The Khepera Ants Experiments: The Roots of Decentralized Coordination in a Population of 12 Robots

Jean-Bernard Billeter, LAMI, EPFL, CH-1015 Lausanne, Switzerland

Michael Krieger, Institute of Zoology and Animal Ecology (IZEA), Lausanne University, CH-1015, Switzerland

This talk will be centered around the presentation of a video showing a series of experiments in which a team of up to 12 real robots is given the mission of keeping their nest's energy at a safe level by collecting food items in the environment. To achieve this mission efficiently, the robots must be controlled by some task allocation mechanism that will modulate in real time the number of individuals staying in the nest (task 1) versus those who go foraging (task 2).

We propose a simple and completely decentralized task allocation mechanism using a single parameter, the activation threshold, i.e. the nest's energy level under which a given robot decides to leave the nest and go collect food items. The experiments show that giving different but individually fixed activation levels to the robots results in an efficient dynamical task allocation. The implications of group size in relation to performance and to mission success are discussed.

In a second set of experiments, we modify the behaviour of the robots by allowing successful robots to lead fellow robots to sources of food items. This additional "social" feature has a positive effect on the team's performance.