TouringMachines:
an architecture for dynamic, rational, mobile agents

Abstract

It is becoming widely accepted that neither purely reactive nor purely deliberative control techniques are capable of producing the range of behaviours required of intelligent computational or robotic agents in dynamic, unpredictable, multi-agent worlds. We present a new architecture for controlling autonomous, mobile agents – building on previous work addressing reactive and deliberative control methods. The proposed multi-layered control architecture allows a resource-bounded, goal-directed agent to react promptly to unexpected changes in its environment; at the same time it enables the agent to reason predictively about potential conflicts by constructing and projecting causal models or theories which hypothesise other agents’ goals and intentions. The line of research adopted is very much a pragmatic one. A single, common architecture has been implemented which, being extensively parametrized, allows an experimenter to study functionally- and behaviourally-diverse agent configurations. A principal aim of this research is to understand the role different functional capabilities play in constraining an agent’s behaviour under varying environmental conditions. To this end, we have constructed an experimental testbed comprising a simulated multi-agent world in which a variety of agent configurations and behaviours have been investigated. Experience with the new control architecture is described.