Human-like Decision Making for Autonomous Vehicles at the Intersection Using Inverse Reinforcement Learning.

Area of Science:

• Autonomous Driving Systems
• Human-Computer Interaction in Transportation
• Artificial Intelligence in Automotive Engineering

Background:

• Future roads will feature both autonomous and human-driven vehicles, necessitating seamless interaction.
• Autonomous vehicles (AVs) must emulate human driving for intuitive understanding and acceptance by human drivers.
• Current AV behavior may not align with human cognitive expectations of driving patterns.

Purpose of the Study:

• To develop an evaluation function for human driving behavior.
• To enable autonomous vehicles to imitate human driving patterns more effectively.
• To enhance the safety and predictability of autonomous vehicles in mixed-traffic environments.

Main Methods:

• Utilized inverse reinforcement learning (IRL) to learn human driver evaluation functions.
• Developed a semi-Markov model to infer intentions of surrounding vehicles.
• Classified surrounding vehicle intentions into 'defensive' and 'cooperative' categories.

Main Results:

• Successfully learned an evaluation function that imitates human driving behavior.
• The semi-Markov model accurately extracts surrounding vehicle intentions.
• Demonstrated the ability to adopt context-appropriate responses based on inferred intentions.

Conclusions:

• Learned driving behaviors enhance AV predictability and understanding for human drivers.
• Intent recognition using the semi-Markov model allows for safer and more adaptive AV responses.
• This approach contributes to more harmonious integration of autonomous vehicles into existing traffic.