Project Summary

Our vision is that small, simple low-power computational devices could be given non-trivial adaptive and learning capacities. We are inspired by animals with miniature brains, specifically the Drosophila larva with c.10,000 neurons, that exhibit rapid autonomous acquisition of associations between relevant stimuli in a complex and variable environment. These ‘simple’ animals can learn arbitrary associations of stimuli with punishment and reward, and behave in an anticipatory fashion in their decisions to approach or avoid learnt stimuli in different contexts. Translated into robotics, we imagine distributed devices that individually detect important regularities between sensory events, and are thus able to act adaptively to exploit natural signals in scenarios such as tracking, search, escape and
exploration. Embedded devices in many different technologies could optimise to anticipate subtle but predictable contingencies. Additional desirable features apparent in animal learning are the ability to generalise or discriminate in appropriate circumstances, to distinguish relevant cues from irrelevant background signals, and to maintain representations of stimuli features needed to guide action (such as signal intensity) while making decisions (e.g. regarding signal identity) that are invariant to such features.
The required breakthrough is to obtain a multilevel understanding of the mechanisms in a miniature brain by which learning alters behaviour, in the form of a one-to-one neural circuit model linked to control of behaviour in the context of complex sensory inputs. This is uniquely possible in our target system of the Drosophila larva due to the intersection of new and powerful methods: in neuroscience for circuit identification and control in behaving animals; and in ICT for integration, analysis and modelling of these circuits in the context of closed loop control of real devices. As a consequence, we believe a small project on a small brain will have disproportional impact for brain and computer sciences.

Project Objectives >>

Leave a Reply

Your email address will not be published. Required fields are marked *