Micro aerial vehicle path planning and flight with a multi-objective genetic algorithm

Abstract

Due to its importance for robotics applications, robotic path planning has been extensively studied. Because optimal solutions can be computationally expensive, the need for good approximate solutions to such problems has led to the use of many techniques, including genetic algorithms. This paper proposes a genetic algorithm for offline path planning in a static but very general, continuous real-world environment that includes intermediate targets in addition to the final destination. The algorithm presented is distinct from others in several ways. First, it does not use crossover as this operator does not appear, in testing, to aid in efficiently finding a solution for most of the problem instances considered. Second, it uses mass extinction due to experimental evidence demonstrating its potential effectiveness for the path planning problem. Finally, the algorithm was designed for, and has been tested on, a physical micro aerial vehicle. It runs on a single-board computer mounted on the MAV, making the vehicle fully autonomous and demonstrating the viability of such a system in practice.