It all depend on your YAML file used when launching the node ethzasl_icp_mapper. The default launch files of the repository are made for large point clouds with a lot of redundant information. The default ICP solution (trigger if no YAML file is given) also assume many points.
I suspect that in your case, the chosen configuration is too aggressive on the downsample and on the outlier filter. We've made the registration solution highly configurable to suit a large range of applications. The downside of that is that you need to tune for your needs.
Here some suggestion to start:
- remove all data filters
- remove all outlier filters
- use point-to-point error
For more information on all module available, you can use the pmicp -l. I think you should find that in the bin folder of the node. This executable can also take two point clouds from disk and a yaml configuration to register point clouds. It might be useful for your tests.
This node rely on libpointmatcher, a library independent of ROS. Lately, I'm putting more effort on maintaining that part. The node ethzasl_icp_mapper is mainly a showcase for it. Don't hesitate to read our tutorial page.
