Visual attention in avian collision-avoidance and homing: what visual features do birds attend to, and why?

Natural environments are too complex to process in their entirety, so visually-oriented animals such as birds and primates focus their attention on selected features of visual scenes. They achieve this by directing their gaze using strategies that maximize the information obtained. For example, our own gaze shifts predict where our next movements will occur as we navigate, but little is known of how birds direct their gaze in flight. Addressing this knowledge gap presents a significant challenge because of the difficulty of measuring a bird’s gaze experimentally, and of manipulating its visual environment on a meaningful scale. Our frontier bioscience project aims to understand how pigeons direct their gaze to attend to specific subregions of the visual scene during biologically significant goal-oriented tasks. In so doing, we seek to unravel the visual features that birds use to guide their flight, with broad implications for cognitive science and bio-informed design. Unlike previous studies that have focused on where birds go, our work breaks new ground by analyzing where they look, what they look at, and why certain features capture their attention. This approach will establish a novel non-invasive model system for exploring cognitive attentional mechanisms in birds, and shed new light on algorithmic attentional mechanisms in computer vision. Our primary research objective is to identify the visual features that birds attend to during collision-avoidance and homing. We will investigate whether birds rely on specific landmarks for navigation, and will explore other visual features that draw their gaze in flight. By making experimental manipulations in the lab and observing how birds respond to landscape-level changes in the environment, we will study how avian flight behaviour responds to visual cues. We will also train machine learning algorithms using the data we collect to mimic a bird's gaze. These algorithms will be compared with the attention mechanisms of the latest models in computer vision, advancing both fields in tandem. To achieve our objectives, we will investigate collision-avoidance behaviours by manipulating obstacles within a state-of-the-art motion capture studio designed for bird flight research. To study homing, we will utilize miniature inertial sensors tracking the bird's position and head direction. By combining open mapping, aerial imagery, and laser scanning with 3D modelling and image-rendering technologies like Google Earth, we aim to visualize the world as a bird sees it in flight. Building on two decades of homing pigeon research in Oxford using GPS tracking, our project leverages significant changes to the environment to enable a landscape-level experiment. This includes seasonal changes such as river flooding and anthropogenic change including housing developments and solar farms in the Oxford greenbelt, facilitating a before-after control-impact study design. The outcomes of our research will not only benefit vision research and movement ecology, but will also extend to experimental psychology, artificial intelligence, and bio-informed design. Applications include enhancing visual attention mechanisms in AI, and developing new visual guidance algorithms for autonomous vehicles operating without GPS. Additionally, the insights from our work may inform the design of the built environment, to better accommodate the animals we share it with; for instance, by designing urban flyways, or enhancing infrastructure to minimize collisions. Our unprecedented dataset of bird's-eye views of the world will be invaluable for machine learning, and will be shared with a broad audience through outreach activities employing the medium of VR.