noisy IMU data.

Quote from: Centaur on March 11, 2008, 12:34:15 AMQuoteThe data we use is taken at 10-20 thousand herts, so as you can imagine it's very noisy. WOW that's some super fast data. I have never heard of an accelerometer that outputs data that fast...

QuoteQuote from: Centaur on March 11, 2008, 12:34:15 AMQuoteThe data we use is taken at 10-20 thousand herts, so as you can imagine it's very noisy. WOW that's some super fast data. I have never heard of an accelerometer that outputs data that fast...An analog accelerometer could . . . the very expensive high frequency kind . . . Centaur, what exactly are you measuring?!

Quote from: Centaur on March 10, 2008, 10:34:15 PMThe nice thing is that data taken at that rate is pretty darn accurate once it's integrated for velocity and then position. So usually when you integrate that many times your error actually builds up quicker since you are also integrating the error at each time step.

So usually when you integrate that many times your error actually builds up quicker since you are also integrating the error at each time step.