Multi-Agent Sensor Fusion Methodology Using Deep Reinforcement Learning: Vehicle Sensors to Localization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Robotics

Background:

• Autonomous driving systems face challenges in detecting all obstacles in complex urban environments and varied conditions.
• Current advanced sensors and processing systems have limitations in perception.
• Artificial intelligence (AI) is being explored to enhance vehicle perception.

Purpose of the Study:

• To present a novel AI methodology for improving vehicle perception capabilities.
• To introduce the CarAware framework for sensor data fusion and vehicle position prediction.
• To apply Deep Reinforcement Learning (DRL) for perception tasks in autonomous driving.

Main Methods:

• Development of the CarAware framework for fusing multiple sensor data types.
• Application of Deep Reinforcement Learning (DRL), specifically the Proximal Policy Optimization (PPO) algorithm.
• Training and evaluation of the DRL model for vehicle position prediction.

Main Results:

• The CarAware framework demonstrates effectiveness in predicting vehicle positions by fusing diverse sensor data.
• The PPO algorithm was successfully trained and evaluated within the CarAware framework.
• The methodology shows promise for enhancing perception in autonomous vehicles.

Conclusions:

• The proposed CarAware framework offers a new approach to perception challenges in autonomous driving.
• Deep Reinforcement Learning can be effectively applied to perception tasks, not just control.
• Further development of sensor fusion techniques using AI can significantly improve vehicle safety and reliability.