Dynamic integration of multisensory turn commands supports active reorienting during ongoing navigation

Summary A longstanding goal of systems neuroscience is to quantitatively describe how the brain integrates cues from multiple modalities over time. Here we develop a closed-loop orienting paradigm in Drosophila to study the algorithms by which stimuli from different modalities are combined during ongoing navigation. We show that flies faced with an attractive visual and an aversive mechanosensory cue exhibit sequential responses, first turning away from the co-localized stimuli before turning back toward them. We also find that the presence of the aversive cue slows flies’ turns toward the attractive target, suggesting that conflicting unimodal turn commands are summed to produce multisensory turns of intermediate velocity. We then test a series of computational frameworks and find that integration is best described by a model in which multimodal stimuli are continuously and dynamically scaled, converted into turn commands, and then summed to produce ongoing orientation behavior.