A Deep Reinforcement Learning Strategy for Surrounding Vehicles-Based Lane-Keeping Control.

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Automotive Engineering

Background:

• Autonomous vehicles (AVs) require advanced safety features as autonomy increases.
• Lane-keeping systems (LKS) are crucial for driver convenience and safety in AVs.
• Sensor failures necessitate robust fail-safe mechanisms for reliable LKS operation.

Purpose of the Study:

• To propose a reinforcement learning (RL) control method for lane-keeping systems (LKS) in autonomous vehicles.
• To utilize LiDAR sensor data for LKS, offering an alternative to traditional camera-based systems.
• To enhance the safety and reliability of LKS through a novel RL approach.

Main Methods:

• A reinforcement learning (RL) framework was developed to control lane-keeping.
• LiDAR sensor data, representing surrounding object information, was used as input for the RL agent.
• The learning environment was created using IPG CarMaker for vehicle dynamics and MATLAB Simulink for data analysis and RL model development.
• Gaussian noise was introduced during simulation to emulate real-world sensor noise and test robustness.

Main Results:

• The proposed RL method successfully used LiDAR data for lane-keeping.
• The system demonstrated the ability to maintain the vehicle's lane effectively.
• Simulated sensor noise did not significantly impede the system's lane-keeping performance, indicating robustness.

Conclusions:

• LiDAR sensor data is a viable alternative to camera data for reinforcement learning-based lane-keeping systems.
• The developed RL control method offers a promising approach for enhancing the safety and reliability of autonomous vehicle LKS.
• Further validation in real-world scenarios is recommended to confirm the system's practical applicability.