This project aims to
design and develop teams of co-operating autonomous agents,
which could generally be used to expand the action-horizon of
humans in inaccessible industrial fluidic applications.
Successful collective performance in these missions critically
depends upon: (a) accurate environmental perception, (b) a real
time decision making and action loop, (c) emergent goal directed
social behavior. In order to meet these requirements, we:
(a) Propose a
novel agent architectural design based on modular Spiking Neural
Network, which will serve as a "blueprint" for agent manufacture
direct/indirect communication mechanisms for collaborative
societies formation, and agent to environment communication.
(c) Design and
develop an accompanying Development Environment, which will
facilitate the evolution of complex agent SNN architectures-
multiagent societies, and the development of succesive CAA
In the current
project we are trying to solve a particularly difficult proble
i.e. fault localization and repair in inaccessible industrial
fluidic environments by using teams of very simple agents. The
mission can be accomplished only if the societies of agents
employed can manifest emergent intelligent collective behavior.
will generally be composed of:
sensor/actuator subsystem which will provide multiple input
data and enable real-time navigation.
communication subsystem that will support self-organization
of behavior at a social level
computational subsystem which will analyze the information
of the sensory channels, control the actuator subsystems of
the agent and initiate and maintain goal-directed individual
and social behavior among agents. The computational
subsystem is composed of simple connectionist spiking neural
networks. The latter are realized in spiking neural
hardware, chosen for its ability to meet the real time
signal processing imposed by the application context.
The project will
demonstrate the feasibility of the concept by developing an
early prototype; the project progress towards the ultimate goals
will be measured by implementing and testing successive
generations of agents. In accordance with potential application
areas, two types of agents will be implemented and tested:
exploring micro-environments in simulations.
containing reduced versions of the architectures for fault
location and repair in actual fluidic environments.
development environment will allow rapid prototyping and
evaluation of successive generations of agents. This environment
will include modules that a) automatically generate agent
architectures and b) evolve optimal architectures based on the
team's performance in a simulated environment.