Hardware flow>Algorithm flow? Who does the future of intelligent driving belong to?

As a popular development direction in the current automotive industry, intelligent driving carries many people's yearning for the future of travel. Discussions on this topic often include terms such as car-grade chips, cameras, and lidar. Let's talk about it today. , these are the hardware used to enable vehicles to achieve intelligent driving.

Nowadays, a vehicle with smart driving as its main selling point would be embarrassed to say that it is a car with smart driving functions unless it is equipped with ten or eight cameras, one or two laser radars, N millimeter wave radars and ultrasonic radars. But there are exceptions to everything. Tesla has always advocated purely visual intelligent driving. In layman's terms, the whole car only has cameras, and there is no hardware such as lidar or millimeter wave radar.

Is the pixels of a car-grade camera not as good as that of a mobile phone?

Among these sensor components, the one that everyone comes into contact with more often is the camera. Mobile phones, surveillance cameras, etc. are all equipped with cameras. As a visual sensor, a car-grade camera used to collect road information required for intelligent driving has the advantage of being relatively cheap compared to other sensors on the vehicle. At the same time, while collecting road information, it can also perform text, color, etc. Identify.

After years of technological updates and iterations, the pixels of mobile phone cameras have now reached 100 million. On the other hand, the pixels of car-grade cameras are hovering around 5 million or 8 million. At this time, some students may ask, is a car comparable to a mobile phone? It’s much more expensive, so why not use a 100-megapixel car-grade camera? This situation is mainly due to limitations in reliability, technology, and computing power.

We all know that mobile phones can perform facial recognition and tracking when taking videos, but this level of recognition technology is really nothing compared to car-level cameras.

Car-grade cameras have different focal lengths depending on their uses. For example, the sensing distance of some telephoto cameras can reach 300 meters. Vehicles, pedestrians, lane lines, and roads existing within this 300-meter visual range Signs, as well as various obstacles, etc., all need to be scanned, identified and transmitted to the data processing library.

Such a huge amount of data needs to be processed one by one. The higher the pixels of the car-grade camera, the higher the computing power required to analyze the collected information. If the chip computing power of the intelligent driving assistance system is not powerful, the entire information processing The slow process creates a delay. Considering that the car moves at tens of meters per second when driving at high speed, the vehicle may hit the obstacle in front of it in just 0.1 seconds, which poses a very large safety risk.

This is like an 80-pound girl competing with a 280-pound big eater. There may not be much difference in the time it takes for two people to drink 1 cup of 200ml milk, but when faced with a table full of 40 cups of 200ml milk, the 80-pound The girl obviously needs more time to finish her drink. Among them, 80 kilograms and 280 kilograms can be regarded as the chip computing power, and 1 cup and 40 cups are respectively the amount of image data recognized and collected by the low- and high-pixel car-grade cameras.

As a visual sensor, the information collected by the camera is essentially a two-dimensional image and cannot directly measure the distance, motion status and other information of the object in front. Therefore, except for Tesla, which is full of personality, other car companies basically cannot Only a single vision sensor is used on the vehicle.

Why don't we do an "X-ray"?

Among the various perception sensors used in current smart driving, vehicle-mounted lidar is the most popular, and its price is also the most expensive among various sensors. The working principle of lidar is to first emit a detection laser beam to the target object, and then analyze and process the reflected laser beam to obtain the target's orientation, distance, motion status and other information.

Compared with cameras, its advantage is that it is not interfered by environmental factors such as light, and can accurately identify the orientation, distance, motion status and other information of objects in front of the vehicle, and finally builds a complete 3D model of the object in the database.

This is like going to the hospital to see a doctor. The doctor can only briefly understand the patient's condition by looking, smelling and asking. If the patient's condition is very serious, he or she will need to conduct in-depth examinations with X-rays, CT and other instruments to clearly know the root cause of the illness. Among them, Wangwenwenqie represents cameras, and instruments such as X-ray and CT represent sensors such as lidar, millimeter wave radar, and ultrasonic radar.

The application of lidar can be traced back to the Hughes Aircraft Company in 1961. After decades of updates and iterations, airborne lidar technology has become very mature. However, as an ordinary passenger-level car, it is naturally difficult to consider many aspects. Will use the lidar set on the aircraft.

First of all, from a safety perspective, lidar is a radar that uses lasers as the radiation source. If the human body is exposed to high-power laser beams for a long time, it will cause varying degrees of radiation damage.

As a vehicle driving on the road, if the laser power emitted by the vehicle-mounted lidar it carries can cause harm to the human body, wouldn't pedestrians on the street suffer? Therefore, according to the requirements of relevant departments, the laser power of vehicle-mounted lidar must be controlled at the safety level of CLASS 1, which means it cannot cause harm to the human eye. Generally speaking, the detection distance of vehicle-mounted lidar is about 200 meters.

Furthermore, there is the level of integration and cost. A car is much smaller than an airplane, so the size of the vehicle-mounted lidar must meet the needs of easy installation, and it must also meet the company's cost control requirements. Therefore, it is enough to ensure the reliability and stability of normal operation under high temperature, low temperature, shock, vibration, water spray and other environments. At the same time, the petite vehicle-mounted lidar was born.

If you see a smart driving test car on the road with a cylindrical shield fixed on the top, then it is equipped with a mechanical lidar. The laser transmitter inside the mechanical lidar can scan and identify the 360-degree environment around the vehicle by rotating. Although this method obtains a richer amount of information, it is too large and costly to use. It is rarely seen on production cars.

Relatively low-cost and smaller solid-state/semi-solid lidar is the type of lidar that can be seen in mass-produced vehicles today. Because the overall size is smaller, there are more choices for their installation forms. Take Hesai Technology’s AT128 as an example. Its length, width and height are 13.7*11.2*4.7 cm respectively. It can be installed at the front, rear, and front of the car. Areas such as the top of the windshield and fenders are installed in different locations according to different vehicle shapes, and can be well hidden with a little design.

However, there are advantages and disadvantages. In order to pursue small size and low cost, solid-state/semi-solid lidar no longer has a laser transmitter that can rotate 360 ​​degrees inside. Therefore, the number of vehicles equipped with this type of lidar is generally more than one. Otherwise, it will be difficult to scan and identify the 360-degree environment around the vehicle.

It is an old friend of veteran drivers!

The name of automotive-grade millimeter-wave radar may sound unfamiliar, but it is an old friend of everyone. Before the concept of smart driving became popular, most vehicles did not have cameras, lidar and other configurations. However, previous vehicles could still support functions such as ACC adaptive cruise, and the major contributor was the car-grade millimeter wave. radar.

The detection range of car-grade millimeter wave radar is also about 200 meters. It is mainly used to measure the distance between the vehicle and vehicles, pedestrians, or obstacles on the road ahead, and also calculates the relative movement speed of the vehicle and the object in front. function. Careful friends may find that after the vehicle turns on ACC adaptive cruise, some vehicles will display the distance between the vehicle and the vehicle in front.

Because the price is lower than lidar, the cost of about a thousand yuan makes it have a very high loading rate. This can be seen in the number of models currently equipped with functions such as ACC adaptive cruise. Most of the newly launched models in recent years have has this function.

Because millimeter-wave radar itself has extremely strong penetrability, its installation areas are mostly inside car logos and anti-collision beams, and some are also directly exposed under the front air intake grille.

Please pay attention when reversing!

Car-grade ultrasonic radar may sound a little unfamiliar to you, so let’s put it another way. When talking about reversing radar, does it feel familiar instantly? When it is necessary to back up, or there are nearby pedestrians approaching the vehicle, the system will emit a rapid "beep" sound to warn the driver. These are the results of the hard work of car-grade ultrasonic radar.

In terms of volume, it is very small compared with lidar and millimeter wave radar, and its price can be said to be "white cabbage" compared with the two. Therefore, most vehicles now not only come with this configuration as standard, but also carry The quantity is also very large, usually around 10.

The cost is cheap because its detection capability is relatively weak. Compared with laser radar and millimeter wave radar that can easily reach 100 meters or 200 meters, the detection range of automotive grade ultrasonic radar is generally between 15 and 500 centimeters. If the distance is too far, Its detection capabilities will also become weaker.