LiDAR Odometry - 5 Minutes with Cyrill
Vložit
- čas přidán 2. 10. 2022
- LiDAR Odometry explained in 5 minutes using KISS-ICP as an example
Code: github.com/PRBonn/kiss-icp
Series: 5 Minutes with Cyrill
Cyrill Stachniss, 2022
Credits:
Video by Cyrill Stachniss
Thanks to the KISS-ICP developers Ignacio Vizzo, Tiziano Guadagnino, Benedikt Mersch, Louis Wiesmann, and Jens Behley
Thanks to Igor Bogoslavskyi and Olga Vysotska
Intro music by The Brothers Records - Věda a technologie
Really good video. I find these 5 minutes are a great way to refresh material.
It is extremely cool that the KISS-ICP developers decided to release their code, especially in such a polished manner. Too much research code is left to rot on university hard drives :(
Thanks
At startup, it gave an error. I updated the firewood and it helped me)
Thank you so much for sharing exceptional content and research. The effort is really impressive and have remarkable impact for whole robotics domain and autonomous vehicles. I love this channel and especially the high quality of the content. Great also that the channel is shared among the other team members. Good luck. Have a nice day!
Thanks!
Thank you so much for your talk, These series are great.
Auf den Punkt in 5 Minuten!
Cannot wait to try KISS-ICP with point cloud from livox HAP lidar, which has FOV 120
Thank you for the explanation.
WOW thank you so much Sir.
It really works, thank you.
Great , happy to hear that!
Очень интересно)
Really nice video. Right now I am researching about the lidar SLAM algorithms. Especially for the tunnel environment where it is dark and not manny features are present. I am wondering about the performance of the KISS-ICP in such environment. Also in the paper there is comparison of KISS-ICP with other state-of-the-art algorithms. I am wondering why algorithms such as Fast-Lio2, LeGO-LOAM, SC-LeGO-LOAM, LIO-SAM were not included in the comparison? Thank you
I'm just making assumptions here. KISS-ICP is a LiDAR Odometry method, which only takes LiDAR's point cloud as inputs, maybe that's why Fast-LIO and LIO-SAM are not counted as comparisons since they are LiDAR-Inertial methods. SC-LeGO-LOAM contains a global point cloud descriptor to perform loop closure and backend optimization, and KISS-ICP mainly focuses on frame-to-frame alignment.
I see, thank you for your input
Hi Prof Stachniss, Thank you for sharing the novel KISS-ICP! I wonder if KISS-ICP works fine as well for 2D laser scanner.
+
Yes it does, out of the box, although not designed for it 🙃
This is great! I figure that, at 40Hz processing frequency, a moving environment won't affect the algorithm too much... But have you been conducting tests in highly dynamic environments? The paper mentions a few standard data sets you tested on, but I'm not familiar with these. Would be great to get a quick idea how dynamic environments can get for this to still work.
It works quite well. Simply run it on your dynamics ans see how it works for you - running it ist faster than me typing in the dataset names on my smartphone 😉
Hey cyrill any lidar odometry which we can use for 2d lidar
Kiss-icp should also work in 2d