search

PRBonn / semantic_suma

SuMa++: Efficient LiDAR-based Semantic SLAM (Chen et al IROS 2019)
MIT License
920 stars 205 forks source link

Is SuMa++ able to do the global localization (i.e. re-localization)? #39

Closed WilliamWoo45 closed 3 years ago

WilliamWoo45 commented 3 years ago

Hi @Chen-Xieyuanli,

Thanks for providing such great project. I'm wondering is the SuMa++ able to do the semantic-based global localization (i.e. re-localization) ?

Besides, is there any plan to extend the current SuMa++ with the capability to work with our own 3D LiDAR dataset?

Cheers!

Best Regards, William

Chen-Xieyuanli commented 3 years ago

Hey @WilliamWoo45,

Thanks for following our work!

We recently have no plan to extend SuMa++ for global localization, but we do have some LiDAR-based global localization methods (overlap-localozation and range-mcl).

You could use SuMa++ for your own 3D LiDAR dataset, as long as you could get the point-wise semantic information for your own scan. You could either re-train RangeNet++ or use your own semantic segmentation method.

WilliamWoo45 commented 3 years ago

Hey @Chen-Xieyuanli,

Thanks a lot for your prompt response! Definitely I'll try the ovelap-localization and I'll stay tuned for your code releasing of range-mcl.

BTW, apart from ovelap-localization & SegMap & ScanContext & NDT_mapping_localization, is there any other open source LiDAR-based global localization algorithm you could recommend?

Best Regards, William

Chen-Xieyuanli commented 3 years ago

Hey @WilliamWoo45, they are all good work, but may focus on different tasks.

You may first specify which problem you want to solve, place recognition or metric localization. For metric localization, you may also tell the difference between pose tracking (with an initial guess) and global localization (without an initial guess).

Since the problem we are discussing is not about SuMa++ anymore, I would like to close this issue. If you have any further questions regarding global localization, we could discuss them via email. My email address is xieyuanli.chen@igg.uni-bonn.de

I hope my answers are clear and helpful. I'll also appreciate it if you could star or folk our repo if you like it.

Best, Xieyuanli

WilliamWoo45 commented 3 years ago

Hi @Chen-Xieyuanli,

Thanks for your comments. Surely we can further discuss by email. I've starred this repo.

Cheers!

Best Regards, William