Implementing sensor fusion using a cost-based approach

A suite of heterogeneous sensors and their controlling agents which are measuring the dynamics of a plant are considered. The sensor agents interact in real time via a communications network. The problem is to find an unbiased estimate of the plant state given that the data available is dynamic, noisy, and given in a multiplicity of representations. The approach proposed is unique because it does not attempt to transform the data to a common representation. Rather we establish a framework which we call the multiple agent hybrid estimation architecture (MAHEA) in which we allow heterogeneous data to flow between agents to improve their estimates of plant state. We give a brief review of Kohn-Nerode extraction procedure for hybrid systems. We show how we can construct a special optimization criterion for a plant estimation optimization problem, the estimation Lagrangian. The type of Lagrangian that we need to construct is special for the plant estimation problem in that we want the Lagrangian to be 0 at each point where the agent has reached the desired estimate of the plant. Such Lagrangians are not suitable for most control problems. Finally, we describe our mechanism to solve the problems of agent synchronization and of how agents with different models can produce coherent and compatible estimates of the plant. Our solutions to these problems use the Noether invariance conditions in a novel way.