Abstract
Human intervention and activities have a?ected the marine and inland water-bodies ecosys-tems thereby leading to an increase in the occurrence of environmental issues such as algal blooms, oil spills, and acidi?cation. In order to study the human impact and understand the complex interplay between dynamic, physical, and biogeochemical processes in the aquatic environments, environmental scientists and oceanographers seek to access a wide array of mea-surements across a range of spatial and temporal scales: e.g. temperature, dissolved oxygen, pH, and biological data. However, there is a lack of e?cacious technology for the robust moni-toring and tracking of dynamic environmental features over the spatial scales of the order of few square kilometres, and with the necessary temporal resolution. Present monitoring techniques rely on either a single autonomous surface vehicle (ASV) or a ?xed network of sensors that are unable to provide high temporal resolution over small spatial scales (0.1 km2 –1km2). This dissertation addresses the gap in terms of both technology and science related to a solution for the pervasive and permanent monitoring of water quality with adaptive resolution in both space and time. We study and discuss an innovative solution for the dynamic monitoring of aquatic environments through the design of a ?eet of decentralised cooperative control ASVs that follow the swarm robotics paradigm. First, we study the system-level design principles and develop a range of cooperative control strategies. Second, a robust ?eet of ASVs is assembled based on the uncovered principles of dynamic collective behaviours. Finally, these cooperative control strategies are tested using the ?eet of ASVs by performing collective behaviours such as ?ocking, navigation and dynamic area coverage. The topology and dynamics of the network of interaction play a key role in the e?ectiveness and responsiveness of swarming behaviours. This research also studies the e?ect of the number of interacting agents on the responsiveness of the system by using two distinct modelling approaches: a distributed linear leader-follower consensus protocol and an agent-based self-propelled particles model. By studying the distinct distributed linear leader-follower consensus protocol and agent-based model, we uncover the important e?ect that the number of interacting agents has on the responsiveness of the collective. While increasing the number of interacting agents improves the system’s capacity to respond to slow changes, an excess of interactions can hinder the swiftness of its response to fast changes. Next, we investigate forced switching—which is the rewiring of the interaction network—with the same two models. For global forced switching, we uncover the existence of a trade-o? between speed to consensus and the collective responsive of the sys-tem. In the case of local forced switching, we reveal that introducing a certain amount of forced switching improves the responsiveness of the system while keeping the number of interacting agents low. Finally, we have successfully developed and tested a ?eet of 48 ASVs to perform a vast range of collective behaviours. This ?eet of ASVs is the largest distributed multi-robot system of its kind reported to date. This research demonstrates that this swarm robotics system is able to achieve scalable deployment and dynamic monitoring in fully unstructured environ-ments and in the absence of any supporting infrastructure. The robustness of the system is validated through the loss of multiple units while successfully performing di?erent collective behaviours during the ?eld experiments. A new metric is introduced to quantify coverage ef-fectiveness for dynamic area monitoring. In addition, this research demonstrates the possibility of water quality monitoring of the system by measuring the water surface temperature and performing real-time temperature ?eld reconstruction. In conclusion, this research demonstrates that our system is capable of performing adaptive deployment, in situ measurement of water quality in the spatial scale of 1 km2,anddynamic area coverage where the changes in area are in the order of one minute. Thus, this system provides an attractive and a?ordable solution to the complex and daunting problem of pervasive monitoring of aquatic environments. The study of interagent interaction shows that it has far-reaching implications for the design of the interaction network and provides insights for the design of arti?cial swarming systems.