The aim of this project is to investigate and design an autonomous swarm of Micro Air Vehicles (MAVs)to do multi-robot exploration of an unstructured indoor environment. These so called pocket drones are small quadroters with a mass in the order of 20 grams and a diameter of 10 cm. It can maneuve ritself through small corridors, windows and can reach different levels of a building. A swarm of these pocket drones is ideal for the fast exploration of a building during a search and rescue mission for instance.
However, these pocket drones have strict limitations on their on-board energy, sensing and processing capabilities. The challenge is to combine the needed functionalities, in terms of obstacle avoidance, exploration and coordination with the other drones. Inspiration can be drawn from honeybees, since an individual bee does not have many capabilities just by itself, but a swarm of them is able explore an entire field of flowers. This inspiration will help to develop efficient algorithms for multi-robot exploration with the pocket drones.
There are aspects and challenges to be considered for the design of a swarm of pocket drones. Low level navigation, which stand for simple behaviors as drift stabilization and obstacle avoidance, is essential to make the MAV fly. This will be executed with an on-board stereo vision system. The emphasis here is on efficiency, as the pocket drone need to do so much more than only these rudimentary tasks. The pocket drone also needs to know its environment and should decide where to explore next, which are more high level navigations tasks. Here the challenge is to design efficient localization and mapping algorithms for limited on-board processing of the pocket drone based on its vision system.
The last challenge is the multi-robot coordination, which enhances the pocket drones performance by enhancing their individual abilities as a swarm. The challenge is to enable the drones to share the generated map and locations and decision making in task coordination with limited communication and sensing capabilities. Since no central computer is used, the focus is on decentralized optimal control as the drones must make decisions themselves to benefit the overall search mission.
Funding: NWO Artificial Intelligence Applications