At the end of this month, Tesla is once again in the spotlight, and this time not because of the “brakes”, but about a configuration to stay or go.
This configuration is one of the most critical sensors for smart cars – radar.
Recently, Tesla announced that from May, the North American market Model 3 and Model Y will no longer be equipped with radar sensors, eliminating the millimeter wave radar in front of the car, using only the camera as the information collection of the autopilot system, which means that Tesla is ready to “risk the world” and become the first car company to achieve a “pure vision” autopilot program in the real sense of the production car.
It is reported that Tesla will not make the same changes to the Model S and Model X models in the North American market for the time being, nor will it make changes to Tesla models produced in overseas markets.
However, the mere elimination of millimeter wave radar in the North American market has caused Tesla to lose the safety endorsement of several agencies in the United States. Several agencies, including the National Highway Traffic Safety Administration (NHTSA), Consumer Reports and the Insurance Institute for Highway Safety (IIHS), have announced that they will suspend or cancel some of the safety endorsements for Tesla’s Model 3 and Model Y models.
In fact, Tesla CEO Elon Musk has been staunchly opposed to LIDAR, and has even disparaged LIDAR in public on several occasions, such as: “LIDAR is pointless and unnecessary for self-driving cars”; “LIDAR is expensive, ugly and unnecessary “; “LIDAR is like a person with a bunch of appendages, the existence of the appendix itself is basically meaningless, it would be ridiculous if a bunch of them grew”; “Any company that relies on LIDAR will probably end up with no problems ” and so on.
So, why is Musk so adamant about “disliking” LIDAR, and indeed all radar, and pushing for an “all-vision” solution?
To understand this question, we may need to have a basic understanding of autonomous driving system solutions first.
Vision vs LIDAR
In order to create an autonomous driving system solution, it is natural that an autonomous driving algorithm is indispensable. And to reach the calculation, you first need enough information.
During the driving process, the source of information collected by the vehicle is the sensors, large and small, on the car. Among them, to achieve autonomous driving, the most important sensors, is the camera and LIDAR (of course, plus the assistance of millimeter wave radar and ultrasonic sensors, etc.).
Therefore, in the field of autonomous driving, it has been divided into the vision school and LIDAR school. The former advocates the use of high-definition cameras + visual recognition algorithms, and the latter advocates the addition of LIDAR to ensure system stability.
The visual faction, or purely visual perception solution, actually simulates exactly the state of a human driving a car, firstly, the “eyes (camera)” see the picture, then transmit it to the “brain (processor)” for processing and judgment, and then give The “legs and feet (driving mechanism)” to give orders.
In addition to Tesla, domestic companies, including Extreme Krypton and Baidu, also use pure visual perception solutions.
The LIDAR faction, or LIDAR solution, is LIDAR-led, with millimeter wave radar, ultrasonic sensors, and cameras to accomplish autonomous driving. The acquisition of a series of information, including object discovery and distance measurement, is mainly achieved by LIDAR, thus creating a 3D environment map to allow the system to analyze and give instructions.
Comparatively speaking, this solution is more like a person walking while holding a crutch. The new products unveiled by many car companies, including Xiaopeng, Weilai, BAIC ARCFOX and Jiji, all use this solution.
Which is better or worse?
The advantage of the pure vision solution is that the video data obtained by the camera is most similar to the real world perceived by the human eye and is closest to the form of human driving. At the same time, in the current stage of the high price of LIDAR, the vision faction only use the camera, it is easier to control the cost.
In addition, the images captured by the camera are well suited for object recognition and classification, and with excellent algorithms, they can effectively exclude interfering terms and enable autonomous learning during the driving process.
Tesla’s pure vision algorithm is already strong in analysis
However, as a visual algorithm similar to the human eye, it is easily affected by light and can easily cause misjudgment in extreme weather or poor lighting. The previous Tesla model, after turning on AutoPilot, had occasional sudden automatic braking when passing under overpasses or ordinary bridges, which was caused by the algorithm taking the sudden shadow in the camera as an obstacle.
Compared with the pure vision solution, the advantage of LIDAR solution is that the monitoring distance is longer, the accuracy is higher, the response speed is more sensitive, and it is not affected by the ambient light. Its processing of three-dimensional information, the calculation of the size of the object, the speed of movement have a very excellent performance.
LIDAR mapping 3D environment
However, LIDAR is not perfect, it also has weak object recognition, high price, in extreme weather such as rain and snow ranging accuracy will be affected and other problems, so LIDAR must use the information collected by other sensors, in order to work together in the algorithm to ensure safe driving.
In addition, a large amount of data collection, the processor’s arithmetic power requirements are also high, which is equivalent to raise the cost of the whole car.
Theoretically, the combination of vision technology and LIDAR is currently a more complete program, but after the integration of existing technology, product costs and other factors, want to achieve the integration of two programs “third program” is too difficult. It is also based on this background, the field of autonomous driving is divided into the visual faction and the LIDAR faction, the “distinct” two factions, while Musk and his Tesla is firmly “team” visual faction.
In fact, abandoning LIDAR, and even now abandoning millimeter wave radar, one of the important reasons Tesla made such a choice is to achieve a lower cost.
After all, the cost of LIDAR is very high, the early mechanical LIDAR price is often 100,000 U.S. dollars, and now with the development of the electronics industry, the price of LIDAR has been reduced, but basically remains at the level of several hundred to thousands of dollars, but this compared to the HD camera “only” a few dozen or even a dozen dollars price is still high. Many.
Lidar is costly
However, beyond the cost, Tesla’s choice to gradually abandon radar also has technical trade-offs.
First, LIDAR can not determine the specific nature of the object, it is easy to cause misjudgment. For example, a large plastic bag floating past the front, the camera solution can identify that it is a plastic bag, and then ignore it, but LIDAR will be judged to be an obstruction, and then stop the car.
Second, millimeter-wave radar likewise has many limitations. The detection distance of millimeter wave radar is directly constrained by its frequency band loss, which can neither identify pedestrians nor accurately model all the surrounding obstacles, and also needs to divide the valuable computing power of the processor due to the large amount of data acquired.
Finally, for Tesla, which is currently the furthest and best in vision solutions, already has a huge investment in vision algorithms. To give up such a large accumulation before, turn to do another route, from the input-output ratio can be said to be “more than worth it”.
Of course, some people combined with the news that Tesla signed a contract with Luminar, a LIDAR technology company, to use LIDAR for testing and development, as well as the photos of Tesla installing LIDAR on the Model Y that were exposed on Twitter some time ago, speculating that Musk might “face” and Tesla might turn to LIDAR in the In the future, Tesla may turn to the LIDAR route.
In this regard, some industry insiders judge that Tesla installed LIDAR on the Model Y, it is likely to “use LIDAR to train the vision algorithm so that it grows rapidly, so as to eventually abandon all radar completely”, in Tesla’s production cars should not see LIDAR.
Tesla is ready to “go to the black”
From the latest display page on Tesla’s North American website, the millimeter wave radar performance display information has been removed from the introduction of fully automated driving (FSD), retaining only the vision and ultrasonic sensor section.
Meanwhile, Tesla also released a driver monitoring system on May 28, which allows the Model 3 and Model Y’s on-board cameras to monitor driver behavior when Autopilot is enabled, detecting and alerting drivers to stay focused through the use of in-cab cameras. Tesla claims that “camera data is stored in the car and the system does not send data back to the company unless the user enables data sharing”.
The current FSD introduction no longer has millimeter wave radar
In addition, Tesla’s FSD V9.0 version will also be officially pushed out this year.
According to Musk’s previous statement, the architecture of the FSD algorithm has been “reconfigured” to enable pure visual perception through eight cameras around the body, and effectively solve the previous “ghost braking” problem. Pure vision Autopilot is now being rolled out in North America,” Musk said. This production version will be updated in 2 weeks, followed by the release of FSD beta V9.0 (also a pure vision version) a week later. the FSD subscription will be enabled around the same time.”
Of course, it will take some “sacrifice” for owners to upgrade to the latest FSD. Tesla said, “For a short period of time during this transition, cars with Tesla Vision may have some features that are temporarily limited or not enabled, including: automatic steering will be limited to a maximum speed of 75 mph (approximately 120km/h) and a longer minimum following distance. Smart Call (if equipped) and emergency lane departure avoidance may be disabled at the time of delivery.”
It is difficult to determine whether LIDAR will win the market in the future with its safety advantage or vision algorithms with their lower cost advantage.
But from the information available, Tesla is going to take the pure vision solution “one way to the black”.
Posted by:CoinYuppie，Reprinted with attribution to:https://coinyuppie.com/one-way-to-the-black-why-did-tesla-and-musk-abandon-lidar/
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.