Ethical and moral decision-making for self-driving cars based on deep reinforcement learning

What is it about?

Ethical moral decision-making in autonomous vehicles refers to the process of making decisions when it faces complex road situations and emergencies. In these situations, the vehicle must balance factors such as user interests, road rules, and legal responsibilities. For example, it involves deciding whether to prioritize the interests of the car owner or protect the interests of other vulnerable groups. Ethical and moral decision-making is a crucial part of the field of autonomous driving, and in previous research, it has mostly been in the theoretical stage. In recent years, with the flourishing development of artificial intelligence and the application of machine learning in various domains, using machine learning to explore and learn optimal strategies for various decision scenarios has become a possible solution. Our research combines deep reinforcement learning with the Rawlsian maximin theory to design a novel method called "Rawlsian DQN." Rawlsian DQN, which takes into account the vehicle's operating conditions, balances factors such as property loss and individual survival rates in the driving environment, resulting in an ethical decision that aligns with the preferences of the general public.

Featured Image

Photo by Samuele Errico Piccarini on Unsplash

Photo by Samuele Errico Piccarini on Unsplash

Why is it important?

Whether in logistics and transportation, public transit, or unmanned military applications, autonomous driving, as a key technology, holds immense potential and importance. Ethical moral decision-making plays an indispensable role in this domain, involving decisions related to human safety, societal acceptance, and the legality of the technology. During the operation of autonomous vehicles, facing complex road ethical and moral decision-making scenarios necessitates making decisions that align with various prospects. The significance of our research is as follows: (1) Innovative ethical and moral decision-making strategy: Our innovative Rawlsian DQN algorithm explores multiple decision possibilities, combining the Rawlsian maximin principle with the DQN algorithm to learn an ethical moral decision-making strategy. This strategy enables vehicles to make decisions that minimize overall loss in ethical dilemmas. (2) Meeting public expectations: Based on experimental results and questionnaire responses, our decision outcomes align to a significant degree with user decision preferences, meeting the public's expectations regarding ethical moral decision-making. (3) Visualization of the decision-making process: This study constructs decision scenarios in the Carla simulation platform, simulating the decision-making process in the operation of autonomous vehicles. The visualization of the experimental process and the scalability of decision scenarios contribute to a more comprehensive assessment of the ethical moral decision-making quality of autonomous vehicles. The innovative method proposed in this research provides a viable solution for ethical moral decision-making in the field of autonomous driving while also offering a new direction for experimentation in this domain.