Visual-Inertial-Wheel Odometry With Wheel-Aided Maximum-a-Posteriori Initialization for Ground Robots

摘要

In recent years, Visual-Inertial Odometry (VIO) has demonstrated remarkable results using low-cost and complementary sensors. However, these methods often encounter initialization failure and suffer reduced robustness or low trajectory accuracy under challenging scenarios. In this letter, we propose a visual-inertial-wheel odometry that provides robust and accurate initialization and high-accuracy estimates for ground robots. We propose a novel Maximum-a-Posteriori (MAP) initialization, coupled with wheel encoder measurements, to address the unobservable scales of the visual-inertial-only initialization when the vehicle moves straight at the beginning. Moreover, we leverage the wheel readings to construct an inertial-wheel joint propagation and optimization, leading to robust pose tracking when encountering weak visual observations. Additionally, we incorporate wheel measurements and planar constraints into the local and global Bundle Adjustment (BA) for positioning, significantly enhancing localization accuracy when the robot moves on an approximately planar ground. Experiments on the public datasets demonstrate the effectiveness and efficiency of our initialization, the robustness of pose tracking, and the improved accuracy of the entire trajectory.