| Popis: |
Recent swarms of Uncrewed Systems (UxS) require substantial human input to support their operation. The little 'intelligence' on these platforms limits their potential value and increases their overall cost. Artificial Intelligence (AI) solutions are needed to allow a single human to guide swarms of larger sizes. Shepherding is a bio-inspired swarm guidance approach with one or a few sheepdogs guiding a larger number of sheep. By designing AI-agents playing the role of sheepdogs, humans can guide the swarm by using these AI agents in the same manner that a farmer uses biological sheepdogs to muster sheep. A context-aware AI-sheepdog offers human operators a smarter command and control system. It overcomes the current limiting assumption in the literature of swarm homogeneity to manage heterogeneous swarms and allows the AI agents to better team with human operators. This thesis aims to demonstrate the use of an ontology-guided architecture to deliver enhanced contextual awareness for swarm control agents. The proposed architecture increases the contextual awareness of AI-sheepdogs to improve swarm guidance and control, enabling individual and collective UxS to characterise and respond to ambiguous swarm behavioural patterns. The architecture, associated methods, and algorithms advance the swarm literature by allowing improved contextual awareness to guide heterogeneous swarms. Metrics and methods are developed to identify the sources of influence in the swarm, recognise and discriminate the behavioural traits of heterogeneous influencing agents, and design AI algorithms to recognise activities and behaviours. The proposed contributions will enable the next generation of UxS with higher levels of autonomy to generate more effective Human-Swarm Teams (HSTs). |
