|dc.description.abstract||This thesis proposes a novel methodology for creating Artificial Agents with semi-realistic behaviour, with such behaviour defined as overcoming common limitations of mainstream behaviour systems; rapidly switching between actions, ignoring “obvious” event priorities, etc. Behaviour in these Agents is not fully realistic as some limitations remain; Agents have a “perfect” knowledge about the surrounding environment, and an inability to transfer knowledge to other Agents (no communication).
The novel methodology is achieved by hybridising existing Artificial Intelligence (AI) behaviour systems. In most artificial agents (Agents) behaviour is created using a single behaviour system, whereas this work combines several systems in a novel way to overcome the limitations of each. A further proposal is the separation of behavioural concerns into behaviour systems that are best suited to their needs, as well as describing a biologically inspired memory system that further aids in the production of semi-realistic behaviour.
Current behaviour systems are often inherently limited, and in this work it is shown that by combining systems that are complementary to each other, these limitations can be overcome without the need for a workaround. This work examines in detail Belief Desire Intention systems, as well as Finite State Machines and explores how these methodologies can complement each other when combined appropriately. By combining these systems together a hybrid system is proposed that is both fast to react and simple to maintain by separating behaviours into fast-reaction (instinctual) and slow-reaction (behavioural) behaviours, and assigning these to the most appropriate system. Computational intelligence learning techniques such as Artificial Neural Networks have been intentionally avoided, as these techniques commonly present their data in a “black box” system, whereas this work aims to make knowledge explicitly available to the user.
A biologically inspired memory system has further been proposed in order to generate additional behaviours in Artificial Agents, such as behaviour related to forgetfulness. This work explores how humans can quickly recall information while still being able to store millions of pieces of information, and how this can be achieved in an artificial system.||en