Dense Point Cloud Mapping in Autoware.Auto not working properly

asked 2021-07-12 13:41:55 -0600

RDaneelOlivaw gravatar image

Hi,

I've been using Ubuntu20 ROS2 Foxy Autoware.Auto stack for generating custom PointCloud maps for autonomous driving. I managed to generate good point clouds, although they are really rough and downsampled.

image description image description

When trying to use smaller voxels, the localization in the mapping phase just doesn't work and it generates corrupted maps like the ones I show here.

error map

I'm using this code and configuration.

ndt_mapping.launch.py

ndt_mapper.param.yaml

scan_downsampler.param.yaml

So my questions are:

1) Why is it not working properly and losing localization so easily when it has perfect TF coming from the Virtual odometry and even with BIG voxels if I drive too fast?

2) Shouldn't this be done using normal localization (meaning using GPS, wheel-steer odometry, and lidar ) and not only hard tfs?

3) How is this done in real cars because in the simulation we can do whatever tricks we want but it won't translate to real-world solutions.

Hope I gave sufficient info, if not please tell me so and I'll try to give better info.

edit retag flag offensive close merge delete

Comments

I faced with the same issue https://answers.ros.org/question/3847...

Aleksandr Savel'ev gravatar image Aleksandr Savel'ev  ( 2021-08-24 06:58:01 -0600 )edit