Mobile robot perception system in ROS 2 on embedded computing platform

Abstract

In this paper a perception system of a mobile robot is proposed and implemented combining a LiDAR sensor with a colour camera. Localisation of unknown objects in an a priori known environment and their classification is required in order for the robot to safety and reliably navigate the working area. To this end, a pipeline is proposed to process and fuse the LiDAR 3D point-cloud data with a monocular colour camera image classifications yielding precisely localised 3D detections with class designations. The proposed pipeline is designed to run on limited computational power embedded platforms in the ROS 2 environment, and has been tested on a robotic testbench.In this paper a perception system of a mobile robot is proposed and implemented combining a LiDAR sensor with a colour camera. Localisation of unknown objects in an a priori known environment and their classification is required in order for the robot to safety and reliably navigate the working area. To this end, a pipeline is proposed to process and fuse the LiDAR 3D point-cloud data with a monocular colour camera image classifications yielding precisely localised 3D detections with class designations. The proposed pipeline is designed to run on limited computational power embedded platforms in the ROS 2 environment, and has been tested on a robotic testbench.