Self-organized Multi-agent System for Robot Deployment in Unknown Environments - Foundations on Natural and Artificial Computation - page 172

Fig. 7. Trajectory described by the robot in one of the experiments. The robot R
navigates from the FOV of camera A, passing through B, C and finally D, which
triggered the call upon detection of the marker M.
scenario was obtained from the robot's odometry and laser logs using the PMAP
SLAM library, compatible with Player-Stage. We run several experiments chang-
ing the initial positions of the robot or the call event, obtaining in all of them a
satisfactory performance of the robot.
5 Summary, Conclussions, and Future Work
In this paper, we proposed an initial approach for a distributed and intelligent
system for robots deployment over unknown environments, consisting on a net-
work of intelligent cameras and autonomous robots. Our purpose was to enable
robots to attend world events happening out of their surroundings, thus increas-
ing robot's initiative. We designed our system on the base of scalability, robust-
ness, and flexibility, independent of the environment and of the number and
layout of the agents deployed. In this regard, we did not use any centralization
or other kind of hierarchy, but biologically inspired self-organization processes,
based on distribution, inter-agent independence, and emergent behaviours out
of local interactions, instead of global plans. This resulted on a system highly
independent of environment changes, redundant, and flexible enough to cope
with a wide range of spatial distributions of the cameras. Also, our system is
able to work with several robots or to attend to simultaneous events.
Real world experiments proofed that our system makes it possible for the
robot to attend events detected by the cameras, regardless of both robots' and
events' positions. Upon distribution in the environment, the cameras detect au-
tomatically their neighbours, forming a global camera network out of local neigh-
bourhood relationships. The routes are planned by a distributed process of local
interactions, allowing the robot to navigate between each pair of cameras, count-
ing on camera's support.