In an attempt to realize decentralized cooperation and coordination in a heterogeneous
team of autonomous mobile agents, we have proposed a platform to model many practical
issues encountered in the process. The system includes an unexplored environment and a
team of agents with different capabilities. The agents' movement follows trajectories limited
by speed and turning radius, and their actions may be affected by their orientation relative
to targets. The stochasticity introduced by task transition makes o-line task assignment
inapplicable.
A basic distributed real-time assignment algorithm and an intelligent predictive decision
making strategy are proposed and compared. The results show that prediction can help
improve the performance of cooperative agent teams engaged in search-and-engage missions.
These algorithms are implemented with the assumption of perfect communication.
The assumption of perfect communication is relaxed in the second part of the work reported,
resulting in a truly decentralized system that is simulated in a somewhat simpler
mission scenario. Each agent in this decentralized model has its own subjective information
base (SIB) based on limited observations and information sharing, and uses a set of information
fusion heuristics to maintain this SIB. The performance of the system is evaluated
relative to the degree of information sharing. The results indicate that cooperation based
on information exchange { even if it is imperfect { can produce excellent performance.
A hybrid communication strategy combining occasional limited-range communication
and opportunistic peer-to-peer synchronization is then proposed. Using Monte Carlo simulations,
it is shown that this strategy can significantly supplement the efficacy of a diffusive
information sharing strategy.