Evolutionary Optimization of Drone Trajectories Based on Optimal Reciprocal Collision Avoidance

• Published in: 2020 27th Conference of Open Innovations Association (FRUCT)
• Main Authors: Alex Bojeri, Giovanni Iacca
• Format: Article in Journal/Newspaper
• Language: English
• Published: FRUCT 2020
• Subjects: artificial intelligence; autonomous systems; evolutionary computation; autonomous agents; unmanned autonomous vehicles; Telecommunication; TK5101-6720; Orca
• Online Access: https://doi.org/10.23919/FRUCT49677.2020.9211037; https://doaj.org/article/c0a538f8b6dd4390a9f2be37864453c5

In recent years, the advent of new hardware and software technologies for navigation and control has made Unmanned Aerial Vehicles (UAVs) ever more autonomous and efficient. As a consequence, it is now possible to have drones moving within complex environments, such as cities or indoor areas. One of the main requirements for intelligent mission planning in such environments is the ability to correctly and efficiently detect and avoid obstacles. For this reason, various libraries have been created for the simulation of UAV navigation in virtual environments, in order to test algorithms for automatic obstacle detection and collision avoidance before deploying the drones in the real world. Usually, the performance of these algorithms depends on various parameters as well as specific application settings. However, while different parameter configurations can be easily tested in simulation, their number can be too large to allow a complete exploration of the parameter space or a manual tuning. Furthermore, a full analytical model of the parameters' influence on the algorithmic performance can be hard to obtain. Yet, it is extremely important to find their optimal values to allow collision-free navigation. In this direction, we propose here a thorough exploration, based on an Evolutionary Algorithm (EA), of the parameter space of the Optimal Reciprocal Collision Avoidance (ORCA) algorithm. Our results show that the proposed EA is a viable solution for finding optimal parameter settings that can be generalizable to different scenarios characterized by different complexity levels.