Global loop closure optimization breaks the map about cartographer HOT 2 CLOSED

First, sorry for the delay.

I experimented with your bag and found some settings that work reasonably well.

The generally guidance is that, as the sensor FOV is decreases, reliance on local SLAM must increase. Since your laser scanner is only 180 degrees, I increased the number of scans per submap, increased the minimum score required for loop closure detection, and decreased the Huber scale:

TRAJECTORY_BUILDER_2D.submaps.num_laser_fans = 270

SPARSE_POSE_GRAPH.optimize_every_n_scans = 270
SPARSE_POSE_GRAPH.optimization_problem.huber_scale = 1e1
SPARSE_POSE_GRAPH.constraint_builder.min_score = 0.8

Enabling correlative scan matching also helps a bit.

I also noticed that the transformation for your IMU seems incorrect. It seems to be oriented z-down and your static transform for it is looks to be around yaw, not pitch or roll.

Lastly, setting the variance of odometry has no effect since you have no odometry. Setting it to 0 in any case would be a bad idea as well.

from cartographer.

Hey damonkohler,

thanks a lot for looking into it! I can confirm that the map is now looking as good as in your screenshot.

You are right that the TF for the IMU is incorrect. I had put the wrong orientation data in the launch file accidently while extracting it from our code.

from cartographer.