Development and Experimental Validation of Visual-Inertial Navigation for Auto-Docking of Unmanned Surface Vehicles

Abstract

Docking is a safety-critical operation for autonomous surface vehicles and requires highly accurate navigation signals. Since Global Navigation Satellite Systems (GNSS) can be unreliable and inaccurate in urban environments, other sensors should be considered for increased redundancy and reliability. To this end, we present a low-cost visual-inertial navigation system that we can use for the automatic docking of small vehicles. The proposed system produces state estimates of the vehicle, including position, velocity, and attitude, based on raw image and inertial data. To simplify the navigation task, we use easily identifiable tags as a reference on the dockside. When the vehicle approaches the dock, a visual fiducial system recognizes the tags and estimates the relative pose between the camera onboard the vehicle and the tags at the dockside. The camera-tag pose and inertial data are then fused using an error-state Kalman filter for robust state estimation of the vehicle. For benchmarking, we use an unmanned surface vehicle equipped with a dual-antenna real-time kinematic GNSS receiver for accurate positioning and heading. We show that the proposed method performs well on regular and adverse weather data. Finally, we demonstrate that the proposed method performs well in feedback control through field experiments and can supplement traditional navigation systems for docking operations.

Development and Experimental Validation of Visual-Inertial Navigation for Auto-Docking of Unmanned Surface Vehicles