Agent deliberation in an executable temporal framework

Abstract

Autonomous agents are not so difficult to construct. Constructing autonomous agents that will work as required is much harder. A clear way in which we can design and analyse autonomous systems so that we can be more confident that their behaviour is as required is to use formal methods. These can, in principle, allow us to exactly specify the behaviour of the agent, and verify that any implementation has the properties required. In addition to using a more formal approach, it is clear that problems of conceptualisation and analysis can be aided by the use of an appropriate abstraction. In this article we tackle one particular aspect of formal methods for agent-based systems, namely the formal representation and implementation of deliberation within agents. The key aspect here is simplicity. Agents are specified using a relatively simple temporal logic and are executed by directly interpreting such temporal formulae. Deliberation is captured by modifying the way in which execution handles its temporal goals. Thus, in this article we provide motivations, theoretical underpinnings, implementation details, correctness arguments, and comparisons with related work.

Formal development of multi-agent systems using MAZE

“…Hindriks et al [24] presents an agent programming theory which provides both an agent programming language and a corresponding verification logic to verify such agent programs. Fisher [25] formalises the deliberation of agents using executable temporal logic. The deliberation behaviour is captured by modifying the execution rules handling the agents' temporal goals.…”

A review of key planning and scheduling in the rail industry in Europe and UK

“…Multi-agent systems are then, by definition, interacting collections of agents (Wooldridge 6 ). Fisher 7 has investigated the design and evaluation of agents providing a framework for the temporal execution of agent designs. This work was motivated by the lack of such a framework and methods to assess the actual behaviour of agents once designed.…”

Where logic and agents meet

“…the current list of 'goals' can be re-ordered or modified at any moment -since computation concerns reducing the first goal on the list, this change of emphasis can be viewed as the deliberation the agent undertakes about its current goal [67]; and 2. the list of facts and rules can be re-ordered -since this ordering describes the next rule/fact to be attempted, then re-ordering corresponds to the agent deliberating about, and changing its view of, how to reduce the current goal.…”