A framework for multisensor data fusion

This paper describes a multi-agent approach for multisensor data fusion (MSDF). We present a theoretical framework for an autonomous mobile robot system that operates in an industrial environment. The robot can be viewed as an intelligent platform using sensors to detect obstacles, build a map of the workplace and navigate to achieve manufacturing tasks opportunistically. Our work focuses on a MSDF system for map building and updating of a structured environment, extendable to dynamical situations. The multi-agent architecture enables problem solving by processes that perform their actions independently. Two agents communicate either by sharing information if they solve a common subproblem, or by sending messages if they need to use data computed elsewhere. Perceptual agents deal with raw data acquisition and are integrated in different sets related to the sensors used, while interpretation agents deal with maintaining a symbolic model of the environment resulting from the combination of information processed from different perceptual sets. The system is interfaced with a Nomad 200 mobile robot and can interact with an external motion planning system by message transmission.