once the robot sees an obstacle it has a very hard time clearing it from the Octomap
if the robot moves a tiny bit, laser rays stop hitting particular cells, which never get cleared
As a result, when using for example Gripper Click for grasping in clutter it is often necessary to "Reset collision map" first, a step which we'd like non-expert users to not have to go through.
The first step would be a more careful investigation of the phenomenon to get an idea of exactly what kind of spurious or ghost data is left behind.
Possible short-term investigations and fixes
1) increase resolution of laser
vertical resolution can be increased by slowing down the laser
horizontal resolution can be artificially increased by adding one more filter on the laser data (there is already a pipeline in place for that) that supersamples the scan.
2) investigate parameters of existing filters, like the shadow filter, in order to reduce spurious points
Possible solutions which would imply more coding effort
1) increase laser "beam width" explicitly in Octomap by casting more rays
2) add decay for cells in Octomap, likely with purging of old data at serialization time
Problem:
As a result, when using for example Gripper Click for grasping in clutter it is often necessary to "Reset collision map" first, a step which we'd like non-expert users to not have to go through.
The first step would be a more careful investigation of the phenomenon to get an idea of exactly what kind of spurious or ghost data is left behind.
Possible short-term investigations and fixes
1) increase resolution of laser
2) investigate parameters of existing filters, like the shadow filter, in order to reduce spurious points
Possible solutions which would imply more coding effort
1) increase laser "beam width" explicitly in Octomap by casting more rays
2) add decay for cells in Octomap, likely with purging of old data at serialization time