Visual inertial perception for autonomous inspection in nuclear power plants

Abstract

We propose a perceptual system enabling visual inertial navigation for autonomous inspection inside a nuclear power plant. The system incorporates robust state estimation using a stereo-camera pair and an Inertial Measurement Unit (IMU) and generates global consistent dense maps which are then used for autonomous navigation. The system is robust to the challenges posed by the nuclear power plant environment such as varying degrees of complexity, varying lighting conditions, clutter and constrained spaces. Efficient representation of and inference from the dense maps minimizes the amount of energy required by the perceptual system thus making it amenable to robots with limited computational resources.