Point-Based Value Iteration and Approximately Optimal Dynamic Sensor Selection for Linear-Gaussian Processes

The problem of synthesizing an optimal sensor selection policy is pertinent to a variety engineering applications ranging from event detection autonomous navigation. We consider such synthesis in the context linear-Gaussian systems over infinite time horizon with discounted cost criterion. formulate this terms value iteration continuous space covariance matrices. To obtain computationally tractable solution, we subsequently approximate problem, which solvable through point-based finite “mesh” matrices user-defined bounded trace. provide theoretical guarantees bounding suboptimality policies synthesized method and numerical examples comparing them known results.