Interactive Lane Keeping System for Autonomous Vehicles Using LSTM-RNN Considering Driving Environments.

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision
• Autonomous Systems

Background:

• Advanced Driver Assistance Systems (ADAS) and autonomous vehicles require sophisticated lane-keeping capabilities.
• Conventional lane keeping often relies solely on lane markers, neglecting crucial interactions with other vehicles.
• Real-time decision-making in dynamic driving environments necessitates advanced algorithms.

Purpose of the Study:

• To develop an interactive lane keeping model for ADAS and autonomous vehicles.
• To enhance steering control by integrating lane marker information with surrounding vehicle interactions.
• To leverage Recurrent Neural Networks (RNNs) for adaptive and responsive lane keeping.

Main Methods:

• Utilized a data collection vehicle equipped with a front camera, LiDAR, and DGPS.
• Designed a Recurrent Neural Network (RNN) model incorporating long short-term memory (LSTM) cells.
• Input features included lane information, surrounding vehicle data, and ego-vehicle states; output was the steering wheel angle.

Main Results:

• The proposed RNN-based model demonstrated accurate lane-keeping performance.
• Evaluations via similarity analysis and case studies confirmed superior results compared to conventional methods.
• The model effectively integrated surrounding vehicle interactions into steering decisions.

Conclusions:

• The interactive lane keeping model significantly improves upon traditional marker-only approaches.
• The RNN-based algorithm provides a robust solution for complex driving scenarios.
• This research contributes to the development of safer and more intelligent autonomous vehicles.