Deep Learning Sensor Fusion for Autonomous Vehicle Perception and Localization: A Review

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision
• Transportation Engineering

Background:

• Autonomous vehicles (AV) promise to revolutionize ground transportation by enabling smart vehicles to perform driving tasks independently.
• Advancements in sensor technology and communication (e.g., 5G) are crucial for AV perception, encompassing both local and extended environmental awareness.
• Sensor reliability is a challenge, as individual sensors can fail due to various factors, necessitating a multi-sensor approach.

Purpose of the Study:

• To provide a comprehensive review of state-of-the-art methods for enhancing AV system performance, particularly in short-range environments.
• To focus on recent deep learning sensor fusion algorithms applied to AV perception, localization, and mapping.
• To identify current trends and future research directions in AV sensor fusion.

Main Methods:

• Review of recent studies on deep learning sensor fusion algorithms for autonomous driving.
• Analysis of methods improving short-range perception in AV systems.
• Exploration of sensor fusion techniques for localization and mapping.

Main Results:

• Sensor fusion, particularly using deep learning, is critical for overcoming individual sensor limitations in AVs.
• Synergistic integration of multiple sensors enhances the reliability and performance of perception, localization, and mapping systems.
• Current research emphasizes deep learning approaches for robust AV environmental sensing.

Conclusions:

• Deep learning-based sensor fusion is a key enabler for advanced autonomous vehicle capabilities.
• Addressing sensor limitations through fusion is essential for safe and efficient autonomous driving.
• Future research should continue exploring innovative sensor fusion strategies and their integration into AV systems.