TU-TESL

• 22

Even the most advanced cars, equipped with sensors and complex data processing capabilities, lack social awareness. Although autonomous driving technology has made great progress, the drivers around it will still be regarded as obstacles, rather than people with specific intention, motivation and personality. (photo source: MIT official website) then can social consciousness improve the performance of self driving cars? Researchers at MIT's computer science and AI Lab studied this. According to foreign media reports, researchers have set up a system to classify drivers' behaviors according to the degree of selfishness (the possibility that drivers show altruistic behavior to other vehicles). The test results show that the algorithm can better predict the driver behavior with an accuracy of 25%. The team's model draws on both game theory and social value orientation (SVO) psychology. SVO represents a person's degree of selfishness (self-interest) and cooperation (pro Society). To build the first mock exam model, the researchers built the model for the driver's road to maximize his own interests. In this scenario, the model learns to predict whether the driver is cooperative, altruistic or egoistic based on the motion clips. In this way, AI can gradually learn to take different driving behaviors at the right time. For example, when merging and turning left, the driver can choose whether to merge other vehicles into the lane. Among them, drivers who let other vehicles parallel to each other think they are more competitive than drivers who don't let cars parallel to each other. Then, the model may be more decisive, making vehicles change lanes in traffic jams. In other cases, such as when a selfish driver makes a left turn, the model may wait for a more pro social driver to drive the car before taking action. The researchers say the system is not sound enough to be implemented on public roads, but they plan to use it for pedestrians, bicycles and other driving environment actors. In addition, the researchers plan to study other robot systems dealing with humans, such as home robots, and integrate SVO into prediction and decision-making algorithms.