HARDWARE IMPLEMENTATION OF ARTIFICIAL EPIGENETIC NETWORKS

An extension of Artificial Gene Regulatory Networks (AGRNs), Artificial Epigenetic Networks (AENs) implement an additional layer of bio-inspired control to allow for enhanced performance on certain types of control tasks by facilitating topological self-modification. This work looks to expand the applications of AENs by translating the existent software architecture into a form suitable for implementation on a Field Programmable Gate Array (FPGA). This opens the possibility of AENs being used in applications where high-performance computational resources are impractical, such as robotic control. This thesis develops a more resource efficient architecture for epigenetic networks based on reduced precision integer mathematics, and then translates it into hardware to provide improvements in resource utilisation and execution speed while not sacrificing the unique benefits provided by the epigenetic mechanisms. The application to robotic control is investigated by utilising the hardware AEN to perform various versions of a foraging task, culminating in one designed to replicate a search and rescue scenario. While the AENs did not demonstrate significant performance improvements compared to their non-epigenetic counterparts, this did indicate that not every type of control task benefits from the inclusion of the epigenetic mechanism. In addition, this work investigates another aspect of AENs, specifically the limits of their topological self-modification with respect to reacting to changes in their environment. More specifically, it is asked if an AEN can maintain its ability to perform a specific task when confronted with factors outside of those it has been optimised to handle. While not conclusively demonstrated, there is sufficient evidence that the answer to this question depends on the performance gains imparted by epigenetic behaviors under normal circumstances