Ant foraging: optimizing self-organization as a solution to a traveling salesman problem.

Foraging ant colonies often face the challenge that food items can appear unpredictably across their territory. This is analogous to traveling salesman/salesperson problems (TSP), wherein solutions to visiting multiple possibly-rewarding sites can vary in cost, travel distance, or site revisits. However, TSP solutions for ants are likely also constrained by cognitive limitations. Rather than envisioning entire routes, ants probably determine their paths by individual-level responses to immediate stimuli, such as nestmate presence or avoiding revisiting an already explored site. Thus, complex group-level search and food retrieval patterns may self-organize from simple individual-level movement rules. Here we derive solutions through simulations that optimize net foraging gains across groups of ant-like agents. Agent search strategies evolve in three spatial networks that differ in travel distances to nests, connectivity, and modularity. We compare patterns from simulations to observed foraging of Argentine ants (Linepithema humile) in identical spatial networks. The simulations and ant experiments find foraging patterns are sensitive to both network characteristics and predictability of food appearance. Simulations are consistent in multiple ways with observed ant behavior, particularly in how network arrangements affect search effort, food encounters, and forager distributions (e.g., clustering in the more connected cells). In some distributions, however, ants find food more successfully than simulations predict. This may reflect a greater premium on encountering food in ants versus increasing find exploitation rates for agents. Overall, the results are encouraging that evolutionary optimization models incorporating relevant ant biology can successfully predict expression of complex group-level behavior.