Description
Hello,
I am currently working on an extension to optimize parameters to achieve better map results relative to its ground truth.
Using my robot I recorded a bag file of a larger environment and used the cartographer on this recording. While I was experimenting with this recording, I generated multiple maps without changing any of the parameters or the bag file. Also I reduced the number of background threads to 1 as well as the ceres threads.
For several maps I aligned them with their ground truth and already visually it was clear to see that those maps differ from each other. I read through the paper and also the issues here and couldn't find any problems related to that.
I have also read through the documentation for the scan matching of ceres and have found out that also other people complained that non deterministic behavior can occur. Is this non-determinism related to ceres?
I was wondering now if there is a way to avoid this non deterministic behavior?
Thank you in advance for the response.