François Pomerleau

Specifications: - [x] Encode our loosely defined formatting rules, see [aseba rules](https://github.com/aseba-community/aseba/blob/master/.clang-format) - [ ] Limit automatic parsing to PR - [ ] Add delimiters to protect from automatic formatting,...

I fine with the idea. Could you prepare a PR?

Yes, we are tracking those solutions (and others). The short terms goals are: - NDT (point-to-Gaussian and Gaussian-to-Gaussian) - GICP (plane-to-plane) We are slowly finishing the math to have them...

I've seen a couple of people implementing ICP on a GPU and loading the point clouds on the GPU takes a lot of time, which reduces a lot the performance...

1. Filters related to [downsampling](https://github.com/ethz-asl/libpointmatcher/blob/master/doc/Datafilters.md#down-sampling) will change the order of points. They should actually squeeze them by removing deleted points. The ones doing descriptor augmentation should not change the order....

You're turning every rocks for optimization ;) So, for organized point clouds, they tend to do a projection on a 2D grid to improve speed. Doing a projection imply loosing...

Part of those were implemented at the application level in [ethzasl_icp_mapper](https://github.com/ethz-asl/ethzasl_icp_mapping/blob/reintegrate/master_into_indigo_devel/ethzasl_icp_mapper/src/dynamic_mapper.cpp#L131).

From what I recall, libnabo only use pointers, so the memory footprint should be small. I'm not sure of what you have in mind for _saving_ the kd-tree. You want...

@aguenette , this seems to be from the Python binding. Any idea?

@aguenette any thoughts?