Extending the RoboCup Rescue to Support Stigmergy: Experiments and Results

Abstract

Social insects have inspired researches in computer sciences as well as
engineers to develop models for coordination and cooperation in multiagent systems.
One example of these models is the model of stigmergy. In this model agents use
indirect communication (comunication trough the environment) in order to coordinate
actions. The RoboCup Rescue simulator is used as a testbed to evaluate this model
in a real world considering a highly constrained scenario of an earthquake. This pa-
per investigates the feasibility of using stigmergy in the RoboCup Rescue and the
improvements of performance can be obtained. We extended the RoboCup Rescue
environment to enable the use of stigmergy by the agents. We compared the results of
a multiagent system that uses stigmergy against two other approaches: a multiagent
system that uses a greedy strategy and no communication, and a multiagent system
where agents communicate via direct messages. Experimental results shown that the
use of stigmergy leads to an improvement on agents’ performance by 9.02% to 38.6%
if comparing to the system with no communication and can be statistically equivalent
to the system which uses messages, depending on the scenario.