Tesla sees “graveyard ghost”? Autopilot systems reveal the technical truth of the “ghost”

Recently, a foreign netizen posted a video on TikTok, claiming that his Tesla had detected a lot of pedestrians when he drove by an empty graveyard, which caused many netizens to post that their own Tesla had detected people on empty streets, too.

Tesla has not officially responded to the video, but its autopilot technology route and inherent defects are worthy of attention.

Analogous to the driving process of human beings, the autonomous driving system of a car is mainly divided into three parts: perception, that is, the car can "see" the outside world; decision-making, that is, the car will decide the next action according to the perceived result; execution, means the car will put decisions into action.

At present, there are two main perceptual solutions in the market: one is the "vision + space" integration scheme represented by Waymo; the other one is the pure visual scheme represented by Tesla.

Tesla’s perception system adopts multiple sensor fusion to analyzes object signals captured by the sensors through algorithm, and then visualize them to the driver. Its sensor configuration includes six cameras (side rear view, front wide view, front main view, front narrow view, rear view, side front view), one ultrasonic sensor and one millimeter wave radar.

If the "graveyard ghost" video is true, then there may be ground obstacles and other interference factors in the surrounding environment at that time, so that the perception system made a false detection.

There are two possible reasons for the false detection: one is that the millimeter wave radar mistakenly detects the echoes of other moving objects similar to human beings as human beings; the other is to identify a person through a visual algorithm with the assistance of a camera.

The common view of several industry technology professionals is that, in the future, the development trend of autonomous driving perception represented by Tesla will have three main characteristics. Firstly, the importance of cameras will increase gradually, so the number will increase, the pixel density will upgrade, and the coverage will also be enlarged from outside the car to inside the car. Secondly, perception from static targets to dynamic scenes, recognition from the whole to the details, the prediction of moving targets will also be enhanced. Thirdly, the development from single-vehicle perception to collaborative perception, however, single-vehicle perception still will be the foundation. (Source: Tencent News)