Navigation in the Service of Enhanced Pose Estimation

Abstract

This paper addresses robust vision-based odometry for underwater robotics by autonomously adjusting the robot trajectory in real-time to optimize the quality of ongoing visual feedback. It is well-known that accurate visual odometry depends on both the presence of sufficient smooth surfaces with manageable reflectance functions and the availability of detectable rigid structures or appearance variations. More generally, some locations in the world are suited to good visual odometry performance while others are not. Our system finds trajectories that pass over such usable visual content by evaluating a localization quality metric (quality score) on points from a forward-facing camera’s image. By planning forward-looking paths that optimize an associated quality score, a downward-facing camera running visual odometry is able to localize our robot more accurately. While our approach is based on a particular underwater vehicle and imaging arrangement, its essential elements can be generalized readily.

Publication
International Symposium on Experimental Robotics (ISER)