WING: Wheel-Inertial Neural Odometry with Ground Manifold Constraints

TL;DR

This paper introduces WING, a neural odometry system for ground robots that leverages ground manifold constraints and neural corrections to improve pose estimation accuracy in challenging environments without relying on external sensors.

Contribution

The paper presents a novel interoceptive odometry system using neural networks and ground manifold constraints, enhancing accuracy and robustness in degraded scenarios.

Findings

Outperforms state-of-the-art learning-based interoceptive odometry methods.

Effectively reduces drift in IMU and wheel encoder measurements.

Utilizes ground manifold soft constraints for improved pose estimation.

Abstract

In this paper, we propose an interoceptive-only odometry system for ground robots with neural network processing and soft constraints based on the assumption of a globally continuous ground manifold. Exteroceptive sensors such as cameras, GPS and LiDAR may encounter difficulties in scenarios with poor illumination, indoor environments, dusty areas and straight tunnels. Therefore, improving the pose estimation accuracy only using interoceptive sensors is important to enhance the reliability of navigation system even in degrading scenarios mentioned above. However, interoceptive sensors like IMU and wheel encoders suffer from large drift due to noisy measurements. To overcome these challenges, the proposed system trains deep neural networks to correct the measurements from IMU and wheel encoders, while considering their uncertainty. Moreover, because ground robots can only travel on the ground, we model the ground surface as a globally continuous manifold using a dual cubic B-spline manifold to further improve the estimation accuracy by this soft constraint. A novel space-based sliding-window filtering framework is proposed to fully exploit the $C^2$ continuity of ground manifold soft constraints and fuse all the information from raw measurements and neural networks in a yaw-independent attitude convention. Extensive experiments demonstrate that our proposed approach can outperform state-of-the-art learning-based interoceptive-only odometry methods.