Multimodal Sensor Fusion in Autonomous Vehicles: Technologies, Architectures, and Open Challenges

Area of Science:

• Robotics and Artificial Intelligence
• Automotive Engineering
• Sensor Technology

Background:

• Advancements in sensing, AI, and computing drive autonomous vehicle development.
• Reliable environmental perception is a key challenge for higher-level driving automation.
• Existing research spans various sensor modalities and fusion techniques.

Purpose of the Study:

• To systematically review multimodal sensor technologies and fusion architectures for autonomous driving.
• To analyze sensor characteristics, fusion strategies, and performance under diverse conditions.
• To identify emerging research directions and future challenges in autonomous perception.

Main Methods:

• Systematic literature review using PRISMA guidelines.
• Analysis of 66 peer-reviewed studies (2014-2025) on autonomous driving sensors and fusion.
• Synthesis of evidence on sensor modalities, fusion architectures, safety, and computational constraints.

Main Results:

• Multimodal sensor fusion significantly improves perception robustness and decision reliability.
• Fusion strategies (early, mid, high-level, transformer-based) enhance scalability and fail-operational capabilities.
• Performance under adverse conditions and adherence to safety standards (ISO 26262, SOTIF) are critical.

Conclusions:

• Multimodal sensor fusion is essential for scalable, robust, and certifiable autonomous driving systems, especially for Levels 4-5.
• Future research should prioritize uncertainty-aware fusion, explainable AI, real-world validation, and hardware-software co-design.
• Addressing real-time constraints and safety frameworks is crucial for advancing autonomous vehicle autonomy.