Robust Visual Odometry On $\text{SE}(3)$: Design and Verification

Abstract

Visual odometry is a crucial technique for estimating a robotic vehicle's trajectory by analyzing images captured by its onboard camera when the vehicle's attitude cannot be retrieved. However, uncertainties such as modeling errors, measurement noise, misidentification of feature marks, and the switching output arising from visual geometric constraints can all hinder accurate estimations. To address these challenges, this article proposes a robust visual odometry that can be implemented in a sampled-data structure. Comprehensive simulations and experiments are conducted to demonstrate the effectiveness of the proposed design and to explore the relationship between design parameters and estimation performance. In addition, tuning guidelines for visual odometry parameters are provided to help address these uncertainties effectively.