Study on sociotype communication in heterogeneous multi-agents environment

Abstract

Heterogeneous multi-agent environment involve a problem space which house more than

1 type of agents. Where multi-agent is concerned, communication has always been the key

thought process. This is because different agents have different architectures and in order

for them to collaborate within the same environment, there should be some form of

communication involved. This paper will examine the communication process of 2 Neural

Network intelligent agents, one is the Multilayer Perceptron (MLP) and the other is the

Temporal Difference-Fusion Architecture for Learning and Cognition (FALCON) with

the help of a grid-based path-finding environment, known as Mine Navigation.

The experiment was broken into 2 groups, each with 1 FALCON and 1 MLP agent. The

first group being the control group consists of the agents with no communication; while

the other, the test group is made up of agents capable of communication. The results from

the two groups were collected and compared to draw out the relationship between

communication and the performance of these agents.

The results show that with communication, FALCON agent is able to perform better under

controlled scenario and the MLP agent is able to be trained faster. With the success of this

experiment, future work may consider increasing the complexity of the problem space by

introducing more dimensions, or alternating the agents of different artificial and

computational intelligence.