Title: Fast and High-Quality GPU-based Deliberative Perception for Object Pose Estimation

Pose estimation of known objects is fundamental to tasks such as robotic grasping and manipulation. The need for reliable grasping imposes stringent accuracy requirements on pose estimation in cluttered, occluded scenes in dynamic environments. Existing methods either require large sets of pose annotated training data to learn object poses or incur a high runtime in a search for the best explanation of the observed scene in a space of rendered scenes. In this work, we introduce PERCH 2.0, a deliberative pose estimation approach that takes advantage of GPU parallelization to achieve an order of magnitude speedup over previous deliberative methods. Further, we introduce a combined deliberative and discriminative framework for 6-DoF pose estimation that achieves a higher pose estimation accuracy than purely data-driven approaches without the need for any ground truth pose annotation. We evaluate our approach in several domains such as object articulation, conveyor picking and tabletop object manipulation.

Maxim Likhachev (advisor)

Oliver Kroemer

Paloma Sodhi