A visual efference copy-based navigation algorithm inDrosophilafor complex visual environments

Abstract

Drosophila visuomotor processing has been intensively studied in recent years, leading to a qualitative understanding of individual neural circuits. However, the collective operation of these circuits during naturalistic behaviors, in which flies encounter a mixture of complex visual stimuli -- including those caused by their own actions -- remains unexplored. In this study, we developed an integrative model of Drosophila visuomotor processing, wherein multiple visuomotor circuits interconnect through an efference copy (EC) mechanism. To derive the model experimentally, we analyzed the wingbeat responses of flying Drosophila to individual, rotating visual patterns. We then combined these models to build an integrative model for superposed visual patterns, using three different strategies: the addition-only, the graded EC, and the all-or-none EC models. We compared orientation behaviors of these models with those of flying Drosophila that rotates their body freely in response to complex visual patterns. Results of these experiments support the all-or-none EC model, in which the amplitude of the flight turn is unimpeded by the background scene. Together, our "virtual fly" model provides a formal description of vision-based navigation strategies of Drosophila in complex visual environments and offers a novel framework for assessing the role of constituent visuomotor neural circuits in real-world contexts.

A control engineering perspective on the advantages of efference copies

“…Yet, when octopaminergic neurons are silenced, there is no change in average flight speed [13]. The physiological [11,12,5,14] and behavioural [13] evidence strongly support and align with, the inner-loop and outer-loop structure for insect sensorimotor control.…”

“…Finally, there is a distinction made in recent work between "graded" and "all-or-none" efference copies [14]; where a graded efference copy is the output of a forward model predicting the sensor response to cancel, and all-or-none is effectively turning off the sensor when enacting a desired state change. In this paper we have analysed a graded efference copy model, however, animals likely employ both methods.…”