This research is being carried out in the field of autonomous collective robotics. The primary goal of the research is investigating principles to automatically design the controllers of autonomous robots that have to coordinate in order to accomplish common tasks. In the long term this research will lead to applications in areas such as open field and marine monitoring, people rescuing, stock automatic transport, traffic control, material routing. The research is also investigating principles that regulate animal and human distributed coordination.
The research is carried out using a realistic 3D simulator of some robots currently under construction (see below). The robots are capable of linking to each other through mechanical grippers whenever this is necessary to carry out common tasks. Our research group focussed on the particular problem of allowing linked robots to coordinate their motion, either to explore a field or to move towards a common target. The controls of the robots, made up by neural networks, has not been designed by hand, but has been automatically "evolved" through "genetic algorithms" that directly mimic natural selection. These algorithms generate random controllers, test the performance of the robots controlled by them, select the controllers with the best performance, produce other controllers from them by adding some random noise, select again the best controllers and so on. This process is repeated several times until some robots successfully accomplish the task under focus. Interestingly, this automatic process showed that the evolved behaviours that lead to the most effective collective coordination strategy are those based on a conformist behaviour ("do what the others are doing").
The research is part of the "SWARM-BOTS" project (2001-2004) funded by the European Commission within the funding brunch "IST-FET, Information Society and Technology - Future and Emergent Technologies". The project is being carried out in close collaboration with two research groups from Switzerland (EPFL, IDSIA), that are building the physical robots used in the project, and two groups from Belgium (IRIDIA, Cenoli), that, together with our group, are tackling the problem of how to control the robots.
In the future, the research will be extended to study: a) the role of explicit communication in coordination; b) different types of social coordination, in particular distributed vs. centralised coordination.
Gianluca Baldassarre, Domenico Parisi, Stefano Nolfi
Vedi anche:
Immagini:
