/elevation_mapping topic data

[Timelees]

I can publish data in the /elevation_mapping topic and display it in rviz, but the data field part of the /elevation_mapping I publish is nan.

I saw in another issue that you solved this problem by reducing the variance. May I ask which specific variance values were adjusted? Furthermore, when using sensors of the perfect type, the effect will be better than that of the struct_light type. Is it necessary to adjust those parameter values in the configuration of the struct_light sensor?

[Timelees]

And why is there also a conical height map at the bottom of the body during scanning

[Muhammad540]

Hello, sorry for late reply! I cant remember the values on top of my head but i can share the following details.
First you need to find the root of the problem, what is exactly causing the variance to explode and only after that go over adjusting the parameters like min/max variance, multi height noise and distance threshold. I have given brief description of my understanding about these parameters. I guess the following things might be causing such high variance: sensor noise , conflicting measurements for a cell, robot position uncertainty.

• min_variance and max_variance : These parameters control the minimum and maximum allowed variance for the elevation of each cell in the map. If the variance of a cell grows too large, it can lead to instability (the Nan's you see)

• multi_height_noise: This parameter adds noise to the variance when a new measurement is significantly different from the existing elevation. This is intended to help the map adapt to dynamic objects, but if it's too high, it can cause the variance to explode.

The mahalanobis_distance_threshold is also important for deciding how to fuse new measurements.A smaller value will make the filter more conservative, relying more on the existing map. A larger value will make it more aggressive in incorporating new measurements.

Also for struct light sensors, they can have different noise profiles and characteristics. I recommend creating a dedicated configuration for this sensor, adjust the variance and Mahalanobis parameters for it.

[Muhammad540]

And why is there also a conical height map at the bottom of the body during scanning

it is very likely due to the sensor picking up parts of the robot's own body. I think the library uses a CropBox filter to exclude points that fall within a specified 3D box around the robot's base. This is configured in your sensor's parameter file. You'll need to define the dimensions of a box that encompasses the parts of your robot you want to ignore.

[Timelees]

mahalanobis_distance

Thank you for your reply. During the troubleshooting process, it was found that the covariance of the rotating part of the robot approaches 0 (that is, the 33 matrix at the lower right corner of the robot pose variance matrix in the figure approaches 0, and even after adjusting the relevant parameters, it is calculated to be 0). This led to mahalanobis_distance tending towards infinity. Then, does the value of my pose covariance matrix meet the requirements? The 33 matrix in the upper left corner should correspond to the x,y, and z directions. However, the first two dimensions seem quite large (the value is around 20, and adjusting some parameters in my Kalman filter doesn't improve it much). May I ask what is the normal range for these data?
During the testing of the structured light sensor, I also found that the data of the /elevation_mapping topic was nan at the beginning. However, after I moved the robot for a period of time, some of its data became numerical values while others were still nan values. This led to an anomaly when constructing the foot constraints of the robot. What might have caused this change?
Thank you for your project contribution and reply. Also, could you recommend any other open-source visual perception projects related to elevation maps implemented with ros2?