Who will Tesla and Waymo take the lead in creating a driverless car that is safer than a human driver? According to foreign media reports, there is now a race to extend from Silicon Valley to Detroit. Who is the first to create a driverless car that is safer than human drivers? Compared with a few years ago, this was a much more arduous task because human drivers knew more about their cars than just about their behavior while driving. In order to achieve the same level of understanding as humans, computerized cars require a lot of data support. At present, the two companies with the most data are Tesla and Waymo. Both Tesla and Waymo are trying to collect and process enough data to create unmanned cars, but they deal with these issues in very different ways. Tesla is using its tens of thousands of cars on the road to collect real-world data related to how its vehicle uses Autopilot, a driver assistance system that is currently semi-automated. Waymo was initially Google’s driverless car project, using powerful computer simulations and using a much smaller test fleet in the real world to drive evolution. Supporters of driverless cars insist that driverless technology will reduce the number of people killed in car accidents each year. The figure in the United States is as high as 40,000. However, applying all these data-driven technologies to the road as quickly as possible may also bring enormous economic benefits. Intel believes that the income of driverless cars will reach 800 billion U.S. dollars in 2030 and reach 7 trillion U.S. dollars annually by 2050. In the summer of 2017, MorganStanley analyst Adam Jonas stated in the report that the data may be more valuable to Tesla than Model3. He wrote: "Only one market is large enough to push Tesla's value to the level of Elon Musk's personal ambitions: mileage, data, and content." Figure 1: Tesla video shows the autopilot in start-up state Tesla is using the customer-owned car to collect all the important data to realize the driverlessness of its vehicle. The company has thousands of customers, many of whom use autopilot on streets around the world every day. According to the privacy policy, Tesla can collect information on the performance of this feature. For anyone who pays attention to SpaceX, another Musk company, this is a familiar strategy. Musk has quietly tested equipment in real rocket launches and even sold the company's many test transmitters. It is difficult to determine how much mileage Tesla has obtained from its autopilot because the company has not issued much public statements about this. In 2016, the head of the autopilot business at the time said at the MIT conference that Tesla had recorded 1.26 billion kilometers of data, of which 160 million kilometers were in autopilot control (or partially controlled). At the end of the summer of 2016, Musk claimed that Tesla's daily collection of data was "just over 4.8 million kilometers." However, as of July last year, the total mileage of Tesla Motors has increased to 8 billion kilometers. As Tesla sells more and more cars, the amount of data it can collect will increase exponentially. Of course, not all miles are done under the control of a drone. Autopilots currently only have semi-automated driving functions. But Tesla also collects data on how the autopilot handles different scenarios, even if the feature is not actually used. The Tesla car can record the autopilot action and the data will eventually be transmitted to Tesla. This collection of so-called "shadowmode" means that Tesla can simulate data from fully automated driving on billions of kilometers of driving. Another company with a similar amount of data is Waymo, which announced earlier this year that it has simulated 8 billion kilometers of unmanned driving mileage. Waymo also stated that its driverless car has reached 8 million kilometers on public roads, which is more than the sum of road mileage measured by almost all other company's driverless cars. Waymo is limited to collecting real-world data from approximately 500 to 600 unmanned Pacifica car fleets. Tesla has more than 300,000 cars on the global roads, and the environment of these cars is much more diverse than Waymo, which is currently only in Texas, California, Michigan, and Arizona. Tested with Georgia. But Tesla can only learn in the real world, because even with autopilot, the current version is only semi-automatic. This balance will also change. Waymo plans to add "thousands" of Chrysler cars to its fleet by the end of this year. Recently, the company also announced that it will collaborate with Jaguar Land Rover to develop a fully automated, fully electric I-PaceSUV. Waymo said that in the next few years, the company will add 20,000 test cars to its fleet. And when all these cars hit the road, it will be able to handle 1 million car trips per day. But before that, Waymo relied heavily on simulations, but computers could not always come up with all strange real scenes. This is why Tesla is now leading in the real world, said Tasha Keeney, an analyst at ArkInvest. “I think everyone thinks that Waymo's technology is the best at the moment, but I think many people are Underestimated the power of the Tesla data set." Figure 2: Data Types The two companies not only collect data of different sizes, they also collect different data. Waymo's driverless van uses three different types of lidar sensors, five radar sensors, and eight cameras. Tesla's car is also equipped with a large number of equipment, including 8 cameras, 12 ultrasonic sensors, and a front-facing radar. But Tesla did not use a laser radar. Lidar is similar to radar, but it does not send and receive radio waves, but emits millions of laser signals per second, and measures the time they rebound. This makes it possible to create a very high-resolution image of the car's surroundings if it is placed in the right place (such as on the top of a car), and it can even include all directions. Even in the dark, it can maintain this accuracy because the sensor itself can provide the light source. This is important because in the dark, the camera's performance is usually worse, and radar and ultrasound are also less accurate. Lidars are expensive and cumbersome, and they also need to move mechanical parts (at least for now). Musk recently called the technology a "pillar" and said that although this will make things easier in the short term, the company will have to master camera-based systems to reduce costs. Gini said that if Tesla can develop a driverless car without this technology, it will have a huge advantage. She explained: "This is a more risky strategy, but it will ultimately bring them greater returns. If Tesla can succeed, then all other companies can only fend for themselves." However, in fact, this assumption is very uncertain. Raj Rajkumar, head of Carnegie Mellon’s Unilever Research Laboratory funded by General Motors, said: “Without lidar data, Tesla may find himself at a disadvantage. "Many in the industry believe that laser radar is an indispensable tool for building driverless cars. Rajkomar said that people are skeptical about Tesla's approach. He said: "We think that hardware is not enough to do this and Tesla is unlikely to succeed in creating a fully automatic driverless car." It is not yet clear what kind of data Tesla started to collect. According to the company's privacy policy, Tesla can obtain data on car speed, acceleration, braking, and battery usage, and can save “short video clips†in accidents. This data can be collected during remote or contracted services. However, for autopilot, the privacy policy only stipulates that Tesla can access "the information about the use and operation of this function." Tesla declined to comment on which sensors collected data and did not disclose the amount of data collected. Tesla can get all the videos on the car. It can also collect camera videos at certain specific moments (such as collisions) or Ultrasonic Sensor data without video. Rajkomar said that it is not yet clear whether Tesla’s collection of full-frame-rate video is still less realistic. Gini agreed, she said: "Waymo's data set is more detailed because they are using Lidar, which is much more information than you can get from the camera." Dealing with challenges Collecting data is one thing. But even Musk also pointed out that processing data is also a daunting task. Musk said in an earnings conference call last summer: “Processing this data and then using data to train the autopilot and letting the vehicle learn from the data effectively is actually quite a challenge because the amount of data is too large. huge." In contrast, Waymo seems more confident with its simulations. According to a report released by the Atlantic Monthly last summer, the company redesigned the full computer model of the city it is testing and tested 25,000 "virtual driverless cars" through them on a daily basis. This helps Waymo to create a tight feedback loop system by recreating real driving data on a computer where it can test “thousands of change scenariosâ€. This data was then downloaded to the Waymo test car. Waymo has also set up a special test facility in California where it can create specific street features or scenes in order to bring maximum trouble to its cars. Raj Kumar said that this closed-loop system "is based on the investment of incredible investment, resources, time and effort. Of course, Waymo has the strong support of the parent company." He said that Tesla is difficult to Match it. Tesla will have to invest more in this area and go through a highly labor-intensive process. In the second Tesla “Master Plan†published two years ago, Musk stated that driverless technology would need to travel about 9.6 billion kilometers if it truly wanted “global regulatory approvalâ€. Tesla is likely to have passed the threshold in the real world, but its car is still not fully autonomous. A Tesla Motors plan to drive from Los Angeles to New York in 2017 without a driver, but this demonstration has been postponed and Tesla's goal of creating a final version of the autopilot has not yet been realized. At the same time, Waymo is also close to 9.6 billion kilometers in simulated driving mileage, the company's virtual mileage growth is faster than ever before, and thousands of test vehicles are waiting to participate in the test. Waymo plans to launch a commercial taxi service using driverless cars later this year, which has already been piloted in Arizona and will further support its data feedback loop. Figure 3: Other competitors Tesla and Waymo are two leading companies that test driverless technology, but they are not alone. One of the most obvious competitors in this area is Uber, the online car giant. Compared to Tesla and Waymo, Uber used a more casual approach to pilotless testing. This is a typical characteristic of the company, which reflects the motto of “fast action, breaking the routine†in Silicon Valley. After starting tests in Pittsburgh in 2016, Uber launched an earlier version of the improved semi-automatic Volvo car on the streets of San Francisco but did not obtain permission from the state. Uber transferred the test to Arizona when the test was blocked. Uber eventually agreed to the basic requirements of California, but its compromise with lawmakers led the company to lag behind competitors like Waymo. When Uber builds test teams in three states, its driverless mileage will increase rapidly. According to the “New York Times†report, Uber’s driverless car has traveled 3.2 million kilometers nationwide in November 2017. It is unclear whether Uber's simulated unmanned driving range, or its technical quality, has also been questioned because a test vehicle killed a pedestrian in Arizona in March. Dara Khosrowshahi, chief executive of Uber, said that the company will still "commit to" the project, but its testing is still suspended. Gini said that in terms of driverless cars, the only competitor that can compete with Waymo or Tesla in terms of quality is a veteran company, General Motors. General Motors has been developing the driverless version of BoltEV with the help of acquired subsidiary Cruise Automation. The company recently announced plans to test its own limited commercial driverless service in 2019. Figure 4: General Motors is designing an all-electric Chevy Bolt with no steering wheel or pedal and will conduct commercial trials with its modified Cruise Automation technology in 2019 Generic is following Waymo's footsteps by generating and processing data that needs to teach cars how to use driverless technology. However, Gini believes that the advantage of GM is its scale of production. She said: "Waymo and Jaguar have reached a deal, which may help them make significant progress in the future, but they don't actually produce cars in-house. I think that having a vertical strategy is good. With autonomous sensors, when When you build from scratch, you can better handle the appearance of the product and how to optimize everything." Similar to Tesla's strategy, GM also has semi-automated products that are currently being used in customer vehicles on the road. However, this product SuperCruise is limited to a Cadillac model, and there is no indication that it will quickly spread to other models. In the eyes of Gini, GM missed a great opportunity. She said: "This is the opportunity they missed. All other car manufacturers also missed the opportunity. Why didn't anyone put the sensor on the customer's car, just like Tesla did to collect data?" Simulation strategy In this driverless car race, there is a dark horse that deserves attention. That is Nvidia. Although Nvidia’s driverless technology does not reach the billions of miles it boasts from Tesla and Waymo, its technology is being used by hundreds of companies in the unmanned field, including Tesla. Last month, Nvidia began selling its product called DriveConstellation, which is actually an analog device for other companies' unmanned projects. In other words, this is a commercial version of the simulation equipment that Nvidia has used for testing, which has been validated in Nvidia's own driver software and hardware. Danny Shapiro, senior director of the Nvidia automotive project, said in an interview that getting good analog equipment is crucial to the development of driverless cars. He said: "We can't drive on real roads, and we can't predict all the crazy things that happen on the highway. Initially, we needed to travel miles of kilometers, most of which were very boring. But after some time You suddenly found yourself ready to deal with this situation." This is the so-called "corner case" that engineers must study, that is, what happens rarely. Shapiro said that when he is driving, he will encounter many such situations, such as red lights, road anger, dangerous weather, and harsh sunlight at sunrise or sunset. In the real world, if you use a test car to do enough tests, you will certainly encounter these events and scenarios, but not yet frequent enough to learn how to deal with them. For example, in the real world, when the sun goes down, you only have a few minutes a day to drive at this particular time. In the simulation, you can drive 24 hours in a sunset scene, and you can also create a variety of potential dangers. This is why any company must first test the cause of drone in a simulated environment. However, by lowering the barrier to entry, Nvidia has made it easier for companies that do not have the scale or financial support of Tesla and Waymo to enter this field. More importantly, the wide adoption of the business model of Nvidia as a driverless technology provider may help create a de facto industry standard for driverless simulation. Nidhi Kalra, a senior information scientist at the non-profit research organization RAND Corporation, said that setting standards for driverless simulations may be an important step in this technology because it is difficult to assess the simulations done by private companies. the quality of. He also said: "The problem with the simulator is that it is a simplification of the real world. Even if it can accurately stimulate the world, if you simulate a sunny day without traffic in Mountain View, then in Mountain View What's the value of simulating 1 billion kilometers on the same dead road? I'm not saying that anyone is doing such a thing, but without this information, we can't know what a billion kilometers of miles really mean." Carla wrote a series of research reports on driverless technology for the Rand Corporation, including a 2016 study that attempted to determine how many miles it would take to drive in the real world to prove that driverless cars are safer than humans. Carla and co-author Susan M. Paddock concluded that driverless cars need "hundreds of millions of kilometers, sometimes hundreds of billions of kilometers," to prove their safety statistically, reliable. Therefore, they write that companies need to find other ways to prove safety and reliability. Kara said that simulation can achieve this goal, but driverless cars need to experience more environments. She said: "If I tell you, I drove a billion miles in Grand Theft Auto, but that didn't make me a good driver. When a company declared: 'We were driving in the simulation. In countless kilometers, I thought: 'Well, I'm glad you have a simulator.'" Kara said that for any company that "simulates driving range" and claims to achieve a milestone, unless they provide more detailed simulation data, its technology is questionable. Kara explained: “The real-world driving range is still very, very important. From a literal point of view, this is where the rubber meets the highway. There is nothing to replace it.†Figure 5: Knowing that Tesla and Waymo have accumulated the most mileage in the simulated and real world, this helps to discuss who owns “the most†data, but this knowledge alone is not sufficient to truly determine who has the ultimate advantage. If Tesla can truly achieve full autonomous driving without a lidar, then theoretically it can push software upgrades to its customers and gain advantages. But how does Tesla prove that his technology is safe? Tesla does have its own small test vehicle fleet and is registered with the California Vehicle Authority (DMV), but their mileage in 2017 is zero. The company has used its current autopilot version to accumulate a large number of miles through its customer fleet, but most of the data collected by the vehicles is related to its application of the autopilot in the real world. After a fatal accident, its technology is again affected. Survey of the National Transportation Safety Board of the United States. When there is a testing fleet of tens of thousands of cars, Waymo may better prove the safety of its technology, but this may also be difficult because it is still limited to a few locations. Even in today's relaxed driverless test environment, it will take time to expand the progress of these efforts. Another question is how to define "security." For all of these companies, the only common measure is the so-called disengagement, which tracks the number of times a safe driver replaces the autopilot system to regain control of the car. This is also an imperfect metric. It was only included in the California DMV, and it proved to be easy to fake because it has only a loose definition. When these companies prove to regulators or customers that they have developed fully autonomous driving technology, the most likely indicator is to judge whether the driverless cars developed by the company are as safe as human drivers, or safer than human drivers. As for how to define the accident rate per X kilometers, the injury rate per X kilometers, and even the death rate per X kilometers, there is another problem. As Carla and Paddock pointed out in their research, it is difficult to prove this in the real world. However, Carla believes that simulation alone cannot prove this, and at the very least, before a more comprehensive and open understanding of the quality and rate of collected data, we cannot rush conclusions. She said: "We may see this technology deployed first, and then there is conclusive evidence to prove how safe it is. This is the contradiction: Before we decided to use a driverless car, we couldn't prove how safe it was. ." Custom Piezoelectric Components The piezoelectric ceramic sound-generating element, especially a lead-free piezoelectric ceramic sound-generating element, is mainly composed of metal flakes and piezoelectric ceramic flakes. The sound pressure generated by this new structure piezoelectric sounder is proportional to the number of piezoelectric ceramic sheets. By increasing the number of piezoelectric ceramic sheets, the required large sound pressure is easily achieved. Custom Piezoelectric Components,PZT Elements Soldered Wires,Element of Piezoelectric,Piezoelectric Elements Crystal Zibo Yuhai Electronic Ceramic Co., Ltd. , https://www.yhpiezo.com