DAP-Pose: Deep Temporal Alignment and Physics-aware Cross-modal Sensor Fusion for Robust Pose Estimation

2026-07-26

AI Summary

This work addresses the challenges of low accuracy and poor robustness in pose estimation within complex environments, where multimodal sensors suffer from asynchrony and a lack of physical constraints. To overcome these limitations, the authors propose DAP-Pose, an end-to-end model that introduces a novel bilateral cross-modal fusion (BCF) mechanism and deep temporal alignment (DTA) to effectively integrate visual, inertial, and GNSS measurements without requiring hardware synchronization. Furthermore, the method incorporates physically informed constraints derived from manifold geometry and GNSS absolute scale to significantly suppress drift. Evaluated on the KITTI dataset, the approach achieves a relative translation error as low as 1.31% and a rotation error of merely 0.46°, demonstrating robust performance even under severe temporal misalignment.

Abstract

Robust and accurate pose estimation with multi-modal sensors is fundamental for autonomous vehicles and mobile robotic systems in complex environments. In this paper, we propose DAP-Pose, a unified end-to-end model for robust multi-modal pose estimation. DAP-Pose introduces a Bi-level Cross-modal Fusion (BCF) module that captures complementary semantic and geometric motion cues from visual, inertial, and GNSS measurements. To handle temporal offsets, we designed a Deep Temporal Alignment (DTA) module that explicitly aligns asynchronous streams in latent space, enabling coherent motion modeling without strict hardware synchronization. Furthermore, we incorporate physics-aware constraints via manifold geometry and GNSS-guided absolute metric scale, enforcing motion consistency and mitigating drift. Experiments upon the public KITTI benchmark dataset were conducted to evaluate the performance of DAP-Pose against existing methods. DAP-Pose achieved the state-of-the-art performance, with the lowest average translation error ($t_{rel}$) of 1.31% and rotation error ($r_{rel}$) of 0.46$^{\circ}$. Furthermore, it accurately estimates poses and maintains robust performance under severe artificially injected temporal misalignment.

Problem

• pose estimation
• multi-modal sensor fusion
• temporal misalignment
• autonomous systems
• sensor synchronization

Innovation

• Cross-modal Fusion
• Temporal Alignment
• Physics-aware Constraints
• Pose Estimation
• Multi-modal Sensor Fusion