Learning-in-Games Approach for the Mission Planning of Autonomous Multi-Drone Spatio-Temporal Sensing

Abstract

Mission planners are one of the major classes of autonomy software and their design is especially challenging in the case cooperation autonomy is required for unmanned multi-vehicle systems. A clear example of this is given by the applications of teams of drones, such as multi-drone spatio-temporal sensing. Here, drone teams act as mobile and cooperative sensor networks to simultaneously collect sensor data in areas of interest and to allow detailed computation on the sensed data. For the design of cooperative and autonomous drone teams, mission planning shall be accomplished in the form of coordinated sensing to optimally assign the different sensing tasks and routes to each drone, employing task allocation and route planning as the basic pillars to maximize the multi-drone mission effectiveness. This work proposes a dynamic and decentralized mission planner for a drone team performing autonomous and cooperative spatio-temporal sensing. The design exploits the learning-in-games framework for the processing of optimal routes in reasonable time frames. Two ad-hoc variants of the binary log-linear learning are proposed as a coordination algorithm to manage both task allocation and route planning, by demonstrating reachability and reversibility properties. Also, the work describes an experimental analysis of the proposed solutions by means of model-in-the-loop simulations, in order to provide a preliminary tune of the main learning parameters for both solutions.

Keywords

• Multi-drone systems
• multi-drone sensing
• learning in games
• multi-agent systems