Learning Hyperparameters in Efficient Spatial Model by Robotic Sensors

Recently, a new class of spatial models over a continuum domain that builds on hidden Gaussian Markov Random Fields (GMRFs) was proposed for resource-constrained networked mobile robots dealing with non-stationary physical processes. The hidden GMRF was realized with respect to a proximity graph over a surveillance region. In this paper, we investigate learning strategies based on the maximum likelihood (ML) and the maximum a posteriori (MAP) estimators to find the locational generating points for the spatial model so that mobile robots can efficiently make the prediction. Some promising simulation results and future research directions are discussed.