Towards Large-scale Occupancy Map Building using Dirichlet and Gaussian Processes

This paper proposes a new method for building occupancy maps using Dirichlet and Gaussian processes. We consider occupancy map building as a classification problem and apply Gaussian processes. The main drawback of Gaussian processes, however, is the computational complexity of O(n) related to the matrix inversion, where n is the number of data points. To enable large-scale occupancy map building, we propose to use Dirichlet process mixture models which cluster input data without fixing the number of clusters a priori and to apply a mixture of Gaussian processes for the clustered data. This approach also has an advantage of dealing with local discontinuities better than one global Gaussian process model. Simulation results will be provided demonstrating the benefits of the approach.