Encoded Scene Panorama Based Information Flow in a Multi Agent System

Abstract

The work presents a multi agent system using backchannels to interact and send environmental information in real time. The performance of the agent system in real time depends on its abilities to cooperate and communicate with each other. This has seen the evolvement of different architectures as well as Agent Communication Languages over time. The amount of data transfer and speed of processing are the two factors which control the real time response of Multi Agent Systems (MAS). Increase in the amount of information communicated between the agents will lead to improved functionality but at the cost of processing and communication. This paper utilizes an approach where an encoded panoramic mosaic is exploited to reduce visual data dimension communicated between agents in real time. Instead of transferring video data in bulk, the sender agent processes the frames to create a panorama. The sender updates the panorama only when there are changes in the scene. A single panorama hence encompasses the full 360 degree view at the $t_0$ instant as well as the dynamic information of the next $t_1$ to $t_n$ instances. The value for n is set according to the amount of dynamic content of the scene. The panorama is refreshed when the Agent changes position. This reduces the data dimension transferred to other agents, which reconstructs the encoded panorama into individual image frames from $t_1$ to $t_n$. The amount of transfer is trimmed from $k \times n$ (k images per panorama and n panoramas) to a single panorama. Also if k images have a dimension $m\times n\times k$, a single panorama can be mapped as $\frac{m.k}{2} \times \frac{n.k}{2}$. The value of k depicts the number of pans used to acquire the 360 degree view and depends on the scene. The model is highly effective when there are minimal changes in the surrounding scene at fixed intervals.