Visual inertial odometry for mobile robotics

Abstract

This work targets the development of a real-time algorithm to track the pose of a mobile robot equipped with an inertial measurement unit and a monocular camera. The method proposed is the Multi-State Constraint Kalman Filter, developed by A. Mourikis and S. Roumeliotis. Its main strength is that it uses the geometric constraints of the features observed without adding them to the state (as in SLAM). The feature processing is delayed until they are out of view, which avoids the usual problems associated to feature initialization. The cost of the algorithm is linear in the number of features (which can be controlled by the detector), which makes it suitable for real-time execution. The method follows the Kalman filtering framework in its error-state variant, which estimates the accumulated error instead of the truestate. The filter is updated with the feature positions, which are computed using Gauss-Newton optimization using the inverse-depth parameterization.