Obstacle Detection and Avoidance Techniques for Unmanned Aerial Vehicles

Unmanned Aerial Vehicles (UAVs) represent a high potential for many fields of applications, from surveying to search and rescue missions, from delivery to photography. For all these applications, the interest in this field become higher for those that require automatic missions and autonomous flight operations. In order to consider autonomous flights, one of the most essential requirements is the capability of detect obstacles and avoid them through reliable mechanisms. Nowadays, obstacle avoidance is an incredibly active field of research and development, considering that several aspects of this area are still unknown. For example, obstacle avoidance in three dimensions is less developed than two-dimensional methods, but represents a cardinal point for all types of systems that can move in three-dimensional space. Especially for multirotor UAVs, the industry is increasingly evolving towards the development of complex systems capable of autonomously carrying out different types of missions. For this reason, it is becoming necessary to develop collision avoidance systems that are capable of handling the vehicle in the three-dimensional space. In this thesis project, a three-dimensional collision avoidance algorithm has been used and optimized. Studying advantages and disadvantages of various methods, it has been chosen the most appropriate one and it has been designed an UAV equipped with suitable sensors, to detect and avoid obstacles providing a reactive collision avoidance system. The chosen algorithm is the 3DVFH+, that computes obstacle avoidance manoeuvres in a reactive manner, using an octomap to determine where the obstacles are located given the vehicle position in a 3D environment. Hence, it has been optimised for the specific studied application and tested in a simulation environment. Parallel to the collision avoidance methods analysis, the specifications of the Guidance, Navigation and Control (GNC) system of the Unmanned Aerial System has been defined and developed as part of the BLUESLEMON project. In particular, hardware and software components of the avionic system have been analysed and the experimental setup has been defined. Subsequently, the development of a simulation environment has allowed to simulate and test the system. Some assumptions have been made on different possible scenarios that the UAV could face in the real world and which could prove the functioning of the simulated collision avoidance system. The main scenario is an outdoor environment in which is required to grant semi-autonomous system for landslide monitoring. A multirotor has been designed to perform this type of flight missions. It has to autonomously detect and avoid obstacles that can be or appear along the path previously planned. Finally, this thesis attempt to implement, simulate and optimise a collision avoidance method with the aim of validating an implementable and testable system in the real world. Simulations results reveal the functioning of the system, which is able to maintain the predetermined distances from obstacles, carrying out automatic missions. For each obstacle detected, the algorithm recalculates the optimal route to be carried out to accomplish the flight mission. However, more work is suggested in order to improve the accuracy and the sensibility of the method and to perform real world flight missions.