Apr 092010
 

I think my odometry is calibrated reasonably well. Well enough considering that I need to go to a conference for the week (giving two talks) before the AVC and my testing time will be severely constrained. Even worse, I wont have my shop and I’ll have to work from a hotel room and test in an as-yet-to-be-determined parking lot. That’ll be the extent of my fun in Monterrey.

Anyhow, check this out: Several back and forth East-West runs on my street. The straight segments are 100 feet long and then I drive off another ten feet or so and turn around and come back to my marker. Not bad, although I do drift. These results fuse odometry, a gyro, and a magnetometer. There is no absolute position reference.

I start at (0,0) (the red circle); drive East; do a few circles to calibrate the magnetometer, and then I head west. That odd diagonal line is a glitch in the save data. OpenLog is still giving me some grief although Nate says he has a new version of the software I need to try. 
Bottom line: The Odometry EKF appears to be working!
 Posted by at 5:13 PM
Apr 022010
 

I’ve got the Kalman set up and running on the robot. Below is a simulation using real robot sensor data from an open loop test run. The Kalman filter fuses the GPS and odometry data. The process and measurement covariance matrices were filled with educated guesses. I need to tune them and try it in the real robot. Things are coming together.
 Posted by at 4:28 PM