Extending the Evolutionary Robotics approach to flying machines: an application to MAV teams

The work presented in this article focuses on the use of embodied neural networks--developed through Evolutionary Robotics and Multi-Agent Systems methodologies--as autonomous distributed controllers for Micro-unmanned Aerial Vehicle (MAV) teams. The main aim of the research is to extend the range of domains that could be successfully tackled by the Evolutionary Robotics approach. The flying robots realm is an area that has not been yet thoroughly investigated by this discipline. This is due to the lack of an affordable and reliable robotic platform to use for carrying out experiments, and to the difficulty and the high computational load involved in experiments based upon a realistic software simulator for aircraft. We believe that the most recent improvements to the state of the art now permit the investigation of this domain. For demonstrating this point, two different evolutionary computer simulation models are presented in this article. The first model, which uses a simplified 2D test environment, has resulted in controllers evolved with the following capabilities: (1) navigation through unknown environments, (2) obstacle-avoidance, (3) tracking of a movable target, and (4) execution of cooperative and coordinated behaviors based on implicit communication strategies. In order to improve the robustness of these results and their potential use in real MAV teams, a more sophisticated 3D model is presented herein. The results obtained so far using the two models demonstrate the feasibility of the chosen approach for further research on the design of autonomous controllers for MAVs.