Pure vision VS lidar, which intelligent driving route is better?

The second half of the competition in electrification and intelligence is undoubtedly about high-end intelligent driving. With Internet companies such as Huawei, Xiaomi, and Baidu directly or indirectly entering the car manufacturing industry, the competition in new energy has entered a new stage of development, and the competition for the right to speak over the advantages and disadvantages of the high-end intelligent driving route has become a must-fight area for car companies.

With the increasing penetration rate of new energy vehicles and Tesla's FSD autonomous driving technology about to enter the Chinese market, as well as news that domestic car companies have successively obtained L3 autonomous driving test licenses, the battle between pure vision and lidar routes will become a new battlefield.

Before analyzing the two intelligent driving routes of pure vision and equipped with laser radar, let's make a simple classification of intelligent driving. Strictly speaking, Tesla's Autopilot assisted driving system is the real pure vision solution, because the system does not carry laser radar, millimeter wave radar, ultrasonic radar, and purely relies on cameras to collect information, and the system then performs centralized calculations and gives instructions.

The entry-level PRO version of the Jiyue 01 and the Zhijie S7 ( parameters | inquiry ) (without laser radar) are equipped with intelligent driving systems. Although the official also claims that they are equipped with pure visual intelligent driving solutions, because they are not equipped with laser radar, they are equipped with cameras, millimeter-wave radars, and ultrasonic radars, so they are classified as pure visual solutions. However, it should be noted that their algorithms are mainly based on vision, but in certain road conditions or automatic parking, millimeter-wave radars or ultrasonic radars are still used. To be precise, they cannot be regarded as pure pure vision solutions, but intelligent driving solutions with "vision as the main and radar as the auxiliary".

The rest is easy. Cars equipped with LiDAR are classified as LiDAR intelligent driving routes. Therefore, in terms of intelligent driving routes, there are basically only two categories: pure vision and LiDAR. However, it does not make sense to distinguish them from the hardware level, because LiDAR, millimeter-wave radar, and ultrasonic radar all rely on emitting laser beams or electromagnetic waves outward. The signal detects the object and is reflected back to the vehicle for analysis. So why does the industry still classify intelligent driving systems equipped with millimeter-wave radar and ultrasonic radar as pure vision intelligent driving systems? In my personal understanding, it is because the initial cost of LiDAR is too expensive, while the cost of millimeter-wave radar and ultrasonic radar is relatively low.

In the early stages of research and development, new energy vehicles equipped with LiDAR, like some new forces using "air suspension", look more advanced and more eye-catching. I just want to tell you that LiDAR is so expensive, and I spent all the money to give you a good intelligent driving experience. I want to create a cognitive advantage at the marketing level. LiDAR is better than pure visual intelligent driving solutions, and it occupies the high ground of intelligent driving theory and law. Therefore, in terms of classification, it is not difficult to understand that everyone generally classifies intelligent driving solutions without LiDAR as pure visual solutions.

But then again, if we analyze it further, infrared cameras may be classified as laser radars, so it is meaningless to use them to distinguish between pure vision and laser radars. Back to the pure vision intelligent driving and the intelligent driving solution equipped with laser radar, which one is the ultimate intelligent driving? We know that Tesla has always insisted on the pure vision intelligent driving route. FSD is a pure vision high-end intelligent driving system, which has been highly praised and recognized by industry insiders and car owners for many years.

Currently, Tesla FSD has been tested and verified for more than 1.6 billion kilometers of driving mileage worldwide, and FSD’s entry into the Chinese market is a foregone conclusion. The pure visual intelligent driving solution can be recognized by the market, and the credit to some extent goes to Tesla’s FSD and Musk.

Musk believes that the pure vision solution mainly provides enough driving data to feed, and through massive data training and AI large model optimization and iteration, and accumulating driving experience, pure vision intelligent driving can be trained to drive like a "human". In addition to Tesla, Baidu and DJI also insist on the pure vision solution. The entry-level PRO of Zhijie S7 is equipped with Huawei ADS2.0 basic version, which can probably be regarded as a pure vision intelligent driving solution (without lidar). Of course, although Tesla FSD is becoming more and more perfect and has strong persuasiveness in the industry, it still cannot prove that the pure vision route is the ultimate intelligent driving in the future. Yu Chengdong once said: Multi-vision driving is the real direction of intelligent driving, and lidar intelligent driving is a higher-level choice!

So, regarding the pure vision and lidar routes, which type of intelligent driving is better? There are still obvious differences in the industry. The previous article made a simple classification of pure vision and lidar solutions, but did not evaluate the pros and cons of intelligent driving. Let's go back to the algorithm and logic level analysis. The lidar solution is obviously better and has better safety redundancy. The visual solution is easy to understand. It is based on the overall analysis of the data collected by the camera. If it is in special circumstances such as severe weather such as heavy rain and night, it is obvious that the pure vision solution is affected by the camera's line of sight, and its upper limit of ability to overcome certain severe weather is not high. The intelligent driving solution equipped with lidar uses cameras, millimeter-wave radars, and ultrasonic radars for comprehensive data calculation and analysis. The dimensions of the data source are wider. In layman's terms, the lidar solution is equivalent to adding lidar protection to the pure vision solution, and an additional safety redundancy. The hardware level is naturally better than the pure vision solution, which is equivalent to others walking on two feet and you running on four feet. The effect is definitely different.

Therefore, from the hardware level, the cost of a car equipped with LiDAR is obviously higher, but the theoretical upper limit of intelligent driving will be higher. Why is the upper limit higher, and it does not mean that it is better than the intelligent driving of pure vision solutions? This has to be said about the algorithm. The reason why Tesla FSD is more and more like a person driving is the effect achieved by algorithm iteration. Because LiDAR needs to emit laser beams outward, it has an additional channel for information collection, and then combines information from cameras and other sources for unified processing, which provides an additional guarantee. However, someone doesn't buy it, that is Musk, who once joked: "Humans don't shoot lasers from their eyes to drive, although it's a bit cool."

Therefore, the height that the intelligent driving system can reach is definitely the result of the joint action of software and hardware, rather than relying on one side alone to raise the intelligent driving to a high level. How to balance safety and intelligent driving also depends on the level of intelligent driving as a prerequisite. The reason why Huawei Intelligent Driving ADS can stand out is that the hardware and software of ADS are all Huawei's own. Good algorithms are adapted to good hardware, and intelligent iteration can naturally do better.

I believe that as the price of LiDAR continues to drop, intelligent driving equipped with LiDAR will become mainstream. At present, high-end intelligent driving systems equipped with LiDAR cannot be quickly popularized due to the high cost of LiDAR. In the future, with breakthroughs in technology and materials, the cost of LiDAR will be reduced, and high-end intelligent driving equipped with LiDAR will become the mainstream of the industry.