Predicting Driver Personality with Artificial Intelligence

2019-11-19

Autonomous vehicles with various sensors and complex data processing capabilities are about to emerge, but even the most advanced cars lack social awareness. Although autonomous driving technology has made significant progress, it still lacks any form of cognition. Therefore, researchers at MIT's CSAIL have been exploring whether autonomous vehicles can classify the social characteristics of other drivers through programming, in order to better predict the possible behaviors of different cars on the road and compensate for their own cognitive deficiencies, making autonomous vehicles safer and more reliable. In a new paper, scientists integrated tools from social psychology to classify driving behavior based on a specific driver's level of selfishness or selflessness. Specifically, they used something called Social Value Orientation (SVO), which represents the degree of a person's egoism and altruism. By estimating the SVO of other drivers, the self-driving system can create real-time driving trajectories for autonomous vehicles, ensuring safe driving. The researchers tested the new algorithm by merging lane and unprotected left turn tasks, and the final results showed that the algorithm could better predict other vehicles with a probability of around 25%. For example, in a left turn simulation, if a selfish driver is approaching, the system will stop the car and wait until the other driver has passed before issuing the turning command. "Creating more human-like behavior in autonomous vehicles is crucial for the safety of passengers and surrounding vehicles, as making decisions in a predictable manner allows humans to understand the system's behavior and respond appropriately," the researchers wrote in the paper. To enhance the social awareness of vehicles, the team combined methods from social psychology with game theory (a theoretical framework for envisioning social situations between competitors) and conducted tests. They modeled road scenarios where each driver tries to maximize their own utility and analyzed their "best response" based on the decisions of all other agents. Based on a short segment of actions from other cars, the team's algorithm can predict whether the behavior of surrounding vehicles is altruistic or selfish, and then issue corresponding execution commands to ensure safety. In the next phase of the research, the team plans to apply their model to pedestrians, bicycles, and other agents in driving environments. Additionally, they will expand their research to robotic systems, not just limited to the field of autonomous driving. This paper will be published in the Proceedings of the National Academy of Sciences of the United States of America (PNAS) this week.

RECENT AI NEWS

RECENT AI TOOLS