Self-organized Multi-agent System for Robot Deployment in Unknown Environments - Foundations on Natural and Artificial Computation - page 165

and ambient intelligence. This integration provides a new way to build intelligent
service robots that opposes the idea of stand-alone robotic platforms. Intelligent
sensor networks, which can be spread out over wide areas, will provide an in-
valuable source of information from beyond robots' immediate surroundings,
allowing the robot to respond to a wide range of events, even when they are
not taking place next to it. This will make the robot look like it has initiative,
and it will improve people's opinion on its role. On the other hand, sensor net-
works connected to the robot will not only enrich robot's perceptions, but they
also could support his movement in the environment and remove or relax the
need of detailed maps of the place where the robot moves. Some researchers
have been already working on enhancing robots' performance on human-based
environments by using intelligent sensor networks. In this sense, some works
focused on camera calibration with robots [1], robot localization and motion
segmentation [2], robot guidance from cameras over small areas [3], new con-
trol architectures aimed towards covering wide areas [4,5,6], and environment
mapping supported by camera networks [7]. Finally, some recent works aim at
building scene semantic models from visual surveillance networks [8]. Although
all these approaches showed interesting results, a scalable, flexible, and robust
system for robot navigation has not yet been proposed.
In this paper we describe a multi-agent system, with which we want to achieve
a fast deployment of autonomous mobile robots in unknown environments, en-
suring robust navigation, and maintaining the cost and the time required for
the deployment to a minimum. Our system consists on two kind of agents: a)
intelligent cameras spread out over the environment, and b) autonomous robots
navigating on it. We based our design on three main requirements: scalabil-
ity (introducing more agents does not involve hardware or software re-design),
robustness (continuity of correct service), and flexibility (independence to the
environment or changes on it). In order to get a scalable and robust system, we
designed it to be fully distributed, favouring inter-agent interactions, and avoid-
ing any kind of centralization or hierarchy. Moreover, instead of designing global
behaviours to fulfil global tasks, our system will be based on low level local in-
teractions amongst agents, leading to self-organized processes, similar to those
usual in biology. Finally, our system will not require any kind of pre-defined
map of the environment to get intelligent robot behaviours. In the future we
plan to use this system to get general-purpose guide robots. Our system will
have to negotiate the movement of several robots in different social events at
schools, forums, conferences, etc. On the other hand, the robots will have to
offer information and advertise the event to people attending or moving around.
In this paper, we present the first prototype of this system. Section 2 gives an
overview of the system. Then, Sec. 3 provides a functional description, explaining
how the most important duties are achieved. After that, Sec. 4 shows some of
the most significant results achieved under real robot operation on real world
experiments. Finally, Sec. 5 summarizes our proposal and points out some of our
future research.