Original title: EfficientPPS: Part-aware Panoptic Segmentation of Transparent Objects for Robotic Manipulation
Authors: Benjamin Alt, Minh Dang Nguyen, Andreas Hermann, Darko Katic, Rainer Jäkel, Rüdiger Dillmann, Eric Sax
In this article, the potential use of autonomous robots in hospitals is explored. These robots could help free up qualified staff and improve patient care. However, the presence of deformable and transparent objects in hospital settings presents challenges to the robots’ vision-based perception systems. The researchers introduce EfficientPPS, a neural architecture that allows robots to accurately perceive and segment these objects. The architecture provides the robots with detailed visual information to assist them in grasping and manipulation tasks. Additionally, the researchers developed a method for unsupervised data collection and labeling to reduce the need for human involvement in the robots’ training process. The effectiveness of EfficientPPS is tested on a dataset containing real-world hospital objects, specifically transparent transfusion bags. The results demonstrate that the architecture is both robust and efficient when used in conjunction with a collaborative robot arm.
Original article: https://arxiv.org/abs/2312.13906