LiDAR in Self-driving Cars

Self-driving Cars

The principle of LiDAR is to emit a laser beam to a target object and then determine the actual distance from the target object according to the time interval between laser beam emission and reflection. The biggest feature is that the distance measurement is accurate, which can reach the level of accuracy. Such high-precision measurement provides data guarantee for the follow-up algorithm of unmanned driving.

In terms of 3D environment perception, LiDAR can scan the static and dynamic obstacles around the vehicle in real time, and rely on the point cloud classification algorithm to segment and classify the obstacles, and output to the downstream control and decision-making module. The planning and decision-making control module is based on different obstacles. Make different behavioral decisions, such as following, overtaking, parking, etc.

In terms of auxiliary positioning, features can be extracted from the point cloud scanning results, and compared and matched with the data of the high-precision map, to obtain an accurate physical location.

Or based on the intensity of the reflection value of the point cloud, do positioning based on the probability of the intensity of the reflection value (this method is used in the Baidu apollo positioning algorithm), which can achieve centimeter-level positioning accuracy.

LiDAR makes up for the shortcomings of other sensors, but it also has its shortcomings, such as sensor noise in rain and snow.