Dynamics-Aware Fast Multi-Drone Exploration of Unknown Environments

Tasks such as surveillance, search and rescue operations, or disaster mapping require exploring unknown environments in the shortest possible time. Furthermore, when these tasks are performed by multiple robots, team coordination is crucial to achieve the desired efficiency. In this spirit, this paper proposes a method that introduces notions about each robot's dynamics into the planning process. We formulate novel costs into the Vehicle Routing Problem (VRP) to maintain each drone's inertia, required for efficiency. The costs of exploring regions are calculated based on the individual state of each robot, such as their speed and orientation, to favour distinct behaviours. The proposed solution is implemented on RACER, a state-of-the-art decentralized exploration system. The original implementation of RACER is modified to integrate our method, including a different optimization library to solve the VRP. To validate our approach, we conduct exhaustive experiments in four maps, with two and four agents and varying initial conditions, comparing our proposal with the state-of-the-art method. The results illustrate the advantages of incorporating knowledge about the team's dynamics into the planning process, demonstrating improvements in the efficiency and effectiveness of exploration in unknown environments.