ADM-Fusion: Adaptive Deep Multi-Sensor Fusion for Robust Ego-Motion Estimation in Diverse Conditions

2026-06-23

AI Summary

This work addresses the challenge of robust ego-motion estimation in scenarios where sensor reliability fluctuates significantly due to environmental degradation. To overcome the limitations of conventional fixed-weight fusion approaches, the authors propose an end-to-end deep multi-sensor fusion framework featuring an adaptive Mixture-of-Experts architecture with a content-aware routing mechanism that dynamically allocates sensor weights. The framework incorporates decoupled translation and rotation branches, enhanced by a cross-task attention module that facilitates information sharing while preserving task-specific characteristics. Evaluated on both the CARLA-LOC simulation benchmark and the real-world KITTI dataset, the method demonstrates strong robustness under sensor degradation, achieving accuracy comparable to or better than state-of-the-art approaches, and enables effective sim-to-real transfer.

Abstract

Robust multi-sensor fusion is essential for reliable autonomy in diverse and degraded environments, where sensor reliability can fluctuate rapidly. Because different modalities fail in distinct ways, effective fusion should adaptively balance complementary cues rather than rely on fixed weighting. This adaptability is particularly important for ego-motion estimation, since accurate updates depend on the consistent integration of complementary sensor information. We propose ADM-Fusion, an end-to-end deep learning based multi-sensor fusion method designed to adapt to environmental changes and sensor degradation. ADM-Fusion employs an adaptive sensor mixture-of-experts framework with content-aware routing to dynamically assign weights to sensor inputs in real time. The system further incorporates separate translation and rotation branches, coupled through a cross-task attention mechanism to preserve task-specific specialization while enabling information sharing. ADM-Fusion is trained on the CARLA-LOC simulated dataset and subsequently fine-tuned on KITTI real-world data, demonstrating effective simulation-to-real transfer. Experiments show that ADM-Fusion remains robust under degraded conditions while maintaining competitive performance against existing methods.

Problem

• ego-motion estimation
• multi-sensor fusion
• sensor degradation
• adaptive fusion
• robust autonomy

Innovation

• adaptive fusion
• mixture-of-experts
• ego-motion estimation
• cross-task attention
• simulation-to-real transfer