Tracking odor traces linked to the surface is essential for the survival of terrestrial animals dependent on smell. Little is known about how animals follow trails at an algorithmic level. In the present study, we propose that a tracking animal maintains a noisy estimate of the orientation of the trail based on its past contact with the trail. We show that virtual agents trained to exploit this strategy reproduce the tracking patterns of ants and rodents. The observed patterns emerge simply because of common geometric constraints, which also impose fundamental limits on the speed at which an animal can follow tracks. A series of experiments are offered to quantify how past experience and track statistics shape tracking behavior.
Ants, mice, and dogs often use surface-bound scent trails to establish shipping routes or to find food and mates, but their tracking strategies remain poorly understood. Strategies based on chemotaxis cannot explain the throw, a characteristic sequence of wide oscillations with increasing amplitude performed during prolonged loss of contact with the track. We propose that tracker animals have an intrinsic geometric notion of continuity, allowing them to exploit past contacts with the track to form an estimate of its direction. This estimate and its uncertainty form an angular sector, and emerging research models resemble “sector research”. Reinforcement learning agents trained to perform sector research summarize the different phases of the monitoring behavior observed experimentally. We use ideas from polymer physics to formulate a statistical description of trails and show that research geometry places basic limits on how quickly animals can follow trails. By formulating tracking as a Bellman-like sequential optimization problem, we quantify the geometric elements of an optimal sector search strategy, effectively explaining why and when casting is needed. We propose a set of experiments to infer how tracking animals acquire, integrate and respond to past information about the track being followed. More generally, we define relevant navigation strategies for animals and biomimetic robots and formulate trail following as a behavioral paradigm for learning, memory and planning.
There is no data underlying this work.
- Copyright © 2021 the Author (s). Published by PNAS.