Data-driven olfactory search in a turbulent flow

Tracking the source of an odor which is advected by a turbulent flow is an important behavior for many flying insects and other animals. This olfactory search problem is rendered especially difficult by the intermittency intrinsic to turbulence, and it requires complex search strategies which properly leverage infrequent odor detections. In this work, we perform direct numerical simulations (DNS) of tracer particles emitted from a points source in a turbulent flow with a mean wind. We study the concentration statistics of the tracer data and use the data to extract model-based policies for search. We compare the empirical performance of near optimal policies (in the sense of partially observable Markov decision processes) to that of several heuristics.