In order to enable a robotic system to reach a certain level of (shared) autonomy, it is crucial to overcome the semantic gap between raw sensor data and manipulable objects. Furthermore, precise object poses have to be estimated to guarantee robust interaction during complex manipulation tasks.
To this end, sensory data of multiple different depth sensing devices (TOF, SGM stereo and PrimeSense) are segmented and known objects are matched in their most probable poses. In order to prevent continuous re-computation, the state of the surrounding scene is tracked over time and efficiently updated in case of changes.
- K. Hertkorn, M. Roa, M. Brucker, P. Kremer, and C. Borst, Virtual reality support for teleoperation using online grasp planning, in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, 2013.
- S. Kriegel, M. Brucker, Z.-C. Marton, T. Bodenmüller, M. Suppa, Combining Object Modeling and Recognition for Active Scene Exploration., in Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, 2013.
- M. Brucker, S. Leonard, T. Bodenmüller, G.D. Hager, Sequential Scene Parsing Using Range and Intensity Information, in Proc. IEEE Int. Conf. on Robotics and Automation, 2012.