In case of a catastrophic event, when ordinary communication infrastructures are out of service or simply not available, it is crucial to create a wireless communication network for rescuers as soon as possible. Our Challenge is to develop self-organizing swarms of autonomous Micro Air Vehicles (MAVs) that provide a fast-deployable and self-managed ad-hoc Wi-Fi network to connect and coordinate rescue teams on the ground. Furthermore the swarm can explore the disaster area in order to localize victims and direct the rescues.
Our objective is to develop novel swarming and networking strategies for self-organizing robot networks and to implement a testbed for demonstrating them with real-world experiments. The testbed is based on eBee micro air vehicles developed by SenseFly and on mini ARM-based computers by Gumstix Inc. The airplanes are fixed-wing aircrafts with an electric motor and integrated autopilot capable of flying with winds of up to 12 m/s, at a cruising speed of up to 15.8 m/s, with an autonomy of up to 45 minutes. In case of emergency, they can be remotely controlled up to a distance of 3 km via a Microhard Systems Nano n2420 link connection. Within this distance, if necessary, the flight mission can be modified on the fly. The autopilot has access to an inertial measurement unit, a barometer, a pitot-tube for airspeed, an optical-flow sensor and GPS receiver. Each eBee also carries a Gumstix Overo Tide computer with a custom embedded Linux distribution and a standard USB WiFi (802.11n) card. We use this embedded computer to establish the wireless network. A serial connection between the auto-pilot module and the embedded computer allows us to access the sensors attached to the autopilot (including the GPS reading) and to give commands to the autopilot, e.g., modify the aircraft mission according to routing needs. Thanks to its small dimensions and weight (under 630 g), flying eBees are not considered a threat. In some countries (e.g., Switzerland) they can be used without specific authorization.
Our research plan is twofold: On the one side, we are examining mobile ad-hoc networks (MANET) composed by MAVs. Previous research works considered star topology networks for micro air vehicles (MAVs) or unmanned aerial vehicles, (UAVs). Star network topology restricts the communication coverage and then the searching area of the swarm, because robots cannot fly out of the control center coverage area. We focus on partially-connected mesh ad-hoc networks, that enable the robots to use multi-hop communication to extend the operation area. Due to the high-mobility of the nodes, these networks are very dynamic and the existing routing protocols fail to provide a reliable communication. For this reason, we presented Predictive-OLSR an extension to the Optimized Link-State Routing (OLSR) protocol which provides reliable communication even in case of very dynamic networks. On the other side, we are examining how to apply swarm intelligence, and in particular how to cope with dynamic swarm size and topology, to collectively explore the area to localize the users. The originality of our approach lies on the local behavior adaptation of robots to improve network communication. This is a reactive exploration approach that dynamically adapts itself to the changes of the environment, whereas predefined approaches fail. Beside exploration, we also address network communication maintenance with mobile users on the ground. In order to accomplish this difficult task we study cooperative strategies between robots, that would not be able to solve the problem independently. In order to have realistic scenario we take into account the significant constraints such as energy and platform's dynamics limitations. As flying robots have limited flight time, we need to force some robots to return to change their batteries. Thus, we aim to apply algorithms that allow for removing robots and adding new ones to the network, to increase the total mission duration. Previous research has demonstrated strategies for either exploration, or communication maintenance in mobile networks, but nobody has addressed these two topics at the same time. Furthermore, algorithms are usually assessed by simulation, where robots are depicted as point masses with omni-directional motion, and do not consider constraints such as the speed of the wind, local communication, and restricted flight time. Whereas, we are aiming at real-world experiments to test the new swarming strategies. We use fixed-wing robots that need to maintain forward speed and follow unicycle trajectories. Therefore we cannot assume omni-directional motion, but we have to address non-holonomic system constraints. In the last few years, several scientific papers have proposed centralized control strategies for swarming. They assume that all the robots are connected to a center server that has the full control of the robots' actions. Other works address hierarchical control strategies, which consider leader robots that are responsible for task allocations among the rest of the robots.We focus on decentralized control strategies in which all the robots are identical and they decide independently and autonomously, based on the communication information exchange with their neighbors. These systems give us the potential to design a system which has a low infrastructure, very short deployment time, and lower cost compared to infrastructure-based systems.
For further details please see the paper: "Speed-aware routing for uav ad-hoc networks".
Predictive-OLSR patch is an open source software. Source files are available in the Download area.
By providing immediately broadband wireless access anywhere, and by monitoring large areas which maybe unreachable otherwise, a flying robot network can play a crucial role during emergency situations. The benefits that make flying robots networks a promising solution are:
This work is supported by armasuisse, competence sector. Science+Technology for the Swiss Federal Department of Defense, Civil Protection and Sport, under the auspices of SMAVNET II Project. We would also like to thank SenseFly for providing the robotic platforms and support.
Contact us via email anytime. It's the best way to reach us!
Email Me at: stefano [dot] rosati [at] epfl [dot] ch
Our office address is:
EPFL IC LCM
Building INR, Station 14
1015 Lausanne, Switzerland
Phone: +41(0) 216937552
Fax: +41(0) 216931200