Predicting the Unpredictable

https://www.wired.com/story/teaching-self-driving-cars-watch-unpredictable-humans/

In the controversy over who accepts blame when crashes occur with self-driving vehicles, we are nearing a new solution to minimize this problem. As discussed in class, the reason that some autonomous vehicles crash could still in fact be due to human error as humans tend to drive erratically when they are in a hurry. In order to address this, researchers are working with scientists to implement a system which adjusts a self-driving car’s patterns based upon the urgency of humans driving around said vehicle. With many complaints about the passiveness of autonomous vehicles in states where they are being tested, companies work toward making their vehicles more similar to confident human driving.

Researchers at MIT have created a mathematical formula which supposedly implements psychological elements to assess if surrounding drivers are “road ragers” or “rule followers”. According to a computer simulation, this technique has improved performance by as much as 25%, making the autonomous vehicles easier to be on the same road with and less likely to crash into. This is all in an effort to make autonomous vehicles more adaptable to human behavior rather than humans being forced to change their own habits. The hope for this adaptive technology is that it can be replicated in other important areas of study including the medical field or in manufacturing. In industries where human-machine interactions are increasing exponentially, this technology is an extremely valuable resource to increase efficiency.

The newest component in researching safe methods of autonomous driving revolves around game theory. Popular in the development of computers playing chess, game theory maximizes the ability of a vehicle to make the right decision during times of uncertainty. Because humans make erratic decisions when driving, especially in unknown areas, autonomous vehicles have a difficult time differentiating what humans are trying to do. The main way to correct this is collecting more data about human driving patterns in order to make the unpredictable more predictable. This could raise the increase in performance from 25% to 50%. This raises yet another concern in the ethics of data privacy but this is an issue for another day.