Self-Deployment Strategy for a Swarm of Robots with Global Network Preservation to Assist Rescuers in Hazardous Environments Pham Duy Hung, Minh-Trien Pham, Tran Quang Vinh Faculty of Electronics and Telecommunication University of Engineering and Technology Hanoi, Vietnam hungpd@vnu.edu.vn Trung Dung Ngo Universiti Brunei Darussalam Faculty of Science The More Than One Robotics Laboratory ngo.trungdung@ubd.edu.bn Abstract—In this paper, we present a self-deployment strat- egy for a swarm of robots that is capable of exploring and identifying victims in an unknown structured building envi- ronment while preserving a global network interconnectivity for information exchange. The strategy are conducted in two phases: self-displacement shifting the robotic swarm from room to room, and self-dispersion and aggregation for exploration and coverage in each room. A decentralised control is built up by decentralised node control governing the dispersion and aggregation and decentralised connectivity control guaranteeing the global network preservation. The self-deployment strategy reduces significantly number of the robots while increasing its capacity of exploration and coverage. The simulation results illustrate technical aspects of the robotic swarm with the application of exploration, search and rescue services. Index Terms— Self-Deployment, Self-Displacement, Self- Dispersion, Swarm of Robots, Decentralised Node Control, De- centralised Connectivity Control, Network Preservation, Rescue Supporting. I. I NTRODUCTION Swarm robotics has been received a lot of attentions due to its challenges and potential applications. Mobile robotics in a swarm are able to interact and cooperate in order to complete many tasks in parallel, which might be quicker, more flexible, and more reliable than a single robot. A swarm of mobile robots can be utilised in hazardous environments for explo- ration and rescue services, surveillance and reconnaissance, and patrolling and monitoring. Deployment strategy for exploration, search and rescue services in large-scale unknown environments is one of the most challenges of swarm robotics. A number of existing deployment algorithms is classified into three main streams: Voronoi graph based unifying geometric method in [1], probabilistic model based controllers in [2], and artificial potential field based controller in [3], [4], [5], [6]. In [1], a Voronoi-based controller was developed to control the robots to move towards centroids of Voronoi cell for coverage maximisation. The geometric strategy has been extended with a limited sensing radius in [7], heterogeneous multi-robot systems in [8], [9], and the adaptive coverage control algorithm in [10]. In the probabilistic approach, in [2] an control algorithm is proposed to navigate the robots to maximise the probability of detecting an event occurring in the environment. Artificial potential field which was originally coined out by Khatib (1986) for single robot navigation with obstacles in the environment. This method was widely used for deploy- ment strategies for multi-robot systems such as deployment of mobile sensor networks in [3], flocking and herding in [5], [6], and consensus in [4]. The deployment strategy was introduced in [3] for an un- known environment where robots automatically disperse out to cover all areas in the environment. However, the artificial potential field based decentralized control is scalable but does not guarantee the global robot network preservation during the deployment process, resulted in collected environmental information not sent out to the human operators. Moreover, this deployment strategy does not control mobility of the robots and their connectivities specifically leading to diffi- culty of governing the robots for specific applications such as exploration and rescues in unknown structured environments. In order to cover whole environment instantly, most of above approaches assume the number of robots is unlimited. However, with the limited number of robots in swarm, a new strategy is proposed to cover whole environment instantly by using backbone network established by anchor robots to create an interconnection between a base station at main entrance and deployed robots at rooms. At the first stage, the swarm shifts from room to room. Then at the second stage, the swarm will be dispersed to coverage the room and identify victims. After that, the swarm will shift to another room as in the first stage. The process will be repeated until the final room is covered. During process, the global network integrity is maintained for exchanging information. As the results, the number of robots is significance reduced compared to strategy in [3]. The rest of this paper is organised as follows: fundamental knowledge of the graph modelled robotic swarm and local connectivity topologies used to develop the decentralised control are described in section II. The decentralised control as hierarchically structured control is presented in section III. The self-deployment strategy are elaborated in section IV. The simulations and discussions are addressed in section V.