Elke Steegmans

TOWARDS ACTIVE PERCEPTION IN SITUATED MULTI-AGENT SYSTEMS

Modeling the environment and agent-environment relationships is not well explored in multi-agent systems, in particular not for software multi-agent systems. This paper aims to contribute with a generic model for active perception in situated multi-agent systems. Active perception enables an agent to direct its perception at the most relevant aspects in the environment, according to its current task. The model decomposes perception into three functionalities: sensing, interpreting, and filtering. The agent first senses its neighborhood through a set of selected foci, resulting in a representation. A set of perceptual laws enforces domain specific constraints on sensing. Next, the agent interprets the representation by means of descriptions, resulting in a percept. Percepts are expressions that can be understood by the internal machinery of the agent. Finally, the percept is filtered by a set of selected filters, restricting the perceived data according to specific context relevant selection criteria.

Self-organising in multi-agent coordination and control using stigmergy

In order to cope with todays dynamic environment, the described manufacturing control system is designed as a self-organising multi-agent system. The design of this novel system implements the PROSA reference architecture [1]. Coordination among agents is done indirectly through a pheromone-based dissipative field as is done by social insects in coordinating their behaviour. In this case, our agents act as social insects interpreting the pheromones put by the others in the environment. This control system is built from the basic elements of any manufacturing controller, namely products, resources and orders. However, the overall control system is constructed not only from those basic elements but also employing the appropriate interaction patterns among the agents who represent them. For coordination purposes, the agents send out a kind of mobile agents - artificial ants - to lay down information on the environment. In our case, where fulfilling the manufacturing orders is the main concern, there are at least 3 types of ant in this system: (1) feasibility ants - to propagate information concerning the feasible finishing routes; (2) exploring ants - to explore the feasible routes; and (3) intention ants - to propagate the route preferences. The overall mechanism enables the system to exhibit a self-organising behaviour.

A design process for adaptive behavior of situated agents

Engineering non-trivial open multi-agent systems is a challenging task. Our research focusses on situated multi-agent systems, i.e. systems in which agents are explicitly placed in an environment which agents can perceive and in which they can act. Situated agents do not use long-term planning to decide what action sequence should be executed, but select actions based on the locally perceived state of the world and limited internal state. To cope with change and dynamism of the system, situated agents must be able to adapt their behavior. A well-known family of agent architectures for adaptive behavior are free-flow architectures. However, building a free-flow architecture based on an analysis of the problem domain is a quasi-impossible job for non-trivial agents. To tackle the complexity of designing adaptive agent behavior based on a free-flow architecture, suitable abstractions are needed to describe and structure the agent behavior. The abstraction of a role is obviously essential in this respect. A modeling language is needed as well to model the behavior of the agents. We propose a statechart modeling language to support the design of roles for situated agents. In this paper we describe a design process for adaptive behavior of situated agents as part of a multi-agent oriented methodology. The design process integrates the abstraction of a role with a free-flow architecture. Starting from the results of analysis of the problem domain, the designer incrementally refines the model of the agent behavior. The resulting class diagram serves as a basis for implementation. We illustrate the subsequent design steps with a case study on controlling a collection of automated guided vehicles.

Protocol Based Communication for Situated Multi-Agent Systems

In this paper we introduce a model for direct communication in situated multi-agent systems. Direct communication is typically associated with cognitive agents, where the information encoded in the messages is related to a mental state. This generally assumed view on communication however, does not fit the approach of situated, behavior-based agents. We propose a protocol-based communication model for situated agents. Communication specified in terms of protocols, i.e. well-defined sequences of messages, shifts the focus of communication from the reasoning upon messages towards the relationship between the exchanged messages. The model decomposes communication into three functional modules: message decoding, communicating and message encoding. The core of the model, the communicating module (1) interprets decoded messages and reacts to them in accordance with the applicable protocol, and (2) initiates or continues conversations when the conditions imposed by the applicable protocol are satisfied.

An agent design method promoting separation between computation and coordination

The development of (internet) agents is often a tedious and error-prone task resulting in poorly reusable designs, since both the internal computation of the agent as well as the coordination support are developed in an ad hoc fashion. To improve the process of agent-oriented software development, we propose an agent design method that imposes the separation of internal computation from coordination aspects. This method comprises two dimensions: a design formalism and an agent design process. As an illustration of the presented method, we present the design of an internet agent that is entitled to deploy a distributed service in a computer network, without breaking the consistency of that network. The presented design method has resulted in the development of ACF (Agent Composition Framework), a component framework to build flexible internet agents. We argue that the presented design method combined with this infrastructure can promote a modular and easy to manage approach to the design and development of internet agent applications.

Integrating free-flow architectures with role models based on statecharts

Engineering non-trivial open multi-agent systems is a challenging task. Our research focusses on situated multi-agent systems, i.e. systems in which agents are explicitly placed in a context – an environment – which agents can perceive and in which they can act. Two concerns are essential in developing such open systems. First, the agents must be adaptive in order to exhibit suitable behavior in changing circumstances of the system: new agents may join the system, others may leave, the environment may change, e.g. its topology or its characteristics such as throughput and visibility. A well-known family of agent architectures for adaptive behavior are free-flow architectures. However, building a free-flow architecture based on an analysis of the problem domain is a quasi-impossible job for non-trivial agents. Second, multi-agent systems developers as software engineers require suitable abstractions for describing and structuring agent behavior. The abstraction of a role obviously is essential in this respect. Earlier, we proposed statecharts as a formalism to describe roles. Although this allows application developers to describe roles comfortably, the formalism supports rigid behavior only, and hampers adaptive behavior in changing environments. In this paper we describe how a synergy can be reached between free-flow architectures and statechart models in order to combine the best of both worlds: adaptivity and suitable abstractions. We illustrate the result through a case study on controlling a collection of automated guided vehicles (AGVs), which is the subject of an industrial project.

Towards commitments for situated agents

Traditional architectures for situated, behavior-based agents take the viewpoint of the individual agent to select the most appropriate action. Action selection is typically based on internal stimuli and stimuli from the agents neighboring environment. As such collaborations between agents have to emerge from the individually selected actions of the agents. In this paper we study the research problem of how to enable explicit collaborations between situated agents. Explicit collaborations are reflected in mutual commitments. Contrary to the traditional approaches of commitment that are based on the mutually dependent menial states of the involved agents and a goal-oriented plan, we introduce the notion of a situated commitment that is based on the roles of the involved agents and the local context they are placed in. Activating mutual situated commitments in a collaboration results in more consistent behavior of the agents towards their commitments. The proposed approach fits the general principles of situatedness and robustness of situated multi-agent systems.

A Nested Layered Threshold Model for Dynamic Task Allocation

Dynamic task allocation is an essential aspect in modeling ant behaviour [1],[2]. In this paper, we propose a new (abstract) model for dynamic task allocation of agents that combines a nested layered architecture with a threshold mechanism. We apply this new model to achieve flexible ant behaviour.

A Basic Taxonomy for Role Composition

Roles are the basic building blocks for defining the behavior of agents in multi-agent systems. Agents typically perform several roles. In this paper, we describe analysis and design issues in defining agents as compositions of roles. In short, specifying the behavior of an agent entails in essence two issues: which roles are assigned to a particular agent, and how does an agent select a role in a particular situation. Both issues can be decided upon either by the designer (i.e. at design time) or by the agent (i.e. at run-time). This paper describes a basic taxonomy for role composition based on both issues, and illustrates the different composition approaches using a case study in the domain of manufacturing control.