Multi-Sensor Fusion Based Localization System for an Amphibious Spherical Robot

Many techniques for robot localization rely on inertial navigation system which suffers from drift and huge computational cost. So, this paper presents a localization method which has a good real-time performance, high-precision and low-cost assumption for a compact amphibious spherical robot. Specifically, this system can navigate the robot along a predefined path without the need for any additional external sensors. Meanwhile, the proposed approach combines various information using extend Kalman filter (EKF), include depth data from a pressure sensor, pose from an Inertial Measurement Unit (IMU), velocity from optical flow and pose estimation from multiple planar markers. A monocular downward facing camera is used to track feature about optical flow and detect artificial landmarks. Moreover, to validate our approach, we conducted experiment in an indoor pool with varying lighting and visibility conditions, and we demonstrate the online localization method is highly accurate, robust and successful application with limited computational capacities and low-cost sensing devices on our compact robot.