George Bard Ermentrout, PhD
Distinguished University Professor
Department of Mathematics
University of Pittsburgh
(February 23, 2021)
Follow your nose: the dynamics of olfactory navigation
We have all seen police dramas where the bloodhounds are called in to locate a missing person. Dogs (and other animals) heavily depend on their sense of smell in order to navigate the world. But how do their olfactory systems use odors to find their way to food when so many other smells are present to confuse the signal? Dr. Ermentrout uses computational methods to determine how the olfactory system can pinpoint location from odor cues.
Olfaction (the sense of smell) is the oldest of our sensory modalities and has been used for millions of years for animals to find mates, find food, avoid predators, etc. In a large multi-investigator collaboration, we have begun to try to understand the algorithms animals use to navigate complex odor landscapes. I will describe several simple algorithms that use local spatial and temporal information about the odor to locate its source. The algorithms fall into two simple categories: differences between two sensors and differences between two different samples. With data from trail-following and spot finding by mice, I attempt to assess the different strategies and how parameters in the strategies affect performance. I also test the algorithms on odor plumes imaged by my collaborators and also in a mobile robot.
Underlying these simple algorithms is some interesting nonlinear dynamics. I will discuss the continuous dynamics of binaral search where the organism uses the concentration differences between two sensors to steer toward the source. Depending on the odor environment, various types of complex dynamics emerge including stable fixed points, periodic orbits, torii, and chaos. Secondly, I will describe a discrete algorithm where the animal samples the concentrations at different time points and uses this comparison to determine the heading. By reducing this algorithm to its very simplest form, we are able to also analyze the underlying dynamics. Finally, I will show the role of “noise” on improving the algorithms and how it can be leveraged as a search strategy by exploring a first passage time problem applied to spot finding.
Visual abstract from George Ermentrout's talk
