Quadrotor UAV 3D Path Planning with Optical-Flow-based Obstacle Avoidance

This thesis is the result of a six-months research activity conducted at University of Denver (DU) Unmanned Systems Research Institute (DU2SRI), Colorado, USA. The objective of the thesis is the design of a bio-inspired 3D local path planning strategy for obstacle avoidance based on the optical flow field obtained by a frontal monocular camera mounted on a quadrotor UAV, with the requirement of being real-time implementable and runnable by onboard embedded hardware. Assuming the quadrotor is controlled in position and its path is defined in term of a list of waypoints, the strategy consists in generating an intermediate waypoint in order to avoid obstacles based on optical flow horizontal and vertical unbalance for horizontal and vertical avoidance, while for frontally approaching obstacles avoidance the concept of expansion of optical flow field is used. The algorithm is implemented in Robotic Operative System (ROS) as a ROS node, where aforementioned quantities are generated by means of OpenCV Python library. Everything is tested in a simulated environment run by Gazebo in three different scenarios representing horizontal, vertical and frontal avoidance and both software-in-the-loop and hardware-in-the-loop tests are carried out. Performance of the algorithm is measured with software profiling and in terms of execution time for two embedded computer boards such as NVIDIA Jetson TX1 and RaspberryPi 4 by running ROS node on them. The tests show the effectiveness of the algorithm in avoiding obstacles in all scenarios and the capability of the algorithm to be run in real time at a frequency of at least 5 Hz on onboard mountable hardware.