Rovio in Carla simulator - covariance grows quickly

I am trying to run Rovio in the Carla simulator (A simulator for autonomous driving research). I am having a major issue. In the frame visualization (the window which shows the tracked features and their covariance), there appear to be plenty of features being tracked very well. However, if I look at the topic /rovio/pose_with_covariance_stamped or /rovio/odometry, the covariance of the pose grows continuously, very quickly (at a rate of 10 to 100 per second). Within typically 1 or 2 minutes, Rovio completely diverges and stops tracking any features.

Here (Or here, same file in a different location) is an example dataset, which includes all of the necessary calibration files, and my results from Rovio. To run rovio with those calibration files, from within that folder, run:

rosrun rovioovio_node _filter_config:=$(pwd)/rovio.info _camera0_config:=$(pwd)/carla.yaml

I know that tracking movement along the optical axis of the camera is the most difficult situation for Rovio. Should it be this difficult, or is this behavior completely unexpected (and therefore indicative of a deeper issue with my specific setup)?

Has anyone else had any luck using Rovio within Carla, or any other simulator?

Thank you all for any help you can provide.

ethz-asl / rovio

Rovio in Carla simulator - covariance grows quickly #232