Unmanned vehicles on the main sensors: laser radar, camera, millimeter-wave radar, GPS, ultrasonic radar and wheel angle sensors. ElonMusk has repeatedly said in public, do not use the laser radar Sensor only camera, but also to achieve Level 4 above unmanned.
Tesla's car is already in sale, sell the car can only update the software, certainly can not change the hardware, such as all re-installed lidar. May 7 this year, the United States, Florida, a Tesla owners in the use of Autopilot when a car accident, and ultimately unfortunate death. Which also led to Tesla to provide computer vision technology Mobileye founder AmnonShashua and ElonMusk between the war of words, the two sides eventually broke: Mobileye announced that after the end of the contract with Tesla no longer continue to cooperate. Speaking of the Suction Control Valve car accident in May, in fact, before the accident, Tesla's millimeter-wave radar has been aware of obstacles, but the camera because of the light problem, did not find the blue sky and white clouds in the background of large trucks, and finally lead to a car accident. Musk certainly knows that the camera does not fly, so the new version of Autopilot in the millimeter-wave radar data as the main reference. Do not use the laser radar only camera, but also to achieve Level 4 above unmanned 'more is out of commercial considerations.
This is intended to use the existing sensor to collect data, while the price of laser radar down. If the price of solid-state laser radar really as the publicity said down to $ 100 to $ 200, in order to ensure the safety of car driving, Musk will certainly be used. One side thinks: the data is king, and then the cow's intelligent algorithm also spells the massive data. While the other is that the data is only building materials, a strong analytical ability to make it into Fuel Rail Pressure Sensor a skyscraper, the pursuit of efficiency led to the algorithm, large data can not replace the algorithm.
Most people think that Google's search is slightly better in quality than Microsoft's Bing search because Google's algorithm is good. But in the former Google engineer Dr. Wu Jun view, this view is correct before 2010, because then Bing in technology and engineering significantly behind Google. But today the two companies in technology has been almost the same, Google can be slightly dominant, largely rely on the power of data. Unlike the search algorithm is not yet mature in 2000, today there is no unknown method, but it will be able to improve the accuracy rate even if a percentage point. Google with the PageRank algorithm to the search results brought about a qualitative change, and good search results can attract more users to use Google's search engine, which unknowingly to Google to provide a lot of click data. With these data, Google can train Temperature Sensor a more accurate 'click model', and click models contribute today to search for at least 60% to 80% of the weight of the search, which will attract more users, the whole process is a typical constant Self-reinforcing positive feedback process.
In fact, as early as 2005, Google's machine translation quality let the world engaged in natural language processing people shocked: never engaged in machine translation of Google, the National Institute of Standards and Technology in the annual evaluation of the lead. In Arabic-to-English closed beta test, Google's BLUE score of 51.31%, leading the second place nearly 5%, while raising the 5 percentage points in the past need to study 5 to 10 years. In addition to Google's usual style of action - to the world's best experts in the field, the University of Southern California ISI laboratory Franz - Ouke (FranzOch) dug up, the most critical or Google hands to improve Machine translation system requires large data.
From the year 2004 to join Google in 2005 to participate in the NIST test, during only one year, so short enough time only in the South Canada system with Google's program style to re-achieve it again, no extra time to do new Research. The secret lies in the use of Google in the use of Google or the use of the method, but take full advantage of Google in the data collection and processing advantages, the use of more than ten times the data of other research institutions, training Out of a machine translation of the six yuan model (generally speaking N-element model Pressure Sensor of the N value of not more than 3). When the data used by Ocho is tens of thousands of others, the accumulation of quantitative changes led to the occurrence of qualitative change, and this is one of the most authoritative experts in the field of artificial intelligence Geoffrey - Hinton (GeoffreyHinton) Professor Adhere to the 'different' it.
It is worth mentioning that, SYSTRAN company is a use of grammar rules for translation of the company, scientists have not thought or conditional use of statistical methods for machine translation before the enterprise in the field of machine translation is the most leading. But now with the use of data-driven statistical model of the translation system compared to its translation system is very backward.
In the current business competition, compared to the algorithm or mathematical model, the importance of the data is indeed much larger, that is, the data is king. Because the former often by the academic community in a few decades ago has been found, all enterprises can be used, but the multi-dimensional complete data is not every enterprise has. Today, many companies in the product and service competition, to some extent is the data of the competition, it can be said that there is no data without intelligence. Because theoretically, as long as you can find enough representative data, you can use the probability of statistical results to find a mathematical model, making it and the real situation is very close, thus saving a lot of manpower costs or give the user more pleasant Experience.
Tesla has accumulated 222 million miles of driving data, and the future will be accumulated data for their R \u0026 D Level 4 above the unmanned car is very helpful, Tesla may be the first step to achieve mass production of Google. Currently for commercial reasons, the production of Tesla with 'camera millimeter-wave radar radar' as the main sensor, but wait Throttle Position Sensor until the low-cost solid-state lidar performance more secure. I believe Musk is sure to be loaded, because it is very helpful to ensure that the vehicle safety of 99.9999% is achieved.