Research

Overview

I am interested in using artificial evolution to create models of animal behaviour. This is a relatively new and unexplored approach to the study of behaviour, and shows great promise. My work on path integration shows how a genetic algorithm can produce a simple and efficient solution to a relatively complex problem, discovering and exploiting mechanisms that are, in part, quite different from those found in the existing path integration modelling literature.

My research has been influenced by the dynamical approach to cognition, particular by the work of Randall Beer. My methodology can be seen as an incremental complexification of his own, with the aim of tackling perhaps more complex, less abstract behaviours and with relatively less emphasis on retaining a neurophysiologically realistic model of neuron dynamics.

Path Integration

My thesis focussed on modelling the animal navigation behaviour known as path integration (PI). This is a method relying only on a compass and odometer (meaning that the animal must have a sense of its direction and speed as it travels), and as such can function without the need for landmarks in the environment. The Saharan desert ant Cataglyphis fortis has been well studied for its PI ability.

During my DPhil, which I did at the Centre for Computational Neuroscience and Robotics at the University of Sussex, I produced a mathematical/computer model of path integration in Cataglyphis fortis using artificial evolution. The computer discovered how to navigate using a modified form of Horst Mittelstaedt's bicomponent model of PI. The evolved model not only returns home, but also searches for the nest when it reaches the point it 'expects' it to be. Searching is generated using a remarkably simple addition to the Mittelstaedt homing system and yet creates a complicated searching behaviour, which never falls into an inefficient repeating pattern. The model can also easy be made to reproduce the systematic navigation errors seen in Cataglyphis, due to its use of leaky integrators in its 'positional memory'.

Salticid Vision

My current research, working together with Duane Harland, is being carried out at the University of Canterbury, New Zealand, in association with AgResearch Limited, and is aimed at achieving a better understanding of visual perception in the Salticid (jumping spider) Portia fimbriata.

The project will proceed by developing a special opthalmoscope for the spider, in order to track its gaze patterns as it is shown images of other invertebrates. The aim is then to produce evolved neural networks capable of reproducing the spider's eye movements and object discrimination abilities. This should provide an excellent opportunity to further develop the techiques developed during my thesis.

References

Mittelstaedt, H. and Mittelstaedt, M.-L. (1973). Mechanismen der orientierung ohne richtende aussenreize. Forschr. Zool., 21:46-58.

Mittelstaedt, H. (1983). The role of multimodal convergence in homing by path integration. Fortschr. Zool., 28:197-212.

Mittelstaedt, H. (1985). Neurobiology of Arachnids, chapter Analytical cybernetics of spider navigation, pages 298-310. Springer-Verlag, Berlin. F. Barth ed.

back