Multilevel Multiagent Based Team Decision Fusion for Autonomous Tracking System

Abstract

Multilevel fusion is a key issue for developing the decision-making kernel of multiagent systems. This article presents a formulation of predictive decision-making algorithm for multilevel multiagent based team decision-making system with the I/O mode characterizations of feature in-decision out or data in-decision out methods. The sequential data fusion is conducted through a dynamic behavior modeling method capable of esti- mating the observed system parameters from the raw sensory measurements over period of time. The temporal es- timated model is used to forward prediction of the observed system output for decision-making. A self-evaluation method to estimate the prediction quality is used to generate the individual decision confidence for final decision integration, which is conducted through a multi-layered fuzzy linguistic reasoning engine. The method is imple- mented for an autonomous tracking system that consists of a target tracking agent whose inputs are visual and ultrasonic range measurements and a collision avoidance agent whose inputs are ultrasonic range measurements. The experimental results conducted by a mobile robot and intelligent electrical wheelchair will demonstrate the feasibility, accuracy, and robustness of the system based on the multisensor fusion method.