Localization problem using cartographer with warthog robot

asked 2019-09-11 06:12:21 -0500

cemartin gravatar image

updated 2019-09-11 08:27:05 -0500

Hello,

I am trying to use the cartographer_node with the warthog robot packages, but I get incorrect results for localization.

If I put the IMU data, the odometry from Gazebo and laserscan (LIDAR values) as inputs of the cartographer node, the localization works : the map frame stays at the starting position, the odom frame stays near the map frame (as it is an approximation of the map frame from the sensor datas), and the base_link frame (robot frame) moves along the robot.

However when I remove the Gazebo's odometry data (just using IMU and laserscan as inputs), it doesn't work anymore : the frame base_link stays close to the map frame and the odom frame moves opposite to the robot.

Screenshots : Frames : https://docs.google.com/document/d/1h... Laserscan : https://docs.google.com/document/d/15...

If I do the same with the husky robot it works (even with only IMU and laserscan).

Does anyone know where can the problem come from ? I tried to compare the husky and warthog packages to find it, but I couldn't resolve my problem. Can it come from a lack of points in my laserscan ?

Packages used : warthog : https://github.com/warthog-cpr/ cartographer : https://github.com/googlecartographer... cartographer_ros : https://github.com/googlecartographer... husky : https://github.com/husky/husky

edit retag flag offensive close merge delete

Comments

Please show us your simulation environment with laser scan overlayed

stevemacenski gravatar image stevemacenski  ( 2019-09-11 07:54:15 -0500 )edit

I added a larger Rviz simulation screenshot, so that you can see the laserscan overlay

cemartin gravatar image cemartin  ( 2019-09-11 08:27:57 -0500 )edit

That laser scan looks pretty not great. Is that what it looks like on your robot as well? I would expect SLAM to potentially have problems with that.

stevemacenski gravatar image stevemacenski  ( 2019-09-11 09:04:09 -0500 )edit

It is the maximum number of points that my computer can handle, then the simulation crashes. So you think my problem comes from the Lidar ? That would explain why it works on the husky, it is not the same Lidar sensor. I will try on a better computer

cemartin gravatar image cemartin  ( 2019-09-11 09:36:14 -0500 )edit

I dont think its related to the computer as much as the laser looks terrible, see the occupancy grid its making? its scattered all over the place. Does your actual laser look like that? Perhaps that's why its not performing well since scan-to-scan matching would be really hard

stevemacenski gravatar image stevemacenski  ( 2019-09-11 10:33:29 -0500 )edit