Fusion Strategy for Time Series Prediction and Knowledge based Reasoning for Intelligent Communication

Abstract

This paper describes a predictive decision making strategy for “Intelligent Communication” in Cellular Robotic Systems (CEBOT). The CEBOT is an autonomous distributed robotic system composed of a number of robotic units called “cell” with limited functions, which configures its optimal structures in terms of hardware and software, according to the task or environment. Intelligent Communication aims at communication reduction through intention reasoning. It regards communication as not only data transmission, but total intelligent activities supposing interaction with other agents. Intelligent Communication synthesizes learning, reasoning and prediction. Additional learning and knowledge acquisition enables more precise inference, and consequently contributes reduction of excessive communication. Since the real world is dynamical, time series analysis will be required and also it will give beneficial clues for intention reasoning to the other robot behavior. However, time series data observed by robot behaviors are nonstationary in general. Because robot behaves based on the planning procedure to execute its tasks, or it might be generated as a result of adaptive motion to the current condition and environment, such as collision avoidance. In order to consider both the trend of state transition and static rules that affect behavior of the robot, we will propose a strategy to synthesize time series prediction and rule / knowledge based reasoning.