Sniffing, Physiology, and Search: Odor Navigation in Embodied Neural Networks

We investigate respiratory odor navigation in a minimal brain-body-environment model where embodied continuous-time recurrent neural network agents are evolved to locate a chemical source using a temporally fluctuating respiratory olfactory signal. Because respiration rhythmically gates sensing, the agent never senses a static gradient but only a time-varying concentration trace tied to its own breathing. We show that evolution reliably produces agents that locate and remain near the source and that a respiratory-neural mechanism supports two robust navigational regimes: continuous oscillatory casting and episodic saltatory search. Furthermore, all successful agents synchronize respiration and motor output, mirroring sniff-synchronized navigation recently observed in rodents. Dynamical systems analysis reveals how a respiratory pattern generator stabilizes internal physiology while reconfiguring motor dynamics across the respiratory cycle, explaining how changes in respiratory period and amplitude switch between casting and saltatory navigation.