Integration and Testing of Planner Methods on AgIRoM: An Agile Vision-based UAV Platform

Abstract

The primary objective of this senior thesis is to continue the development of the AgIRoM, focusing on the integration of various planner methods to demonstrate the capabilities of the platform in a live navigation example. AgIRoM is a vision-based quadrotor platform that largely extends upon the work conducted by the Robotics and Perception Group (RPG) at the University of Zurich (UZH) on Agilicious through the addition of a depth-based motion planning pipeline. In particular, the development of AgIRoM was the main focus of my work during the past two years in the Intelligent Robot Motion Lab (IRoM). This report aims to discuss the process of successfully integrating two novel planner methods: the first is a method described in Perceive with Confidence (PwC) developed by researchers in IRoM, and the second is Ego-Planner (and its successor, Ego-Planner Swarm), a lightweight gradient-based planner developed specifically for quadrotors. The project was able to reach the live-testing phase with Perception Guarantees (the name of the GitHub repository of PwC, these terms will be used interchangeably) with some initial success but was unable to conduct full extensive testing due to the deprecation of AgIRoM's state estimation systems (discontinued end-of-life support for the visual-inertial odometry camera, and calibration deterioration for the motion capture system). Consequently, testing for the integration of Ego-Planner was done fully in simulation. During the integration and testing phase, it was found that a majority of the challenges arose from incompatibilities in hardware and their respective proprietary software packages. In an effort to address this, a Zed Mini camera - which has both state estimation tracking and depth estimation capabilities - was tested as a substitute for both cameras onboard the Agilicious framework.