Ask Your Question
0

RTAB-Map with odometry and laserscan

asked 2016-01-29 15:56:14 -0500

Jack000 gravatar image

just getting started with ROS. I've got RTAB-map working with a kinectv2, but I'm seeing loss of tracking when there aren't enough feature points in the camera view.

I was wondering if adding odometry input would alleviate this problem in the actual robot? Would a laser scanner improve accuracy as well?

I'm contemplating buying a laser scanner as per: http://wiki.ros.org/rtabmap_ros/Tutor...

but I'm not certain how the laser scan interacts with RTAB-map. It says that the laser scan is used to construct a 2d occupancy grid, but that can already be done without RTAB-map. Is the 2d laser scan data used to improve performance in the 3d SLAM at all?

edit retag flag offensive close merge delete

Comments

As you said the problem is when there are not enough features on the current image. The odometry info probably improve the perfomance. I don't think that laser scan alleviate that, it is use to get better results on the mapping but not have any influence on the issue that you mention.

jcardenasc93 gravatar image jcardenasc93  ( 2016-03-18 20:23:43 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted
0

answered 2016-03-21 14:39:59 -0500

matlabbe gravatar image

Hi,

RTAB-Map can use laser scans to refine odometry transformations (e.g., wheel odometry). So yes, if you have a LiDAR, this will improve the map. See Kinect+Odom+2D laser configuration for example (many RTAB-Map videos used this configuration). Visual odometry is not used in this configuration to avoid the problems you mentioned (the environment has many white walls).

cheers

edit flag offensive delete link more

Comments

Thanks for the info and I really like your loop closure method

jcardenasc93 gravatar image jcardenasc93  ( 2016-03-21 22:49:05 -0500 )edit

Can you use RTAB-Map to build a 3D map with Laser/Odom information ONLY? Any tutorial I have seen uses some form of depth image/optical image to build up a pointcloud/point map.

I know that the whole purpose is to use imagery, but I would my current robotic application my not allow for that.

atomoclast gravatar image atomoclast  ( 2016-12-19 16:11:48 -0500 )edit

2D laser scans (LaserScan input) or 3D laser scans (arbitrary PointCloud2 input) can be used to output a 2d occupancy map or a 3d cloud/octomap respectively. A camera is required only for loop closure detection (if you sync fake empty images, you could get what you want but without loop closure).

matlabbe gravatar image matlabbe  ( 2016-12-22 12:05:54 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

Stats

Asked: 2016-01-29 15:56:14 -0500

Seen: 1,731 times

Last updated: Mar 21 '16