Some questions on: Which seq is the preview demo video used? whether close-loop and how to initial when rotating without IMU, data structure

[Kin-Zhang]

Hi, Thanks for your work!

I have following question after view this repo and the paper.

1. which seq is the preview demo video used? https://user-images.githubusercontent.com/38326482/189950820-030fd9e4-406b-4d14-8171-43b134344223.mp4
2. whether close-loop use in kisss-icp? since the drift happen on ICP and NDT normally happen with really large map, it perform great with small size
3. How to initial motion when agent rotate really quick without IMU but only odometry velocity? since I met the problem when I refactor the autoware ndt_mapping and running in the kitti dataset in real-time 10Hz in i7-9700H, but when it met the big rotate it will have error calculation(but when slow down the bag play it perform well.
4. Whether the hash table better than octrees in kiss-icp? is there any ablation study to see which data structure is better under this paper? mentioned in the paper: the used data structure can be easily replaced with VDBs [16], [30], Octrees [28], [32], or KD-Trees [2].
5. Is there any comparsion on how sparse the LiDAR that the system cannot provide accurate odom? like 16, 32, 64, 128 LiDAR

[nachovizzo]

1. Sequence 08
2. As explained in the paper, we do not implement loop-closure
3. I can't understand this question, but if the rotation is huge, there is nothing KISS-ICP can do.
4. We did ablation studies internally, and indeed there is never a real difference. We even tried with iKDtrees. The acceleration data structure is just used to speed up NN-seraches, nothing else.
5. We don't have an experiment for supporting this, but we've seen the system struggle with 16-beams datasets.

[Kin-Zhang]

Thanks for the reply.

As explained in the paper, we do not implement loop-closure

Without this one, I think maybe it may fail a lots of suitation. Did you also do some experiments on the how long will make the odom not so accurate and how to reduce this effect?

---

Sorry in advance because the following may be considered an offense. Actually not, just my confusion.

I rechecked the paper, and after reading it (but not so detailed sorry), I was still confused about why the kiss-icp could make the system simple. It's pretty much similar core-method to years ago papers like icp and ndt

for the discussion of the parameters. Maybe A-LOAM only has two parameters in their launch file to change:

    <param name="mapping_line_resolution" type="double" value="0.2"/>
    <param name="mapping_plane_resolution" type="double" value="0.4"/>

also autoware.ai 1.13 who open-sourced at 2019 also involved with ndt methods and their parameters could also not so many for 7/8 if you set same resolution all the time.

And there are also many methods (of course you mentioned some in the paper) like with only one LiDAR could also get the odom/pose for point cloud: hdl_graph_slam, BLAM, RAL2022 VoxelMap

Through all these comparisons (all methods I mentioned above could run only on one LiDAR without IMU, camera etc), maybe you could explain more on how your method could be simple without tuning so many params, and also which part makes the result better than others or above methods.

And of course, and still, you did really great thing to make the icp method open for everyone through python pip install. I also learned a lot from your other works.