I am running mrpt_ekf_slam_2d (http://wiki.ros.org/mrpt_ekf_slam_2d) on an indoor robot equipped with an RPLidar. Landmarks are fed to the ekf by running a clustering algorithm on the lidar pointcloud.
I am having a problem that the ekf does not seem to associate observations with landmarks located behind the robot, that is with a yaw angle outside the interval [90, -90]. Those observation are simply just added as new landmarks. Landmarks are mapped and associated correctly when in front of the robot, but as soon as they end up behind it, the ekf spams out multiple landmarks: 1
Here is the result of the same run when filtering the lidar scan to only include points within [90,-90]: 2
Has anyone encountered a similar problem using mrpt?
Interesting! It looks like a bug in mrpt-slam... please, could you file a bug report here for us to investigate it?
A example dataset / test case would be great.