DB-VIO: Dual-Branch Visual Inertial Odometry with Enhanced Visual-Inertial Representation

2026-07-24

AI Summary

This work addresses the limitations of existing learning-based visual-inertial odometry (VIO) systems, which suffer from degraded pose estimation accuracy due to the entangled modeling of heterogeneous rotational and translational dynamics, the lack of explicit geometric constraints in monocular vision, and the implicit nature of rotational information in inertial measurements. To overcome these challenges, we propose the first dual-branch learning-based VIO architecture that explicitly decouples rotation and translation estimation. Our approach enhances multimodal fusion through depth-augmented visual features, explicit integration of pose priors, and dedicated temporal modeling for each motion component. Evaluated on KITTI and EuRoC benchmarks, the method outperforms baseline approaches by 20% and 33%, respectively, achieving state-of-the-art performance—most notably reducing rotation error by 65.7% in EuRoC’s high-dynamic sequences.

Abstract

Visual inertial odometry (VIO) is essential for accurate 6-DoF motion estimation in mobile robotic systems. Recent learning-based VIO methods have shown promising progress, but they often rely on unified visual--inertial representations and a single temporal model for full-pose estimation, limiting their ability to capture the heterogeneous dynamics of rotation and translation. Moreover, monocular visual features often lack explicit geometric structure, while raw inertial encoding leaves the underlying rotational kinematics implicit, weakening the rotation-related cues in IMU features. To address these issues, we propose DB-VIO, a dual-branch visual inertial odometry framework with enhanced visual--inertial representation. DB-VIO incorporates depth cues to improve monocular visual perception, injects an explicit integrated-attitude prior to strengthen rotation-aware inertial representation, and decouples pose estimation into dedicated rotational and translational branches for motion-specific temporal modeling. Experiments on autonomous driving and aerial robot benchmarks show that DB-VIO achieves state-of-the-art performance, improving the corresponding baselines by 20\% on KITTI and 33\% on EuRoC. Notably, under the more agile motion patterns of EuRoC, DB-VIO improves the rotational metric by 65.7\% over prior methods. These results demonstrate the effectiveness and generalization of DB-VIO across different platforms and motion scenarios.

Problem

• Visual Inertial Odometry
• Rotation-Translation Decoupling
• Heterogeneous Dynamics
• Monocular Geometry
• IMU Kinematics

Innovation

• dual-branch architecture
• enhanced visual-inertial representation
• explicit attitude prior
• rotation-translation decoupling
• depth-augmented monocular vision