search

leggedrobotics / delora

Self-supervised Deep LiDAR Odometry for Robotic Applications
https://arxiv.org/pdf/2011.05418.pdf
BSD 3-Clause "New" or "Revised" License
238 stars 49 forks source link

About Scan2Map #12

Closed yuqJin closed 2 years ago

yuqJin commented 2 years ago

The experimental results in your paper show that the performance of self-supervised LO+map has improved significantly. Now I have a question to combine the trained odometry with mapping. How did you embed this project in LOAM?

cynosure4sure commented 2 years ago

Hi @yuqJin! We used this implementation of LOAM: https://github.com/laboshinl/loam_velodyne and in LaserMapping.cpp we switched laser odometry topic "laser_odom_to_init" to our learned odometry topic. You have to take care of some frame transformations as LOAM uses a different frame convention than ROS (Z-forward, X-left and Y-up) and that's about it.

hhongwei1009 commented 5 months ago

Hi @yuqJin! We used this implementation of LOAM: https://github.com/laboshinl/loam_velodyne and in LaserMapping.cpp we switched laser odometry topic "laser_odom_to_init" to our learned odometry topic. You have to take care of some frame transformations as LOAM uses a different frame convention than ROS (Z-forward, X-left and Y-up) and that's about it.

Can you be more detailed? Thank you very much!

nubertj commented 4 months ago

Hi @hhongwei1009, Thanks a lot for your message. In what sense are you missing the information? Instead of using the LOAM scan2scan, we were simply using the learned scan2scan. Best, Julian