K. S. Barber

Comprehending agent software

Software comprehension (understanding software structure and behavior) is essential for developing, maintaining, and improving software. This is particularly true of agent-based systems, in which the actions of autonomous agents are affected by numerous factors, such as events in a dynamic environment, local uncertain beliefs, and intentions of other agents. Existing comprehension tools are not suited to such large, concurrent software and do not leverage concepts of the agent-oriented paradigm to aid the user in understanding the softwares behavior. To address the software comprehension of agent-based systems, this research proposes a method and accompanying tool that automates some of the manual tasks performed by the human user during software comprehension, such as explanation generation and knowledge verification.

Dynamic reorganization of decision-making groups

Agents in a multi- agent system must coordinate to achieve their goals, in general. Establishing an organizational structure that specifies how agents in the system should work together helps multi- agent systems achieve effective coordination. Among other things, an organizational structure specifies agent decisionmaking frameworks. A decision- making framework identifies the locus of decision- making control for a given goal and the authority of decision- makers to assign subtasks in order to achieve that goal. Agents may participate in different decision- making frameworks for each goal they pursue. Agents who implement the capability of Adaptive Decision- Making Frameworks (ADMF) are able to dynamically modify their decision- making frameworks at run- time to best meet the needs of their current situation. Through ADMF, agents are able to reorganize decision- making groups by dynamically changing (1) who makes the decisions for a particular goal and (2) who must carry out these decisions. This paper presents experiments exploring the following hypothesis: Multi- agent systems that implement ADMF can perform better and more robustly across run- time changes in situation than systems that maintain static decision- making frameworks. Experimental results show that no one decision- making framework performs best across various situations that may be faced at run- time. Further experiments show that the implementation of ADMF in multi- agent systems can improve system performance across changing situations. These experimental results clearly motivate the implementation of Adaptive Decision- Making Frameworks in multi- agent systems.

Conflict representation and classification in a domain-independent conflict management framework

This paper explores the process of representing and classifying conflicts in agentbased systems. These capabilities form the basis for an in-progress implementation of a domain-independent conflict management framework for multi-agent environments. The following hypothesis is to be tested under this implementation: the most appropriate conflict resolution strategy for a given conflict is contingent on the type of conflict, the agents’ organizational role, and the agents’ solution preferences. A three layered representation of conflict is employed (i.e. goal, plan, belief). The representation serves to trace the source(s) of conflict and provides a map to potential resolutions. Each layer of the representation is assigned an associated conflict type reflecting the source of the conflict (i.e. what must be changed to eliminate the conflict: goals, plans, or beliefs). A formal definition of these three conflict classes is provided. The ongoing implementation of these concepts through the Sensible Agent testbed is discussed. Technical Report TR98-UT-LIPS-AGENTS-02

Using Policies for Information Valuation to Justify Beliefs

This research presents a belief revision algorithm, based on intuitive policies for information valuation. Policies for valuation of information, based on characteristics of the information and the sources providing it, are delineated as guidelines for algorithm construction. In addition, each policy is traceably incorporated into the belief revision process to provide justification for calculated belief certainties. Finally, since modeling of information source trustworthiness can be complicated, significant effort is devoted to constructing these reputation models. Experimental results show that application of information valuation policies to belief revision yields significant improvement in belief accuracy and precision over no-policy or single-policy belief revision.

Sensible agents: an implemented multi-agent system and testbed

Sensible Agents have been engineered to solve distributed problems in complex, uncertain, and dynamic domains. Each Sensible Agent is composed of four modules: the Action Planner, Perspective Modeler, Conflict Resolution Advisor, and Autonomy Reasoner. These modules give Sensible Agents the abilities to plan, model, resolve individual conflicts, and change agent system organization. Two component suites provide a variety of user- oriented features: the Sensible Agent Run- time Environment (SARTE) and the Sensible Agent Testbed. The SARTE provides facilities for instantiating Sensible Agents, deploying a Sensible Agent system, and monitoring run- time operations. The Sensible Agents Testbed facilitates automated generation of parameter combinations for controlled experiments, deterministic and non-deterministic simulation, and configuration of Sensible Agents and data acquisition. Experimentation is a crucial step in gaining insight into the behavior of agents, as well as evidence toward or against hypotheses. Using a real- world example, this paper explains and demonstrates: (1) the functional capabilities of Sensible Agents, (2) the Sensible Agent Run- Time Environments facilities for monitoring and control of Sensible Agent systems and (3) the experimental set- up, monitoring, and analysis capabilities of the Sensible Agent Testbed.

Infrastructure for Design, Deployment and Experimentation of Distributed Agent-based Systems: The Requirements, The Technologies, and An Example

This paper discusses infrastructure for design, development, and experimentation of multi-agent systems. Multi-agent system design requires determining (1) how domain requirements drive the use of agents and AI techniques, (2) what competencies agents need in a MAS, and (3) which techniques implement those competencies. Deployment requirements include code reuse, parallel development through formal standardized object specifications, multi-language and multi-platform support, simulation and experimentation facilities, and user interfaces to view internal module, agent, and system operations. We discuss how standard infrastructure technologies such as OMG IDL, OMG CORBA, Java, and VRML support these services. Empirical evaluation of complex software systems requires iteration through combinations of experimental parameters and recording desired data. Infrastructure software can ease the setup, running, and analysis of large-scale computational experiments. The development of the Sensible Agent Testbed and architecture over the past six years provides a concrete example. The design rationale for the Sensible Agent architecture emphasizes domain-independent requirements and rapid deployment to new application domains. The Sensible Agent Testbed is a suite of tools providing or assisting in setting up, running, visually monitoring, and chronicling empirical testing and operation of complex, distributed multi-agent systems. A thorough look at the various Sensible Agents infrastructure pieces illustrates the engineering principles essential for multi-agent infrastructure, while documenting the software for users.

Constructing and dynamically maintaining perspective-based agent models in a multi-agent environment

This paper describes a model that explicitly represents the declarative and behavioral knowledge of a goal-driven agent. The developed declarative and the behavioral models allow an agent to reason about itself, other agents, and the environment. The declarative model specifies data and services (capabilities) assigned to an agent while the behavioral model specifies the execution model of an agent (defined as states, transitions between states, and events affecting the transitions). An Extended Statechart (ESC) is used as the execution model. To maintain these models, a self-contained computational module called the Perspective Modeler is proposed and incorporated into the Sensible Agent Architecture. With the Perspective Modeler, a Sensible Agent has the capability to model itself, other agents, and the environment to generate more desirable behaviors. The Perspective Modeler is implemented as a CORBA object, as demonstrated successfully at the AAAI‘’98 Intelligent Systems Demonstration session held in Madison, WI. Submitted to Agents’99

Agent trust evaluation and team formation in heterogeneous organizations

This demonstration highlights different aspects of the bottom-up assembly of multi-agent teams; illustrating trust evaluation of potential partners via experience- and reputation-based trust models, multi-dimensional trust evaluation of potential partners, task selection through personality-based modeling and team selection strategies that maximize a teams ability to function in dynamic environments. The demonstration format will be a software live demo with supporting slide shows.

Verifying and Explaining Agent Behavior in an Implemented Agent System

As agent systems become more sophisticated, there is a growing need for agent-oriented debugging, maintenance, and testing methods and tools. This paper presents the Tracing Method and accompanying Tracer tool to help verify actual agent behavior in the implemented system against expected (or designed) agent behavior. The Tracing Method captures dynamic run-time data as actual agent behavior, creates modeled interpretations in terms of agent concepts (e.g. beliefs, goals, and intentions), and compares those models to the agent behavior expected by the designer or developer; thereby, gaining insight into both the design and the implemented agent behavior. The Tracer tool can help: (1) determine if agent design specifications are correctly implemented and guide debugging efforts and (2) discover and examine motivations for agent behaviors such as beliefs, communications, and intentions.

Conflict detection during plan integration based on E-PERT diagrams

Abstract This paper describes techniques developed for conflict detection during plan integration based on E-PERT diagrams. PERT diagrams were originally developed in 1980s for project management to provide a global consistent view of parallel activities within a project. We E-PERT diagrams for use in the plan integration activity within multi-agent systems. The E-PERT diagram contributes to maintain traceable temporal relations among agents’ local scheduled actions. Combined with pattern matching, plan conflicts due to resource sharing or conflicting conditions (i.e. post-conditions of one action disabling pre-conditions of another action) can be detected. This conflict detection technique is implemented in Sensible Agent Testbed to promote deployment and performance analysis.This paper describes techniques developed for conflict detection during plan integration based on E-PERT diagrams. PERT diagrams were originally developed in 1980s for project management to provide a global consistent view of parallel activities within a project. We E-PERT diagrams for use in the plan integration activity within multi-agent systems. The E-PERT diagram contributes to maintain traceable temporal relations among agents’ local scheduled actions. Combined with pattern matching, plan conflicts due to resource sharing or conflicting conditions (i.e. post-conditions of one action disabling pre-conditions of another action) can be detected. This conflict detection technique is implemented in Sensible Agent Testbed to promote deployment and performance analysis. Technical Report TR99-UT-LIPS-AGENTS-08