Efficiency Through Communication In A Multiple Agent System

Efficiency Through Communication in a Multiple-Agent System

Abstract

If a fully autonomous multi robotic system were expected to operate freely in the world to accomplish any task, the robots must be prepared for the enormous variety of situations that they would encounter on a constant basis. Even though the task of preparing robots for these encounters seems unsurmountable at first glance, there are ways to reduce the size of the problem. The use of communication in a multi agent system is one such way of reducing the complexity and scope of the problem. In this paper, I will give a brief survey of the state of multi agent communication to date. I will cover issues such as direct and indirect communication forms, visual and verbal communication, and communication evolution. I will also briefly cover the topic of Agent Communication Languages and how they may help us to minimize the growing pains of agent communication and information sharing. Finally, I will discuss communication decisions such as what to communicate, who to communicate with, when to communicate and, probably the most important part of communication, the cost of communication. Communication cost can make or break a system, depending on what system resources are available to use in the communication process as well as what resources are needed receive and process a message.

1. Introduction

Over the last twenty years, research in topics such as autonomous robot navigation, has shifted from single robot systems to multi robot systems. This shift has occurred mostly because single robot systems have limitations to what tasks they can perform.

Tasks such as object retrieval and large area survey through navigation can both be done by single unit systems, but the use of a multi-unit system will, in most cases, improve the efficiency to which the task can be completed.

There are several areas in which one could focus their attention when it comes to multi-agent systems. For the purpose of this paper, I have chosen to focus on communication between agents and how its use can influence task completion.

When we think about agent communication, it is important that we do not confine ourselves to the realm of human communication. There are several forms of communication that can be used in the artificial sense, but all forms of communication fall into two distinct groups, direct communication and indirect communication.

Direct communication is communication in which the message directly affects the sensors and possibly the behavior of another unit (McPartland, 2005). .

Indirect communication is communication in which the environment is altered by the behavior of one unit, which may eventually cause another unit to modify its own behavior to compensate for the new state of the environment (McPartland, 2005).

Designing a system that uses communication to improve efficiency must be done carefully. We must consider things such as the cost of using communication, when to use communication, and what to communicate. Answering any one of these questions may be difficult, as the answer could possibly depend on how the others are answered.

Ultimately, the efficiency of a system is determined by its ability to coordinate each individual agent's actions in a manner such that the task is completed correctly and as quickly as possible. This makes communication a crucial part of the equation.

On its own, a single agent has a limited view of its environment. The process of gathering information about the environment and sharing that information with other agents could significantly widen the view an agent has of its environment. Improved knowledge of one's surroundings can several implications on efficiency of task completion.

In this paper, I will discuss the outcome of several experiments in which multi-agent systems were assigned tasks. One system used direct communication, another used indirect communication, and the remaining system used no communication at all.

2. Communication

In the human realm, communication is typically described as the exchange of information between two people. Humans use verbal, written, and visual forms of communication to convey their intentions. These same principles can be used in robotics to enhance task completion efficiency. Some of the forms of communication appear to work better than others when it comes to artificially intelligent agents.

2.1 Direct Communication

The use of visual communication has been shown to have little effect on task completion. As shown in several experiments with MinDART, the use of communication through visual means shows little effect on the amount of time it took to complete a task. Instead of completing the task more quickly, which is what the hypothesis was, they found that the use of communication reduced the variance in completion times. This was attributed to robots spending less time wandering aimlessly around the course (Rybski, 2005).

Visual communication is a logical next step in the study of multi-agent communication because it is fairly simple to implement. In the MinDART experiments, a simple search and retrieve task was assigned to a team of robots. The targets were fixed with lighted beacons which were visible to each robot when they got within a predefined distance.

When a robot found a "beacon", it would start signaling others by flashing its own light for a certain amount of time. When the time had expired the robot tried to retrieve the target on its own.

This form of communication would be described as direct communication. One unit discovers some new and important information about its environment and tries to communicate that information to others. Though this method could work in many instances, it tends to be inefficient due to a lack of information.

In the MinDART experiments, it sometimes caused robots that were near a target of their own, to turn around and move towards the robot that had found a target (Rybski, 2004). Also, in some instances, the timer on the signaling robot would expire. This meant that the signaling robot would turn off its signal when the second robot had only covered part of the distance. The second robot would then begin to search for a new target, which would add to the time it took to complete its task.

The use of verbal communication works well for humans but may not be practical for robots. Most languages humans use to communicate have inherent ambiguities. When we speak to one another, we often need to decode the messages we get. Robots would operate in much the same way.

Robots would