Risk-aware path planning and replanning algorithm for UAVs

During last decades employing Unmanned Aerial Systems (UAS) to perform different types of operations has surely brought enormous benefits. Nonetheless allowing such operations requires great attention, the people safety is in fact one of the main obstacles to be evaluated before any operations could start indeed because of their high density on working environments. Therefore, the aim of the project conducts to the creation of a structured low-altitude airspace able to achieve autonomous operations of low-cost unmanned vehicles, while maintaining the safety for people on the ground. Traditionally, the system that acts as a supervisor of such kind of issues is the Air Traffic Managers (ATM) but in our case we are working with different vehicles from traditional aircrafts, and a system able to cover the equivalent role of an ATM is required. UTM (UAS Traffic Management) is a structure able to support safe and efficient UAS operations following the characteristics of low-cost unmanned vehicles. The upcoming generation of mobile networks, 5G, and the improvements in the Cloud Robotics paradigm constitute a suitable framework to address the problem into account here. So, after studying the state of the art of the involved technologies, we propose and developed a networked control system (NCS) called CBUTM (Cloud-Based UTM) to accomplish the project’s goal. Developing a such kind of manager involves several category of knowledge, so the working group was divided into smaller ones, each one with different aspects to treat. My thesis, especially, will look to the research of the best path planning algorithm capable to be risk-awared and to execute suitable replanning. After deep examination of today’s state of art, taking to account our precise condition of risk-aware and dynamic map, two variations of the Rapidly-exploring Random Trees (RRT* and RRTX) have been designated to generate a safe flight minimizing our specific cost function. Going deeper, the proposed approach dwells on two distinct stages. At first, an off-line path planning is accomplished; it computes the optimal global path in a static environment without time constraints depending only on the risk-map, in which cells describe a specific location and have been associated with a risk-cost. Afterwards, there is an online path planning, in which, considering a dynamic risk-map, the off-line path is fixed and fitted to the new map, always taking to account to minimize the cost function. Ergo, a quick response constitutes a fundamental design parameter because the system needs to revise the pathway in very short time. The implementation is based on ROS (Robot Operating System), that played a key role in this project. Simulation results are still exposing, demostrating both the validity of our approach and some design limitations, especially in the online part, while supporting further developments.