Online High-Precision Probabilistic Localization of Robotic Fish Using Visual and Inertial Cues

This paper focuses on the development of an online high-precision probabilistic localization approach for the miniature underwater robots equipped with limited computational capacities and low-cost sensing devices. The localization system takes Monte Carlo localization (MCL) as the main framework and utilizes the onboard camera and low-cost inertial measurement unit (IMU) for information acquisition to provide a decimeter-level precision with 5-Hz refreshing rate in a small space with several artificial landmarks. Specifically, a novel underwater image processing algorithm is introduced to improve the underwater image quality; two key parameters, including a distance factor and an angle factor, are finally calculated to serve as the criteria to MCL. Meanwhile, the accurate orientation and rough odometry of the robot are acquired by onboard IMU. Moreover, a Kalman filter is adopted to filter the key information extracted from the sensors´ data processing. In principle, when visual and inertial cues are both obtained, visual information with higher reliability has the priority to be used in the algorithm, which finally results in rapid convergence to the real pose of the robot. A series of relevant experiments are systematically conducted on the robotic fish, which prove that the online localization algorithm herein is highly accurate, robust, and practical for the miniature underwater robots with limited computational resources.