Perception Pipeline of Autonomous Mobile Robots Using 3D Point Cloud Classification.

Robotics can be used in structured environments such as industrial manufacturing plants and warehouse storage. The paper discusses the necessity of a strong perception pipeline for mobile robots in dynamic environments such as malls and airports. To create accurate maps of the environment, we highlight the importance of both 2D and 3D mapping and point out the issues with manual 2D map cleaning, manual creation of restricted zones, and manual classification of dangerous locations. The process of manually gathering and processing the 3D point clouds from real-world environments was tedious and time-consuming. Hence, one of the contributions of the paper is the generation and labelling of such point clouds from NVIDIA Omniverse scenes. We reviewed the technological advances that have made it possible to deploy robot fleets in structured environments and gave examples of sectors that have successfully used mobile robots with perception pipelines. The advantages of the perception pipeline implementation are also discussed, including improved path planning, navigation, and obstacle avoidance.