SporeAgent: Reinforced Scene-level Plausibility for Object Pose Refinement
This work addresses pose refinement for objects in cluttered environments, which is crucial for robotics and augmented reality applications, but it is incremental as it builds on existing RL-based registration approaches.
The paper tackles the problem of inaccurate object pose estimates due to observational noise, segmentation errors, and ambiguity from symmetry and occlusion, by introducing a method that uses scene-level physical plausibility to reduce ambiguity, resulting in state-of-the-art performance on LINEMOD and YCB-VIDEO datasets.
Observational noise, inaccurate segmentation and ambiguity due to symmetry and occlusion lead to inaccurate object pose estimates. While depth- and RGB-based pose refinement approaches increase the accuracy of the resulting pose estimates, they are susceptible to ambiguity in the observation as they consider visual alignment. We propose to leverage the fact that we often observe static, rigid scenes. Thus, the objects therein need to be under physically plausible poses. We show that considering plausibility reduces ambiguity and, in consequence, allows poses to be more accurately predicted in cluttered environments. To this end, we extend a recent RL-based registration approach towards iterative refinement of object poses. Experiments on the LINEMOD and YCB-VIDEO datasets demonstrate the state-of-the-art performance of our depth-based refinement approach.