Occlusion Resistant Object Rotation Regression from Point Cloud Segments
This work addresses a key challenge in robotics for dexterous manipulation, offering an incremental improvement over existing methods by simplifying the pipeline and enhancing occlusion resistance.
The paper tackles the problem of rotation estimation for known rigid objects from point cloud segments by directly regressing a pose vector using a convolutional neural network, achieving competitive performance and increased robustness against occlusion without requiring post-processing.
Rotation estimation of known rigid objects is important for robotic applications such as dexterous manipulation. Most existing methods for rotation estimation use intermediate representations such as templates, global or local feature descriptors, or object coordinates, which require multiple steps in order to infer the object pose. We propose to directly regress a pose vector from raw point cloud segments using a convolutional neural network. Experimental results show that our method can potentially achieve competitive performance compared to a state-of-the-art method, while also showing more robustness against occlusion. Our method does not require any post processing such as refinement with the iterative closest point algorithm.