Decentralized Estimation and Control for Multisensor Systems

Explores the problem of developing scalable, decentralized estimation and control algorithms for linear and nonlinear multisensor systems. Such algorithms have extensive applications in modular robotics and complex or large scale systems, including the Mars Rover, the Mir station, and Space Shuttle Columbia.

Most existing algorithms use some form of hierarchical or centralized structure for data gathering and processing. In contrast, in a fully decentralized system, all information is processed locally. A decentralized data fusion system includes a network of sensor nodes - each with its own processing facility, which together do not require any central processing or central communication facility. Only node-to-node communication and local system knowledge are permitted.

Algorithms for decentralized data fusion systems based on the linear information filter have been developed, obtaining decentrally the same results as those in a conventional centralized data fusion system. However, these algorithms are limited, indicating that existing decentralized data fusion algorithms have limited scalability and are wasteful of communications and computation resources.

Decentralized Estimation and Control for
Multisensor Systems aims to remove current limitations in decentralized data fusion algorithms and to extend the decentralized principle to problems involving local control and actuation.

The text discusses:

Generalizing the linear Information filter to the problem of estimation for nonlinear systems

Developing a decentralized form of the algorithm

Solving the problem of fully connected topologies by using generalized model distribution where the nodal system involves only locally relevant states

Reducing computational requirements by using smaller local model sizes

Defining internodal communication

Developing estima