Revision history [back]

You're correct, there is no way to handle this situation currently. The process noise covariance matrix is currently assumed to be static, but there are clearly situations where this isn't ideal. This should really be handled by monitoring the velocity and freezing the error growth when the robot stops moving. There are forks that already do this, and I will look into merging those changes upstream. Feel free to file a ticket on the r_l GitHub page.