Bayesian Learning of Occupancy Grids

Information-Theoretic Interactive Sensing and Inference for Autonomous Systems

“…An illustration of the proposed active perception problem is given in Figure 1. The search area is discritized into grid cells where the knowledge about object locations in the environment is captured through an occupancy grid estimation process [9]. The knowledge about the class of each detected and localized object is then provided in the form of a Dirichlet-Categorical model (DCM) [10].…”

“…(1) Generating a map of the scene through an occupancy grid estimation process, which produces a set of marginal posterior probabilities that any one point in an area is occupied [9]. (2) Classifying occupied regions through a sequential classification process, which produces a set of marginal posterior probabilities of class membership for each occupied region.…”

“…Occupancy grid estimation is a popular process for generating an occupancy map of an area given a set of measurements taken from that area [7]- [9]. The map is partitioned into a set of B disjoint grid cells {g i } B i=1 , all with the same shape and size.…”

“…Using the following occupancy grid formulation presented in [9], these marginal posterior probabilities at time step S can be expressed as…”

“…The first cost function is associated with object detection and localization, and measures the mutual information between the occupancy state variable for a single grid cell and a binary measurement random variable. Solving for its closed-form representation relies on the measurement model and posterior occupancy distributions produced through the occupancy grid estimation process presented in [9]. The second cost function associated with the classification of detected objects measures the mutual information between a class state variable for a single grid cell and random variable that is the parameter to the class state variable distribution.…”