Abstract
Aquatic organisms offer compelling evidence that local flow sensing alone, without vision, is sufficient to guide them to the source of a vortical flow field, be it a swimming or stationary object. However, the feedback mechanisms that allow a flow-sensitive follower to track hydrodynamic trails remain opaque. Here, using high-fidelity fluid simulations and Reinforcement Learning (RL), we discovered two equally effective policies for trail following. While not apriori obvious, the RL policies led to parsimonious response strategies, analogous to Braitenberg’s simplest vehicles, where a follower senses local flow signals and turns away from or towards the direction of stronger signal. We analyzed the stability of the RLinspired strategies in ideal and simulated flows and demonstrated their robustness in tracking unfamiliar flows using diverse types of sensors. Our findings uncovered a surprising connection between the stability of hydrodynamic trail following and sense-to-response time delays, akin to those observed in the sensorimotor systems of aquatic organisms, and could guide future designs of flow-responsive autonomous robots.
Competing Interest Statement
The authors have declared no competing interest.
