Covariance on wheel odometry is incorrect during start-stop (ros ticket #4138)

[ahendrix]

In some situations, the navstack gives the base controller many very short commands. The result is that the base odometry thinks it is moving for a very short time, and then thinks it is standing still for a very short time. Because the covariance of the base controller jumps to a small number when the base is stopped, and a large number when the base is moving, this covariance is also changing rapidly.

When driving, the covariance is large, when standing still the covariance is small. And when the wheels are vibrating, we have something in the middle. But, when the wheels are vibrating rapidly, the wheel odometry is actually very poor. So the covariance should actually get larger than the driving covariance; instead now it gets smaller than the driving covariance.

We can try to detect this situation, and modify the covariance. Or, when this is too hard, we can simply keep the driving covariance until the base has stopped for e.g. one second. Then, vibrating wheels will have the same covariance as in the driving situation. This would already be a great improvement over the current situation.

This problem is for example causing intermittent failures in the door opening on HW, messing up our demo's occasionally.

trac data:

• Owner: sachinc
• Reporter: wim
• Reported at: Tue Apr 6 16:39:55 2010
• URL: https://code.ros.org/trac/wg-ros-pkg/ticket/4138