In this package we estimate the grasp position for the robot.
Therefore we assume that there is a PointCloud which describe the recognized object.
We filter these PointCloud with a PassThrough filter.
Here you get more informations about the PassThroughfilter
Then extract the leftest and rightest point from the filtered_cloud as grasp_points for the PR2.
Publish these two points and for visualization on rviz the whole filtered_cloud.

For information on how to use estimate_grasp_positions, read the "How does it work" section of the chair_grasping page.