Prioritized Experience-Based Reinforcement Learning With Human Guidance for Autonomous Driving.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Robotics

Background:

• Reinforcement learning (RL) faces challenges with complex optimization and control problems due to high computational demands.
• Integrating human guidance into RL presents a viable strategy to enhance learning efficiency and overall performance.
• Autonomous driving systems require advanced control strategies that can adapt and learn effectively.

Purpose of the Study:

• To establish a comprehensive framework for human guidance-based reinforcement learning.
• To develop a novel prioritized experience replay mechanism that leverages human guidance.
• To create a behavior model that mimics human actions, reducing the burden on human participants.

Main Methods:

• A human guidance-based reinforcement learning framework was developed.
• A novel prioritized experience replay mechanism was proposed, adapting to human input.
• An incremental online learning method was used to build a behavior model mimicking human actions.
• The framework was evaluated on two challenging autonomous driving tasks.

Main Results:

• The proposed algorithm demonstrated improved learning efficiency and performance compared to state-of-the-art methods.
• Experiments validated the effectiveness of the prioritized experience replay mechanism and behavior model.
• The algorithm showed robustness in autonomous driving scenarios.

Conclusions:

• The human guidance-based RL framework significantly enhances learning efficiency and performance.
• The novel replay mechanism and behavior model effectively reduce computational load and improve learning.
• The proposed approach shows strong potential for complex control problems, particularly in autonomous driving.