Biorobotic investigation of multimodal cue integration for insect orientation

Abstract

Insects are capable of great feats of navigation; these range from relatively simple straight-line orientation over a few metres to migration over hundreds of kilometres. A robust compass is therefore critical, and to this end, an animal may use multiple orientation cues simultaneously (a behaviour known as cue integration). Recent work has indicated that ball rolling dung beetles perform cue integration during straightline orientation. This thesis explores dung beetle cue integration using a biorobotic approach, with the goal of understanding how this process could be implemented in the insect brain.

We start by adapting the standard Bayesian approach from human psychophysics to compare mathematical models of cue integration behaviour, assuming that dung beetles weight cues according to their reliability (the inverse of their variance). This modelling indicates that beetle cue integration is likely represented by a vector sum which: (1) aligns well with a candidate neural substrate, and (2) can generate a variety of behaviours depending on how the cue weights are set (providing behavioural flexibility despite functionally conserved neuroanatomy).

It has previously been proposed that orientation cues are integrated in a set of neurons known as EPGs which receive input from ‘R’ neurons via highly plastic connections. Recent evidence has shown that different sets of R neurons encode different cue modalities. Using a computational model, we argue that R neurons could encode vectors as sinusoidal activity across their population and the resultant input to the EPGs represents a vector sum. The model can account for the vector sum, and provides a possible mechanistic explanation for the dung beetle orientation snapshot. Further, the work suggests that a number of factors may contribute to total cue influence; specifically, cue weight, cue reliability, and synaptic strength (which is affected by weight and reliability independently). Moreover, we show that reliability does not seem to be the major factor which determines cue influence in dung beetles (using a combination of modelling and animal behaviour). This result stands in contrast to the vast majority of literature on cue integration, including all previous accounts in insects which discuss concrete models, and prompts a critical re-examination of the explanatory power of the Bayesian approach to cue integration.