Outdoor Navigation under Tree Canopy for Unmanned Ground Vehicle (UGV)

Abstract

TThis paper is focusing on implementing the Global Localization System (GPS)-an independent localizing method using a single sensor so that the UGV can drive autonomously in the GPS-denied environment. This paper aims to measure the performance of the localization between using GPS navigation and Visual Odometry (VO) navigation. The UGV platform is built with Intel Realsense T265 as VO camera for visual navigation, Ublox M8N as GPS receiver for navigation evaluation purposes, Pixhawk as the autopilot, and Raspberry Pi 4 as the companion computer for the autopilot. To perform this paper, the Mission Planner will be used as Ground Control Station to set the parameters and settings for the autopilot and compose the missions to the UGV to be executed in form of waypoints. From the results, this paper has achieved its objective where the error when using VO navigation compared to GPS navigation is 0.21 meter. Hence, using the VO camera itself can provide comparatively the best option to replace the GPS receiver for navigation by using the same settings set in this paper where the PID gains of the steering are P=0.00528620, I=0.14430414055, D=0.019767, IMAX=79, FF=0.811, and the setting for the throttle are P=0.9681, I=0.9966, D=0.0484, IMAX=140. Different values of PID gains, data filters, or more recent camera technologies can improve the results of the VO navigation.