Intelligent Land-Vehicle Model Transfer Trajectory Planning Method Based on Deep Reinforcement Learning.

Area of Science:

• Intelligent Transportation Systems
• Robotics and Control Theory
• Machine Learning

Background:

• Intelligent vehicle motion planning faces challenges with model error and tracking dependence.
• Existing methods often struggle with continuous input/output control for complex maneuvers.

Purpose of the Study:

• To propose an intelligent vehicle model transfer trajectory planning method using deep reinforcement learning.
• To directly obtain effective control action sequences for intelligent driving maneuvers.
• To enhance generalization performance and reduce lateral control errors.

Main Methods:

• Extracting an abstract model of the real environment.
• Jointly training a deep deterministic policy gradient (DDPG) model with a vehicle dynamic model.
• Transferring actual driving scenes to equivalent virtual abstract scenes using a transfer model.
• Calculating control actions and trajectory sequences via the trained deep reinforcement learning model.
• Selecting the optimal trajectory sequence using an evaluation function.

Main Results:

• The proposed method effectively handles continuous input and output for intelligent vehicle trajectory planning.
• Model transfer significantly improves the generalization performance of the planning model.
• Continuous rotation-angle control sequences are outputted, outperforming traditional methods.
• Reduced lateral control errors were observed compared to conventional trajectory planning techniques.

Conclusions:

• The deep reinforcement learning-based model transfer method offers a robust solution for intelligent vehicle motion planning.
• This approach enhances maneuver control and reduces errors in real-world driving scenarios.
• The method demonstrates potential for improving the safety and efficiency of autonomous driving systems.