Files
Abstract
In nature, many animal species developed social communication and heterogeneous relationships and demonstrated emergence as an evolutionary trait over millions of years. While these traits are ubiquitous in natural systems, it is imperative to extend the findings to artificial Heterogeneous Multi-Agent Systems (HMAS) to develop domain-specific design solutions that can demonstrate emergence. For a robust emergent HMAS design, determining the suitable group composition and ensuring effective knowledge transfer is paramount.It is motivating to find these complex HMAS behaviors in sub-cellular processes, which inspired the current work on investigating emergence, understanding the need for heterogeneity, and establishing knowledge transfer processes.
This dissertation research exploits the emergence, heterogeneity, and knowledge transfer in HMAS. Specifically, the research objectives include: (i) Developing a simulation framework for designing emergence in complex HMAS; (ii) Investigating the impact of knowledge heterogeneity, and (iii) Developing knowledge transfer strategies by exploiting heterogeneity.
In the first contribution, a new approach for simulating self-organization in microtubules was developed. Here, the proteins are defined as agents with specific geometries and intelligence encoded in behavior trees (BT). The simulations built from this approach successfully demonstrated growth, shrinkage, dynamic instability, and severance by Katanin, assimilating the in-situ emergent behaviors of microtubules.
The second contribution investigated the effect of heterogeneity in a robot search and rescue (SAR) simulation, in which robots varied in capabilities and knowledge among heterogeneous groups. The simulations showed that the heterogeneity measure positively correlated with rescue efficiency, demonstrating the benefits of heterogeneity. However, some cases contradict this, pointing to the need for determining optimal group composition by prioritizing the agents' functional overlap.
Finally, in the third contribution, we developed two frameworks for direct (KT-BT) and indirect (IKT-BT) knowledge transfers that utilize BTs for knowledge representation and sharing. The KT-BT framework follows a direct query-response-update process, and IKT-BT follows an indirect eavesdropping mechanism. When tested on SAR simulations, both strategies efficiently improved knowledge spread, mission performance, and balanced heterogeneity compared to no transfer, even with additional memory constraints imposed.
Overall, the above contributions and the developed frameworks significantly advance the research on emergence, heterogeneity, and knowledge transfer in HMAS.
This dissertation research exploits the emergence, heterogeneity, and knowledge transfer in HMAS. Specifically, the research objectives include: (i) Developing a simulation framework for designing emergence in complex HMAS; (ii) Investigating the impact of knowledge heterogeneity, and (iii) Developing knowledge transfer strategies by exploiting heterogeneity.
In the first contribution, a new approach for simulating self-organization in microtubules was developed. Here, the proteins are defined as agents with specific geometries and intelligence encoded in behavior trees (BT). The simulations built from this approach successfully demonstrated growth, shrinkage, dynamic instability, and severance by Katanin, assimilating the in-situ emergent behaviors of microtubules.
The second contribution investigated the effect of heterogeneity in a robot search and rescue (SAR) simulation, in which robots varied in capabilities and knowledge among heterogeneous groups. The simulations showed that the heterogeneity measure positively correlated with rescue efficiency, demonstrating the benefits of heterogeneity. However, some cases contradict this, pointing to the need for determining optimal group composition by prioritizing the agents' functional overlap.
Finally, in the third contribution, we developed two frameworks for direct (KT-BT) and indirect (IKT-BT) knowledge transfers that utilize BTs for knowledge representation and sharing. The KT-BT framework follows a direct query-response-update process, and IKT-BT follows an indirect eavesdropping mechanism. When tested on SAR simulations, both strategies efficiently improved knowledge spread, mission performance, and balanced heterogeneity compared to no transfer, even with additional memory constraints imposed.
Overall, the above contributions and the developed frameworks significantly advance the research on emergence, heterogeneity, and knowledge transfer in HMAS.