Autonomous exploration of visually-degraded environments using aerial robots

Abstract

This paper presents a combined perception systems and planning algorithms approach to the problem of autonomous aerial robotic navigation and exploration in degraded visual (dark) GPS-denied environments. A perception system that comprises of a synchronized near-infrared stereo camera system, flashing LEDs, inertial sensors and a 3D depth sensor is utilized in order to derive visual-inertial odometry and dense mapping in conditions of complete darkness. Exploiting this ability within the framework of a localizability-aware receding horizon exploration and mapping planner, the proposed approach ensures robotic autonomy in dark environments for which no prior knowledge exists. A set of experimental studies in a dark room of complex geometry, as well as a city tunnel at night were conducted to evaluate and verify the abilities of the system and the proposed solution.