Empowering safer socially sensitive autonomous vehicles using human-plausible cognitive encoding.

Area of Science:

• Robotics
• Artificial Intelligence
• Neuroscience
• Ethics

Background:

• Autonomous vehicles (AVs) are increasingly common, necessitating ethical decision-making capabilities.
• Current AVs lack social sensitivity, failing to consider the distinct yet interrelated impacts of multiple road users.
• Integrating ethical considerations into AV operation is crucial for safe and socially acceptable deployment.

Purpose of the Study:

• To propose a novel scheme for AV ethical decision-making that incorporates social concern and human-plausible cognitive encoding.
• To enable AVs to differentiate risk assessment based on road user categories.
• To achieve a holistic ethical decision-making process by encoding individual impacts into a collective behavioral belief.

Main Methods:

• Developed a scheme combining social concern and cognitive encoding for AV ethical decision-making.
• Assessed individual road user impact based on risk and categorized users.
• Holistically encoded independent impacts into a behavioral belief to support ethical decisions.
• Evaluated the scheme using two thousand benchmark scenarios from CommonRoad.

Main Results:

• The proposed scheme led to safer and more ethical AV decisions.
• Overall risk was reduced by 26.3%, with a 22.9% decrease for vulnerable road users.
• In accident scenarios, self-protection improved by 8.3%, overall road user protection by 17.6%, and vulnerable road user protection by 51.7%.

Conclusions:

• The human-inspired scheme renders AVs socially sensitive, addressing ethical challenges in driving.
• The approach enables differentiated consideration of road users and a holistic view of collective impact.
• This work paves the way for more responsible and human-like ethical decision-making in autonomous vehicles.