| dc.description.abstract | Robotic collectives are large decentralized robot groups of more than fifty individuals that coordinate using interactions inspired by social insect behaviors. Collectives make group decisions that are facilitated by information pooling within a shared decision space, similar to an insect colony. Biological collectives must frequently choose the best option from a finite set of options and execute an action based on that choice. Discrete collective consensus achievement algorithms have enabled robotic collectives to make decisions, but most research does not consider scenarios in which the collective must ignore biasing features within the environment. Whether the collective decides between occupation sites, routes, or future actions, the environmental features (e.g., distance between a resource and the collective) alter robot interactions and bias collective decisions towards options that are the easiest to find, evaluate, and reach, but may not be the optimal choice. Robotic collectives that must ignore biasing environmental features during decision making are likely to be inaccurate and inefficient. Robotic collectives do not have centralized control; thus, they are challenged to synchronize the initiation and execution of the chosen actions, which is critical to future collectives that must respond to the environment and complete complex tasks. Discrete collective consensus achievement strategies have only recently been considered in the field of Human-Swarm Interaction, which has largely focused on enabling humans to control artificial swarms. Typically, swarms are comprised of agents that interact according to a protocol that causes a desired emergent behavior, such as flocking. Most Human-Swarm Interaction research assumes the human has near-perfect knowledge of the swarm and global communication with the swarm's agents. Robotic collectives have the potential to share decision making functions with humans; however, methods of human interaction with collective discrete consensus strategies have not been designed or evaluated.
This dissertation develops a new algorithm influenced by biologically inspired discrete consensus achievement strategies in order to enable robotic collectives to choose and implement the best actions, despite the presence of environmental bias. The new model enables future human-collective teams to make decisions when the human does not have perfect global knowledge. Further, human-collective interaction mechanisms are developed in order to facilitate collaborative decisions between a human and a simulated robotic collective. The robotic collective model is implemented and evaluated for its ability to act independently and as a part of a human-collective team in trials featuring the human supervision of multiple targeting collectives. | |
