Two sessions in May | NVIDIA DLI Deep Learning Introductory Course
May 19/May 26 One-The text consists of 2131 words, 4 pictures, and the estimated reading time is 8 minutes.
Article source: AI Technology Base Camp (Public ID: rgznai100)
Author: Mavis
At the Baidu AI Developers Conference in 2017, the live video connected Li Yanhong, who was driving towards the venue in a driverless car on the Fifth Ring Road. He sat in the co-pilot and explained that the driver's hands did not touch the steering wheel. It is precisely because of this sentence that the unmanned vehicle received the first ticket from the traffic police. In the 2018 Spring Festival Gala, a fleet of more than 100 unmanned vehicles showed off their driving skills on the Hong Kong-Zhuhai-Macao Bridge, and hundreds of millions of viewers watched the video through live broadcast.
When we are still feeling that driving is a laborious thing, the progress of driverless technology has gradually begun to liberate our hands to the development of the brain. At the O'Reilly and Intel Artificial Intelligence 2018 Beijing Conference, the battalion commander and Dr. Li Liyun, the former founding core member of Baidu Silicon Valley R&D Center, talked about it. Li Liyun said that the most urgent technologies in driverless technology are: the balance between sensor capabilities and its value, and the planning and control of driverless vehicles.
The development of driverless technology not only liberates the hands, but also liberates the brain for human beings. Our attention does not need to be focused on driving, which will lead to great progress in economic and social benefits.
Undoubtedly, in the traffic environment of Beijing and Silicon Valley in the United States, driving is not a kind of enjoyment for us. It takes a lot of time and effort. Check email in the car, read the news, and even take a break. In addition, unmanned driving will also bring some changes to the economic ecology. For example, the unmanned industrial chain may change the location of the business district.
Data shows that human beings drive about one fatal accident every one million kilometers. Since the development of driverless vehicles, they have traveled tens of millions of kilometers, and an Uber accident has occurred. Relatively speaking, driverless vehicles are better than humans. Driving safety factor is higher.
Replacing drivers with driverless vehicles is a long process, and it can be seen that driverless tests are equipped with drivers. If unmanned driving technology is popularized, we can optimize the overall smart city. For example, everyone can have a centralized scheduling system to optimize the people who go to the same destination and promote shared travel.
Unmanned driving technology not only liberates human energy and efficiency, but the ultimate direction is to build smart cities and intelligent transportation planning. With the development of this overall planning, the proportion of drivers may gradually decrease, but the day it will eventually replace drivers, There may be a long way to go.
Li Liyun believes that the modern driverless technology in the United States is still far ahead of the domestic one. From the data reported by the California Department of Transportation, we can see that the domestic top Apollo still has a certain gap compared with Google Waymo and Uber.
Another point is that there are many autonomous driving talents in Silicon Valley in the United States, which is a very important difference. Google Waymo, Uber, etc. have accumulated a lot of autonomous driving talents, and China is still in the stage of accumulation in this regard.
The domestic testing environment is more challenging. The government provides a lot of support, whether it is traffic control regulations or technical facilities. In addition, the Chinese people are quick to accept new things, such as these mobile payments, O2O. This is a model that the United States has never seen before, and it is quickly accepted in China, which has a great advantage in this regard.
Therefore, China has great advantages in this kind of landing and transformation. The United States has a deeper accumulation of technology than China. In addition, in terms of talents, with the gradual accumulation and explosion of Chinese talents, the final gap will not be very large.
There are two parts to the technology that needs to be solved most urgently in driverless technology:
The first is the balance between the capability of the sensor and its value.
According to the statistics of Yole Développement, an authoritative French market analysis agency, intelligent driving mainly realizes perception through cameras (long-range cameras, surround cameras and stereo cameras) and radars (ultrasonic radar, millimeter-wave radar, lidar); the current most advanced smart cars 17 sensors are used (only for autonomous driving functions), and it is expected to reach 29 sensors in 2030.
The cost is difficult to come down, and cheap sensors cannot meet the safety requirements, so the balance between price, safety and capability is an important problem that needs to be solved urgently.
For example, lidar technology is not "atomic bomb technology", this technology just needs more precipitation and more energy to make it better and more refined. Technically speaking, there is the possibility of cost reduction.
Now every lidar manufacturer says that as long as you give me a lot, I can reduce the cost, so as long as the technical plan is finalized, the cost reduction can definitely be reduced, and its more challenge is how to deposit this radar. More stable, more accurate, and more suitable for the use of unmanned vehicles.
Second, the planning and control of unmanned driving.
The driverless technology has solved the problem of normal driving very well, but when encountering some abnormal situations, such as some pedestrians not obeying the traffic rules, or some extreme situations, how can we solve the long-term problems through algorithms? It's a challenge.
Perhaps the scene of an Uber accident occurs only when the driverless test is several million kilometers, and this kind of thing will be avoided as much as possible during the driverless vehicle test. In this field, planning control and simulators are the ones that can exert their strength. point.
Simulators and artificial intelligence are used to detect the extreme capabilities of some vehicles, or the reaction of vehicles in some extreme situations. These scenarios are often difficult to learn and test through data collection or normal means.
Many people think that artificial intelligence is not "intelligent" enough. This is because people have high expectations for artificial intelligence. From the perspective of unmanned vehicles, the human brain is a neural network that has been iterated for many years. When it is a designed network, this network is called the genetic and biological brain.
In addition, for example, you start driving when you are 16 years old. In fact, your brain's perception has been trained for more than 10 years. Your understanding of the world is not like a self-driving car, you make a lot of pictures, and then train , The perception ability of the human brain is very powerful, so there is still a long way to go before artificial intelligence can really achieve the perception ability of this person.
With the development of computer vision, artificial intelligence has obvious applications in perception and prediction, but in decision-making planning, the application is not so direct. With the development of artificial intelligence, decision-making planning has also begun to be driven by data. change.
We train our algorithm by collecting the difference between human driving data and machine driving data. Let our algorithm drive more and more like human behavior. This is a direction that artificial intelligence has begun to penetrate into decision-making planning. In the future, artificial intelligence will also become a mainstream algorithm for decision-making and planning.
The regulations on unmanned driving in various cities have just been introduced, and they are not so sound at present, but this is also a good embodiment of embracing the changes in unmanned driving technology. In addition, under the supervision of these regulations, it is more legal and effective to improve the stability and capability of the entire system, and then make the system better.
Many people regard the development of driverless technology as a game between technology and law. In fact, it is more like a process of mutual development and adaptation.
Finally, as a holiday benefit, the battalion commander sent everyone a PPT (with algorithm) of part of Dr. Li Liyun's speech at the conference.
Original link: https://mp.weixin.qq.com/s/jNKhLnVwN4Q3Y8gKkhip6A
For more concise and convenient classified articles and the latest course and product information, please move to the newly presented "LeadAI Academy Official Website":
www.leadai.org
Please pay attention to the artificial intelligence LeadAI public account to view more professional articles
Everyone is watching
![640.png?](https://ss.csdn.net/p?http://mmbiz.qpic.cn/mmbiz_png/fgnkxfGnnkQJsfHzk0g4uu7Oout4kXavwIic9jfUkqsJ3kVjyFcic0icWTHEnMCv7E8h7ciclOmNpspcnduk5PJia3w/640.png?)
Application of LSTM model in question answering system
TensorFlow-based neural network solves the problem of user churn overview
The Most Common Algorithm Engineer Interview Questions (1)
Sorting out the most common algorithm engineer interview questions (2)