Exploring symmetries in 6D pose estimation models

To carry out robotic tasks in the real world it is often required to know the 6D pose of an object. A big hurdle here is that the pose might be ambiguous because some objects are or appear symmetric. There is already a great variety of existing models for pose estimation with very diverse approaches. A lot of them are aware of the symmetry problem and have come up with different approaches that not always succeed to solve it. In my thesis, the handling of symmetry will be investigated further. Existing models will be analyzed with this emphasis and different possible solutions will be presented. While the majority of state-of-the-art models has only been evaluated on synthetic data, in my thesis they will also be tested on real data.