I presented in this literature review core concepts on which the research on this thesis is based. The demands of MAS require planning-agents to communicate and manage their plan resources effectively. Furthermore, the complexity of MAS limits the use of single-agent planning strategies in distributed problems because the local beliefs of an agent are often not sufficient to generate a satisfactory plan.
Argumentation in Artificial Intelligence is a formal discipline that tries to implement models of argument used by humans through computational models to enable agents to cooperate and coordinate efficiently in order to reach their goals. Argumentation provides a powerful framework for interacting agents taking decisions, assessing the validity of information, or otherwise resolving differences of opinion.
In particular, Argumentation has been used successfully in AI to build computer systems able to exchange arguments to establish dialogues as a medium of negotiation, persuasion, deliberation, etc. where agents try to persuade other parties to accept of- fers through argumentation. In particular, research in argumentation and automated dialogues is useful to define how self-interested autonomous agents can interact with one another. The categorization of dialogues given by Walton and Krabbe has inspired researchers in MAS to design and implement models of communications based on argu- mentation theories. The correctness and implementation of teamwork dialogues is a key issue for this approach. This is the area in which the research on this thesis focuses.
The practical reasoning process in autonomous agents consist of agents generating a possible set of actions to execute using its current beliefs and desires and choosing the best option. One way to model practical reasoning in autonomous agents is using argumentation schemes due to their potential for making improvements in the reasoning capabilities of agents and the automation of agent interactions using critical questions. Planning tasks in MAS require communication, and the exchange of beliefs desires and intentions through interactions in different steps in the planning process. The following chapters present concepts on how agents can engage in interactions to propose, evaluate and refine plans in a multi-agent scenario.
A Framework for Plan Proposals
This chapter presents an Argumentation Scheme for Plan proposals (ASP) to propose multi-agent temporal plans based on the promotion or demotion of values. The multi- agent plans considered are conceptualized as sequences of scheduled actions for each agent involved in the achievement of the goal. The argumentation scheme presented builds on an argumentation scheme that allows agents to create proposals for action based on the promotion or demotion of values in [11]. The objective is that autonomous agents can use the ASP scheme and its associated critical questions to evaluate multi- agent plans.
Contrasting with the approach in [11], the type of actions I consider are durative, that is, actions have a start time and a duration. Durative actions were originally defined in the context of temporal planning. Together with the argumentation scheme, I present an analysis of critical questions which provide pointers to question and/or attack the plan proposal.
The remainder of this chapter is structured as follows: Section 3.1 presents an ap- proach to provide agents with mechanisms to argue about cooperative plans. Appendix A presents an example where agents need to propose an agree on a plan in a war zone. The example will help to exemplify the elements presented throughout the thesis. Sec- tion 3.2 presents an argumentation scheme for durative action proposals. Section 3.3 present a discussion on multi-agent temporal plans and the implications of having indi- vidual action dependency in multi-agent plans. Section 3.4 presents the argumentation scheme for plan proposals: ASP. Section 3.5 presents an analysis of critical questions that match the ASP argumentation scheme where questions are categorized into seven layers. Finally, Section 3.6 presents a chapter summary.
3.1
An Approach to Propose Cooperative Plans using Ar-
gumentation Schemes
Argumentation-based dialogues (cf. Section 2.2.3) can be used to specify how agents can engage in a dialogue about plans. The approach presented in this thesis assumes agents need to propose, justify and evaluate multi-plans taken from a private plan repository (i.e. plans created by or supplied to the agents) and then engage in a dialogue to justify, evaluate and agree on a plan to execute.
Planning in multi-agent scenarios is a complex process. This thesis propose the use of a structured dialogue using argumentation principles in order to agree on a plan in situations where agents have different views about the world and/or different interests. Agents may have different information about the world because they cannot update continuously their beliefs or because they have a partial view of it. Whatever is the reason, the aim is to deal with these conflicts through dialogue.
A structured argumentation-based communication process where agents can apply their preferences in the selection of actions and plans can help to solve problems related to the selection and modification of multi-agent plans. Agents should be able then to propose and evaluate plans in a structured way while obtaining information about all the possible factors that may affect the plan execution. It is out of the scope of the thesis all the issues related to theexecution of the plan.
Figure 3.1 an approach to plan creation where agents merge their Knowledge Bases and then create a plan. Figure 3.2 presents the approach considered in this thesis where agents discuss plans already formed and then engage in an argumentative dialogue to agree on which plan to execute. I assume that all the agents involved in the dialogue are willing to cooperate to achieve the goal and that they may have differences on their beliefs about the world and their preferences over the values promoted by the plan. An important difference between these two approaches is that agents do not merge their complete Knowledge Bases, but only exchange such information as is relevant to the proposed plans. Therefore, the objective is that agents take advantage of an argumentation-based dialogue structure to resolve their differences and provide them with options where they can use their preferences as the dialogue progresses. An agent ag for the purpose of this thesis has the following elements:
• a set of state predicates used to represent the world;
• a belief assignment to each of the predicates that represents the agent’s beliefs about the current state of the world. The beliefs of agents do not have any uncer- tainty. They can be wrong about their beliefs, but they are determined;
• a set of durative actionsAcagthe agent can perform with the elements that conform the action i.e. the preconditions, start effects, invariant conditions, termination conditions and end effects;
Agent 2 Agent 1 Final Plan Knowledge Base Planner Plan selection (Preferences applied here)
Merged Knowledge Base Knowledge
Base
Knowledge Base +
Figure 3.1: A Distributed Artificial Intelligence Approach to Plan Creation.
• a condition or effect of an action Ac is defined for an action if the predicate the condition is part of the set of conditions in actionAc. This will be furter explained in the next subsections;
• a plan goal G;
• a set of values Vag and a preference over these valuesδ;
• an element indicating if the agent has evidence to support the element (e.g. a predicate, the belief assignment, an action, a value);
• an internal planner that receives as input the initial state of the world and a set of actions and outputs a multi-agent plan that consist of set of scheduled actions for each agent involved in it that achieve the goal;
• a setpossibleof multi-agent plansP Lgenerated by an internal planner that achieve the goal and promotes/demotes values in Vag;
• aplan checker that receives as input the initial state of the world , a set of actions, a multi-agent planP Land the goalGand verifies if the planP Lachieves the goal G.
Agent 2 Agent 1 Dialogue Best Agreed Plan Planner Planner Agent Plans
Persuasion and Deliberation Dialogue (agents’ preferences influence the plan modification and selection)
Figure 3.2: Argumentative Approach to Plan Creation.
In Appendix A, an example is presented to explain the elements introduced in the thesis. The next section presents the concept of durative actions in the context of agents proposing multi-agents plans for execution.