Posted by
g.bartoli on
URL: http://official-rtab-map-forum.206.s1.nabble.com/A-bunch-of-general-questions-tp856p861.html
1) Ok, I understood the LccIcp parameters. Is there a reason why PoseScanMatching is not used in Localization mode? Shouldn't it give better accuracy also for navigation by registering laser_scan onto grid_map?
2) Ok!
3) Ok, I will recompile and try this again, thanks! I'm also looking forward to improve wheel odometry, in particular for in place rotations.
4) Ok, we don't have a IMU, so Odom/ResetCountdown seems the best choice, thanks! Furthermore, I see that a seemingly more complete package for EKF fusion exists in ROS, it is called
robot_navigation. Do you have any experience with it? Do you think it can be better than robot_pose_ekf? Here is the developer
announcement.
5) Our motor_driver node publiches both /odom (without covariance) and /tf (using differential wheel model) at 50 Hz. I'm planning to integrate RGBD Visual Odometry, so I think connecting /odom to rtabmap is the best choice.
6) Ok!
7) Ok, understood. However, I noticed that when I tried ICP 3D (this also happens with RGBD/OptimizeSlam2d = false), I had problems with navigation, since the graph is optimized in XYZ and it is not planar, so every time I set a Navigation Goal I receive an error that the final pose has an invalid Z coordinate, but it is not a problem, since my application is for non-holonomic 2D navigation.
8) Do you think is there a way to apply the calibration obtained in the paper to the ASUS Xtion Pro? I read that they perform a computationally expensive SLAM approach to derive the calibration, but once it has been obtained, it can be applied in under 2ms to the depth image to get a far better point cloud. However, they did not publish any opensource library or final calibration file for the sensors they tested.
9, 10, 11) Nice, thanks!
However, as others already stated, let me say that RTAB-Map is one of the best written piece of software that I ever used, congratulations Mathieu! :)
~Guido