The primary aim of mammalian behaviour simulation is to allow “behaviourologists” to extend their current knowledge without needing to resort to expensive and intrusive real life experimentation. A useful mechanism for realising mammalian behaviour simulation is provided by the idea of Multi-Agent Based Simulation (MABS) where each "player" in a simulation is represented by an agent with a particular set of features or capabilities. This thesis proposed the Mammalian Behaviour MABS (MBMABS) framework. The fundamental idea presented in this thesis is that each mammal featured in the simulation can be modelled as an agent that has a set of desires and a set of behaviours. The desires may be static, in that they do not change for the duration of a simulation, or dynamic in that they change with time during a simulation (influenced by some internal or external event). In the work presented behaviours are modelled using the concept of a behaviour graph comprised of vertices representing states and edges indicating possible state changes. State changes occur as a result of an agent completing some self-appointed task or as a result of some external event. Each state has one or more predefined potential follow on states. Where there is more than one follow on state selection is made according to a weighted random selection process. The weightings are derived dynamically according to individual agent’s desires. A particular novel element of the proposed approach is that it features a degree of randomness, agents will not behave in the same manner on each occasion that a simulation is run. The operation of the MBMABS framework is illustrated in this thesis using a collection of mouse behaviour case studies, in which real mice are represented as individual agents. The reported evaluation of the case studies demonstrated that the proposed framework readily supports rodent behaviour simulations. The reported evaluation also indicated that the proposed simulation framework readily allows users to observe the behaviour of the simulated entities. More specifically the evaluation of the simulations was conducted by: (i) comparing the operation of the proposed MBMABS with video data, (ii) visual observation and (iii) reference to domain experts. The MBMABS experiments conducted using video data successfully indicated that there was a similarity in the behaviour of mouse agents operating within the framework and real life mice (as recorded using video data). Mouse behaviour such as thigmotaxis and nest site selection was observed in both the simulation and video. The evaluation also indicated that the MBMABS framework readily supported the addition of states and desires. However, is was also noted that: (i) as the number of states increased the behaviour graph became more complex and difficult to visualise and (ii) as the number of agents interacting with the behaviour graph increased, the performance of the proposed framework was also affected in the sense that it required more resources to operate optimally.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:706933 |
Date | January 2016 |
Creators | Agiriga, E. I. |
Publisher | University of Liverpool |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://livrepository.liverpool.ac.uk/3004462/ |
Page generated in 0.0118 seconds