Embedded Mobile ROS Platform for SLAM Application with RGB-D Cameras

Abstract

Mapping and localization are essential for various robotic applications. Without a GPS signal, indoor navigation remains a challenging task. With recent progress in machine learning, embedded systems, and sensor technologies it has become practical to use computer vision algorithms to perform localization and mapping tasks on low power, portable robots. The focus of this project was to create a tracked vehicle with autonomous navigation capabilities for use in traversing difficult terrain. Such an unpredictable environment posing additional challenges to our navigation system. Navigation is achieved via processing source stereo image provided by an Intel RealSense D435i depth camera using the NVidia Jetson Nano, a CUDA-enabled single-board computer. The design heavily relies on the usage of the Robot Operating System (ROS) and Simultaneous Location and Mapping (SLAM) algorithms to enable flexibility in future feature expansion and ease of communication.