why some locations are sparse in new pointcloud map

[improve100]

@koide3 I have a question to ask, the following picture is the effect of my test with hdl_graph_slam. Above is the g2o graph, below is the laser point cloud map, the red circle indicates the corresponding position of the two maps, I don't understand what caused the position g2o map is very sparse. The point cloud map is also very sparse. Can you give me some advice?

[koide3]

@improve100 , If you did it with a ROS bag, could you try to slow down the playback speed? I guess the CPU power was not enough, and the SLAM module dropped some frames...

[improve100]

@koide3 Thank you for replying so quickly.This is the configuration of my test computer, the resources should be enough. I also tried to play at a rate of 0.5 and 0.2, the effect is similar to the previous one. What other reasons can cause this phenomenon? Or how should we remedy this phenomenon?

[koide3]

@improve100 , OK, your PC is definitely powerful enough for this package. Could you upload your launch file so that I can check some parameters?

[improve100]

@koide3 I also found another problem. In the scan matching node, ndt can easily calculate that the car is backward, that is, it falls into a local optimal solution. The actual car is moving forward. Especially in the corners.

[improve100]

this is my launch file test.txt

[koide3]

@improve100 , About the scan matching problem, maybe ndt_resolution is too small for outdoor. Larger resolutions (2m ~ 5m) are better for such environments.

I'm not sure what is the cause of the sparse map problem. But, could you try to change the parameters as follows:

• g2o_solver_type: gn_var_cholmod
• g2o_solver_num_iterations: 64
• graph_update_interval: 1.0