New statemachine

[lexihaberl]

New version of the statemachine which splits the FindGraspPose server into multiple smaller parts:

• Image fetcher, which fetches the most recent images and then unsubscribes to save bandwidth.
• object dector, which only calls the configured object detector
• pose estimator, which calls the pose estimator for known objects
• direct grasppose estimator, which estimates the grasppose without needing a pose estimation of the object
• grasppose estimator, which estimates the grasppose for known objects, based on manually annotated grasp points.

Additional smaller changes include:

• Prefering side-grasps over top grasps to make placement into the shelf easier
• Fall-back to Handover if all placement poses failed.
• Directly doing handover in the case of a top-grasp because most top-grasped objects can not be placed into the shelf due to the robots geometry.