International Union for the Study of Social Insects (IUSSI2018), August 5-10, 2018 in Guarujá, Brazil.

only days left!
Find us: Ft

Using robots to test hypotheses about cooperative transport in ants

Stephen C. Pratt, Aurélie Buffin , Sean Wilson , Takao Sasaki , Spring M. Berman , Stephen C. Pratt
School of Life Sciences, Arizona State University; Mesa Community College ; School of Electrical and Computer Engineering, Georgia Institute of Technology ; Department of Zoology, University of Oxford ; School for Engineering of Matter, Transport, and Energy, Arizona State University ; School of Life Sciences, Arizona State University
Animal groups can accomplish tasks beyond the abilities of solitary individuals, but the benefits of cooperation must be balanced with the costs of coordination. For example, many ant species can form groups to transport items too large for a single porter, but the groups are often slower and less efficient than are lone ants, for reasons that remain unclear. In this study, we tested and rejected one hypothesis for this phenomenon, and then used a robotic testbed to find support for an alternative explanation. The rejected hypothesis held that groups are slower when porters must encircle the load to carry it, because this arrangement places ants in a variety of postures relative to the load and the direction of travel. Porters may therefore have difficulty maximizing individual forces and aligning them with those of other group members. Experiments on the desert ant Novomessor cockerelli, an adept cooperative transporter, found that groups were slower than lone transporters, even when all porters pulled in the same favored orientation, and all carried the same per capita weight. Simulations supported an alternative scenario in which ants vary in their intrinsic speed and the group's speed is limited by that of its slowest member. To test this model, we translated it into a decentralized controller that we implemented on teams of robots. The controller incorporated a real-time reinforcement learning algorithm that successfully reproduced the observed relationship between ant team size and transport speed. The controller required no direct communication among robots or global knowledge of team composition or behavior. Social animals are increasingly recognized as inspirations for swarm robotics. These results show that the relationship can also work the other way, with robotic swarms helping to test mechanistic hypotheses about biological collectives.