Dottorato in INGEGNERIA INFORMATICA

Titolo della tesi: Monitoring, Mapping and Exploitation by Self Organizing Robot Swarms

Swarm robotics, is a peculiar approach to the study of multi-robot systems that emphasizes self-organization and decentralized control. Swarm robotics assumes limited individual abilities, local sensing, and local communication, and provides design methods to ensure flexibility, robustness, and scalability of the robotic systems. This is achieved thanks to well-designed collective behaviors relying on coordination and collaboration among multiple robotic units. In this thesis, we draw inspiration from swarm robotics practices and focus on the exploration and exploitation of points of interest scattered in an unknown environment. With the term ``exploitation'', we identify processes whose role is to exploit the resources of an area, which may be either tangible assets---e.g., plants to be mapped---or abstract utilities---e.g., expected information about a process to be monitored. We start by introducing two approaches for collective monitoring and mapping. Then, we propose a bio-inspired algorithm reproducing the nest selection process of honeybees and, later, extend it to the exploitation task. We apply the proposed algorithm to several study cases and analyze it both with static and dynamic scenarios. The first and principal application that we analyze is related to the field of precision agriculture while, other applications such as service robots and long term exploitation, are considered later. Within the precision agriculture domain, we focus on mapping the presence of weeds in a cultivated field by means of a group of autonomous flying drones. Initially, we present a decentralized solution for monitoring the field by means of a reinforced random walk with inhibition of return, where the information available to all the agents is used to bias the individual decision about the next area to visit. Then, we improve the decision process accounting for the level of information of an area and the uncertainty reduction, represented respectively by the information entropy and the information gain. The latter provides a representation of the degree of additional information associated with each possible observation. The approach, based on information gain, is independent from many parameters of the system, resulting in a solution suitable for direct implementation without a preliminary configuration step. Next, we extend the above studies and propose an adaptive strategy for wide-area monitoring. In many applications, high-resolution data are required only in certain areas while others can receive lower attention, making non-uniform coverage strategies efficient both in time and energy expenditure. To enable non-uniform coverage for robot swarms, we build on top of the previously presented reinforced random walk and design a decentralized dynamic strategy to assign parts of the group only to those regions where high-resolution data is required, leaving the rest of the swarm to the exploration task. To this end, we present two different algorithms based on the above model and respectively apply these approaches to a precision agriculture scenario and to a service robotics scenario. Last, we abandon the slow environmental dynamics typical of precision agriculture and move to environments characterized by dynamically-changing utilities. Amongst this, we study the dynamics of a self-regenerating environment subject to external exploitation. We introduce a polynomial function to generalize the effects of the agents on the environment, while we model the regeneration of the resources thanks to the logistic growth function. We study how the proposed approach reacts to renewable resources and cope with the concept of respectful exploitation---i.e., avoid complete depletion of the resources so as to allow them to regenerate, granting new possibilities of exploitation in the future.