Research into Autonomous Vehicles Following and Obstacle Avoidance Based on Deep Reinforcement Learning Method under...

Area of Science:

• Artificial Intelligence
• Robotics
• Computer Science

Background:

• Traditional rule-based algorithms have limitations in autonomous driving response times.
• Autonomous vehicles operate in complex urban environments with map constraints like lane boundaries and rules.
• Handling dynamic and variable numbers of external obstacles is a challenge for current autonomous systems.

Purpose of the Study:

• To propose a deep reinforcement learning (DRL) approach for autonomous driving that incorporates map elements.
• To address vehicle following and obstacle avoidance challenges in urban driving scenarios.
• To develop an effective obstacle representation method for DRL models in autonomous vehicles.

Main Methods:

• A deep reinforcement learning framework is developed for autonomous driving.
• Map elements (lane boundaries, rules, center lines) are integrated into the DRL model.
• A novel obstacle representation method is proposed to process variable external obstacle information.

Main Results:

• The DRL approach demonstrates reduced response times compared to traditional methods.
• The proposed method effectively handles autonomous driving tasks including vehicle following and obstacle avoidance.
• The obstacle representation method successfully addresses the challenge of non-fixed obstacle states and numbers.

Conclusions:

• Deep reinforcement learning, considering map elements, offers significant advantages for autonomous driving.
• The developed DRL approach with integrated map information and advanced obstacle representation enhances vehicle safety and performance.
• This research contributes to more robust and efficient autonomous vehicle navigation in complex urban settings.