Acronym MAIS-S
Funding Reference FCT - CMU-PT/SIA/0023/2009
Dates 2010-09|2013-08
Summary

With the generalized use of intelligent technology, the interaction between multiple smart devices poses interesting challenges both in terms of engineering and research. One interesting aspect of this phenomenon in the context of this project is the appearance of networks of heterogeneous devices that must operate in a fully distributed manner while sharing information necessary to complete some preassigned task. In this project, we propose that such complex networks be modeled as multiagent systems where each node corresponds to an agent. This interpretation suggests several interesting research avenues, some of which will be the focus of this project. We propose the use of a class of decision-theoretic models – Dec-POMDPs and specializations thereof – that naturally captures the decentralized nature of these networks in terms of local perception (the information that each node can acquire per se), interaction/communication (the exchange of information between the nodes) and local actuation (each node processes locally the available information and acts accordingly). In fact, one can argue that several systems typically considered in a fully centralized fashion (such as surveillance systems) could potentially benefit from this multiagent view of the network (e.g. in terms of robustness, more efficient communication, etc).

We are interested in heterogeneous surveillance networks that include different kinds of nodes, possibly with different perceptual and actuation capabilities, as well as different processing power. From this perspective it is natural, for example, to have nodes corresponding to cameras mounted on mobile robots. The control and even the positioning of the robots should be decided locally but in such a way to globally optimize the performance of the network.

The project will address the following problems:
Scalability: General Dec-POMDPs have a worst case complexity that is NEXP-complete. As such, significant efforts have been devoted to finding simpler models that, while capturing the fundamental features of Dec-POMDPs, are still amenable to efficient computation. We explore how local interactions can simplify the process of decision making in these multiagent models.
Efficient communication: We study the tradeoff between using communication to simplify the decision process against the cost of overloading the communication channels.
Robustness: How should the network adjust when one or more of the nodes fail?
Implementation: We investigate the problem of navigating the mobile robots as well as efficiently interfacing mobile and static nodes. We will implement a simple prototype of the system in a structured environment (ISR), using an existing infrastructure, and we consider the possibility of conducting tests in a real scenario.

Research Groups Intelligent Robots and Systems Group (IRSg)
Project Partners INESC-ID (coordinator)
ISR/IST Responsible
Matthijs Spaan