Autonomous Quadrotor 3D Mapping and Exploration Using Exact Occupancy Probabilities

Abstract

This paper deals with the aerial exploration for an unknown three-dimensional environment, where Bayesian probabilistic mapping is integrated with a stochastic motion planning scheme to minimize the map uncertainties in an optimal fashion. We utilize the popular occupancy grid mapping representation, with the goal of determining occupancy probabilities of evenly-spaced grid cells in 3D with sensor fusion from multiple depth sensors with realistic sensor capabilities. The 3D exploration problem is decomposed into 3D mapping and 2D motion planning for efficient real-time implementation. This is achieved by projecting important aspects of the 3D map onto 2D maps, where a predicted level of map uncertainty, known as Shannon's entropy, provides an exploration policy that governs robotic motion. Both mapping and exploration algorithms are demonstrated with both numerical simulations and quadrotor flight experiments.