has learned to buy a car, November 12, “the Seventh China and South Korea Automobile Industry Development Seminar” organized by the National Information Center for Information and Development, Industry and Hyundai Motor (China) Investment Co., Ltd. Cooperation (hereinafter referred to as “China and South Korea automobile Forum”) held in Beijing.
It is understood that the forum of “Eastern wisdom, autopilot drive a new era” as the theme. As we all know, the current automotive industry is experiencing an unprecedented large change in order to autopilot core of the intelligent network linking technology is reshaping the entire automotive industry supply chain. China as the world’s largest car producer and the consumer market, how will grasp the opportunity in the automotive industry, “the new four modernizations” Eve degenerated into a future-oriented leader in power? Hyundai Motor Group to become the world’s Top 5 on behalf of the automotive industry, so in the moment of great change in the industry, which in turn how to “future car strategy” as the cornerstone of sustainable development? Forum site, and Korean automotive industry experts, business leaders, as well as guests from Chinese and foreign enterprises also discuss the latest developments and future direction of the automatic driving skills from multiple angles.
on
seminars, company COO drops autopilot Meng woke said autopilot industrialization have a lot of problems, technical problems still exist, the whole industry is very concerned big challenge. But also think about how to cooperate with industry, government support, local tests have to do better, get better data iteration and build operations. Today we have to do is move forward in parallel to his, and gradually try to explore some of the ways can be solved, the autopilot and operational services pushed to a commercial stage.
The following is the speech Record:
all the leaders, guests, Good morning! I am privileged to have the opportunity to share with you together and I think commercialization drops on autopilot this level.
drops 3 years ago autopilot team set up, and then gradually build up to today are relatively forming a relatively large autopilot team.
with issue facing the industry over the past three years, mainly all face are the same, are technical challenges. In recent years, technically we will continue to face challenges, but we have overcome difficulties one after another after. We enter into the next scene is that we start to think about this from the business of askingquestion. How do we operate, how do commercial. We still have many challenges Technical, how can we put the art so that we begin to see the possibility of using and landing.
just take this opportunity today to share with you.
explain by bit, by bit is the largest one-stop travel platform, we are also the largest travel network platform on a global scale, there are more than 10 billion passengers travel experience annually, equivalent to the world of the population at least once per person, per ride. In addition to the domestic market, in fact drops in Latin America also did relatively large scale, especially as in Brazil.
how the pieces will provide a safer and more efficient travel service as a target, we want to give passengers a better travel experience. Future autopilot, will further realize the vision and pieces. Until now, we use three years to do a full stack of software development autopilot level, from more than perception, decision making, control, maps, etc. simulator, a full implementation of a fully autonomous vehicle automatically run up the entire software R & D system.
now probably has about 200 employees and engineers in the United States and China, China and USA together about four cities over the past year and a half to two years time to road test.
In August, the team drops the autopilot upgrade to a separate company. I was joined by bit until three months before that I have been doing venture capital, from 2015-2016 years to see autopilot industry.
In 2016, I voted in the industry its first autopilot company in two to three years time, China and the United States than we are concerned about this matter in terms of investment autopilot, investment perspective, the perspective of capital human perspective and social point of view this question, there are no good criteria, the only criteria is to do the company’s team, have not done anything, in related fields such as computer vision , the accumulation of automatic control, the automotive sector has not deep enough, to what extent we guess by this technology goes. And then a deep understanding of the industry that people may understand that speaking in terms of technical route, you choose the purely visual or should I say with a laser, sensor fusion program, is not it more reasonable. Further down to see how fast that is to come up with a car technology demo, began to run on the road, first openBegan to run wide at the road, ran on during the day easier road, the next day in the black, rain and snow, the driveway is not clear where to run, if I quickly come up with a good demo, means He says you can become a very successful company.
perception of the industry, relatively speaking, is relatively simple, in fact, behind the assumption that what was just said? Who can become a pioneer in technology, who can pull the technological gap, who is likely to be the next biggest company or the company’s success in the autopilot industry.
By 2018, we see that the competition between the autopilot, or iteration is not a single company to complete the mission.
If we put in terms of the competitive environment is beginning to feel more between the Union and the Union, many people in the industry we have found that the autopilot is a long-term need to invest thing while the resources required are not any company independent can do this thing, and need an industry, from upstream to downstream, to the government policy to all aspects, to go with this thing can be achieved, so that after 2018 we see soon to a phenomenon that a large number of industry alliances began to build, but the industry says the alliance is not generally a lot of companies, dozens of companies in the industry formed, possibly one or two or two or three companies to do this industry Alliance kinds of very deep, then the formation of such a system.
on the nature of this system that is behind us began to focus on the autopilot win or not, or can make it or not, not only the core level of technology. But that car is not correspond with the development progress, whether policy responses development progress.
If you do operate a vehicle, whether you can set up traffic system, etc., all things become very important, behind become more complex, the skills needed are becoming increasingly complex .
In fact, the environment is becoming increasingly complicated after, we need more resources, so the industry to form a broader alliance. From our point of view, to see the very clear point that, for example, in terms of policy, in fact, in 2017, in some domestic areas, there are some road test policy gradually open up. Especially after the second half of 2019, local governments have great efforts to carry out this example fromFloor of the autopilot, and the use of manned demonstration scene, gave the company largely autopilot space, you can put it together with the technical start of passengers, and then put this into practice scene. Although the scene has its limitations, the vehicle will not be particularly large, we can see a little bit of the future looks like. In the past we imagine the future looks like, through video, through our imagination. But soon we can play to a self-driving car by drops of APP, then get real feedback from the experience. This is difficult to do in the past.
From the beginning of the second half of the year it is getting closer to us. We are in a very interesting era, on the one hand try to push our technology, the speed of our advance technology iterations; on the other hand within a limited range, provide a more complete experience, let everyone know that this is what the whole experience of.
bit is automatically made to the driving scene L4 and above. L4 scene is directly related to the service, hoping to operate this service. Realize this thing takes a lot of links. The first is the degree of realization of autopilot technology itself. Just about the perception of these decision-making systems. There are many important node in fact we were going to suppliers, cooperation car prices, car prices, for example, we will decide L4 a car, it is able to deliver a steering system with redundant at what point in time, the same sensor is also very important, in a large number of road test car fitted with many sensors, these sensors companies can do the test, but can not do mass production vehicle operators, when they mature, they can help promote mature together, a lot of the problem is a mature industry with chicken egg . Another thing is the need for government support. This is very very important for us to different scenarios floor, to incorporate more interesting environment to test technology to them, let’s technology iterations more quickly.
In addition to this problem has been taken into account, we have to do for the L4 and above autopilot we have to consider two questions. The first one is data, the vast majority of the time we ran autopilot, our ability to collect data is relatively scarce, we limited the car to collect iteration speed is slow.
The other is the driving behavior of the autopilot when it comes to operations, passenger demand data, the data is send a single match. We assume that there is a self-driving car today, trying to turn it into a network service, you have to operate and manage, to scheduling, toPricing, repairs and maintenance and so on these things, for in autopilot will introduce some new challenges, such as remote takeover. These are the new challenges require very sophisticated operations team to do. These challenges become more and more true today.
drops want to use network operating system built around cars embodied the advantage to autopilot operations in the past.
on the first data bit transport passengers a year about 10 billion alone, there are tens of millions of single per day, and a large track of the vehicle. Then we need to do route planning and matching supply and demand of passengers and vehicles, we have accumulated a large amount of data.
In the autopilot to solve the problem, especially when the perceived problem, a very important issue is the problem of the long tail. The problem is to solve the long tail you have encountered plenty of equipment acquisition and collection vehicles to collect saw this thing, it can be done, we are about cars or other network systems, you as a good collection channels, establish good collection system, making us faster faced with this problem long tail.
middle of the picture overloaded tricycle, the question now facing is, after re-classify see, if you do not see as a possible static obstacles or other obstacles, not better predict its behavior with normal tricycle is the same, if not done such a prediction might be wrong. Only after more than enough to capture the magnitude of these problems and resolve gradually classification.
For general autopilot company, limited vehicle data collected. The company has more than 100 vehicles in China is still very small. After the data put into the simulator then iterates inside, and then to machine learning, re-fitted to the car, make a loop, this is an effective data structure, but a small amount of data, do we all know the depth of learning and intelligent learning the process requires a lot of data to fill, to complete the loop.
However, the amount of data is not enough, so the slow iteration. In terms of drops, in fact, we can use data more readily available to help iteration, a large number of network about cars can also collect data as part of our team. These data are also scenes of the real scene, the test value is very high, and then iterated to machine learning, the feedback to our team, so we speed up the rhythm becomes faster iterations.
This is the Rand Corporation last year to do aResearch, assuming that today very confident autopilot, autopilot capabilities and driving skills, like humans, this technology has arrived.
at 95 percent confidence level, need 275 million miles of driving continuous no accidents, to tell mankind about the same level, or say accident, but a succession of 8.8 billion miles, at least 270 million miles without an accident, there are 100 cars, 8.8 billion miles to open 400 years, if it is 275 million miles need to open 12.5 years, which is to verify that instead of developing the entire number of test mileage. 100 vehicles made difficult to make such verification, need to use a larger number of vehicles or other means to verify this part of the run through.
just said partial data, I repeat what network operators. Some areas particularly suitable autopilot, some areas are not suitable for automatic driving, never even suitable autopilot, the autopilot is not just a relationship with the car, a great relationship with the scene.
what we want to achieve scene. The first is the most unbalanced supply and demand, lack of supply of local passenger car hit areas. What are the greatest demand, but supply a minimum. What followed was a scene to find suitable autopilot.
Based on this we could determine that the areas in which we use the autopilot of the car, which used to go running in the scene, which gradually to run.
to do the autopilot is not a panacea to solve the problem is to solve the problem of the scene, the scene has a lot of dimensions, and then put these dimensions into the past, and then to operate the service.
If, as the manner just solved only address a small part of the scene, which is actually a very safe way too valuable. Net bilateral market is about cars, you need to have enough supply enough vehicles and drivers. Drops of autonomous vehicles will have them put the vehicle network system drivers, become mixed to send a single model, we put the platform in autonomous vehicles as a supplementary capacity.
even have a car or a very small car we can also operate the service up and running, the automatic driving traffic and demand conditions as well as the situation in terms of a single ride, if appropriate, sent there, this or this list passengers suitable for automated driving car to service, we put the car assigned to him; if not for us to send a car with a driver to him, this gradual implementation,This ensures safety and technology can be achieved gradually.
back to our topic, the biggest point industrialization autopilot difficulty faced by many of our difficulty, technical difficulty still exists from this year, in terms of the whole industry is a great challenge to solve technical difficulties at the same time we began to think about how we cooperate with parallel industry, how to get support from the government, how to get a better place to do the test, and how we build a better iteration data and operations.
what we do today is to move forward in it, drops gradually try to explore some of the ways can be solved, we hope to explore cooperation with industry partners, to cooperate together, the autopilot and operational services pushed to a commercial stage, thank you!