An Evolutionary Algorithm for Quadcopter Trajectory Optimization in Aerial Challenges

Machine learning methods have been widely employed in robotics over the years, and recent developments in machine learning have completely re-shaped problem-solving in the area. Indeed, if we consider multi-objective planning, these models' optimization and learning capabilities can derive more robust strategies. Inspired by the species natural selection mechanism, Evolutionary Algorithms (EA) are among the best known computational approaches available for this purpose. In this scenario, this work proposed an EA model developed to find the best travel trajectory for a quadcopter in the “Desafio Petrobras” challenge. In the challenge, a set of landing platforms that the robot has to visit are displaced in the 3D-space. To find the best trajectory possible, we optimize an EA over a low-level control that can take the quadcopter from point A to B. We vary our fitness function to support more complex decisions. The software-in-the-loop technique was applied for a simulated quadrotor in the Coppelia simulated environment. The proposed approach has shown the capability to generate short trajectories while considering variables like UAV dynamics and energy consumption.