科技食谱

它会警告您具有识别驾驶和自动驾驶的能力吗?

The CSAIL research team at MIT’s Artificial Intelligence Research Institute announced that it was able to study how to classify the social personality of other drivers in autonomous vehicles and to some extent more accurately predict movement.

Specifically, it uses a parameter called social value orientation SVO (Social ValueOrientation), which quantifies the degree of human driver’s voluntary or altruistic or cooperation. Lets you make it.

The research team tested the algorithm at the place where the lanes merge into the left turn. As a result, it revealed that with a 25% chance, it was possible to predict the behavior of other vehicles more accurately. For example, if an oncoming vehicle makes a selfish judgment, it shows a prudent waiting state.

At this stage, it seems that it detects a self-centered driver to avoid contact accidents, but if the behavior considered by SVO is refined, it may lead to countermeasures to reckless driving in the future or to prevent vehicle movement that causes reckless driving.

This insight into human behavior can be said to be an important factor in protecting human lives in situations where autonomous vehicles and human drivers run on the same road. The Uber self-driving test vehicle, which caused an accident that killed a pedestrian crossing the road, turned out to be a system that did not assume a person crossing a place without a crosswalk.

The research team says that working with humans means reading intentions to better understand human behavior, and trying to determine whether cooperative or competitive tendencies can be quantified as they influence driver behavior.

It also explains that it is essential for the safety of passengers and surrounding vehicles that autonomous vehicles behave like humans, and that acting in a predictable manner allows humans to understand the behavior of autonomous vehicles and respond appropriately. This means that driving away from humans can embarrass human drivers around them.

The next step in the study is to expand the predictive model for autonomous vehicles such as pedestrians and bicycles. In addition to fully self-driving vehicles, it could also help to quickly detect and warn dangerous-driving vehicles if it is installed in systems where human drivers will ride. Related information can be found here .