Human brains and vision-based robot­ics require inten­sive compu­ta­tion to recog­nize visual pattern features in various contexts and augmen­ta­tions, known as invari­ant pattern recog­ni­tion. The humming­bird hawkmoth (Macroglos­sum stellatarum) similarly uses pattern features on flowers to select suitable forag­ing sites, with only a fraction of the ‘compu­ta­tional power’. Aiming to under­stand how they do so with such efficiency, we will use behav­ioural, neural, and compu­ta­tional methods to uncover the algorith­mic basis of (invari­ant) pattern recog­ni­tion in insect pollinators.

Active sensing, or using movement to actively acquire sensory input from the environ­ment, is achieved through diverse mecha­nisms across species. In the case of vision, humans can use eye movements, and robots can manoeu­vre cameras to actively collect and process infor­ma­tion such as pattern features. Insects, however, tend to be much more restricted by morphol­ogy in how they can rotate or bend their visual sensors, if at all.

Insect polli­na­tors are partic­u­larly inter­est­ing in this regard, as they need to recog­nize and respond to the visual patterns of flowers when decid­ing where to forage. To overcome their morpho­log­i­cal restric­tion, the humming­bird hawkmoth (Macroglos­sum stellatarum) steers its’ body while hover­ing over flowers, gather­ing infor­ma­tion about their pattern features. Recog­niz­ing such features in varying condi­tions and contexts can require millions of neurons in humans. Limited compu­ta­tional power in hawkmoths suggest they may overcome this challenge through active pattern recog­ni­tion – gather­ing visual infor­ma­tion using a highly special­ized senso­ri­mo­tor system.

This PhD project attempts to uncover the algorith­mic basis of pattern recog­ni­tion in insect polli­na­tors through three approaches: Using behav­ioural and psychophys­i­cal methods to analyze how hawkmoths guide their flight when inspect­ing differ­ent pattern features; Neuro­phys­i­o­log­i­cal techniques to under­stand how varia­tion in pattern proper­ties (e.g. contrast or orien­ta­tion) affect neural compu­ta­tions for visuo­mo­tor circuits; Develop compu­ta­tional models that mimic both active sensing strate­gies in the hawkmoth, as well as those used by the visual system for pattern recog­ni­tion, with poten­tial appli­ca­tion to computer vision and aerial robotics.