Encoded depth Panorama based Information flow in a multi agent system

Abstract

Multiagent Systems is a research area of distributed Artificial Intelligence where agents are used for modelling any entity like web service, robot, or nodes of a grid for example. These artificial agents require to communicate and coordinate efficiently; retrieving, processing and exchanging expertise with other agents in-order to solve real time complex problems that cannot be solved individually. Research in this domain has a two-fold focus. One end concentrates on developing architectures of MAS while the other end works with the communication language between the agents. This usually requires a common language, also known as an Agent Communication Language (ACL). Although ACLs were popular in software agent systems, their use was limited in robot systems where low level data like video transfer was necessary. This continued until the introduction of backchannel for transferring the low level data. Different MAS developed and discussed in literature considers target specific data processing and communication mechanisms. The thesis addresses part of a research problem which deals with a heterogeneous system with two agents used for high level parsing of the environment. The work in its entirety involves Agent 1 as a robot equipped with a pan-tilt camera, seven IR sensors , three ultrasonic sensors and a SBC. This agent is responsible for acquiring environmental information using the sensors, processing them and sending them to Agent 2, which is a computer where semantic parsing on the received data is performed. This thesis work has two main objectives. Primarily it aims to reduce data flow between the agents for efficient usage in terms of network bandwidth and storage requirements. This is done by exploiting a encoded panoramic mosaic to reduce visual data dimension communicated between agents in real time. Secondly, it proposes a model which can be utilized to recover approximate depth information using a pair of overlapping images from a single camera.