Preface
“Unmanned driving Q&A” is an extra part of the basic unmanned driving technology series. Here, I will summarize and answer many high-quality questions that readers have asked in the replies and articles. The answers to these high-quality questions will serve as supplements and extensions to unmanned driving technology.
Mainbody
Question One
Q: Personally, I don’t agree with unmanned driving or car-to-car networking. We have to entrust safety to computers, and the stability of these computers must be extremely strong, and they cannot fail. For example, short circuits or other issues. In the future, if cars are networked with driving computers, can’t high-end hackers manipulate cars? Just like in “Fast and Furious”? I feel that it is dangerous to walk on the road.
A: Regarding the stability problem you mentioned, there is a special team called “Functional Safety” to research this issue in unmanned driving, just to ensure the absolute safety of passengers in the event of functional failure. As for the concern about high-end hackers manipulating cars, it is just like everyone’s fear of personal account hacking in the popularity of mobile payments. It is not unreasonable to be afraid, but more security experts will design more secure architectures. Technological development is not achieved overnight and requires long-term accumulation. Insiders consider more and more details about what may happen than people outside the industry. It is an inevitable trend that highly repetitive mechanical actions like driving will be replaced by higher-tier machine intelligence.
Question Two
Q: I am a senior in vehicle engineering. I have studied OpenCV before and used it for recognition, but the effect was not very good. Later, I got into the deep learning track and found that its performance was very good. Now I am also facing the task of finding an internship or job. So as a professional in the field of vehicles, do you think it is necessary to continue to study deep learning? Which aspect is more important for job searching? I look forward to your advice!
A: Deep learning is a very promising and prospective direction. From the perspective of enterprise HR, the competitiveness of undergraduate students is definitely not as good as that of graduate students. Therefore, if you can enroll in graduate school, learning deep learning is still worthwhile. If you cannot enroll in graduate school, evaluate the difficulty of taking the postgraduate exams and consider attending schools that are good at this field. If you do not plan to pursue higher education, look for internships in companies in this field. These companies welcome capable students, such as autonomous driving field Yushi Technology, Baidu, and even vehicle factories. If you have trouble finding an internship, just apply for all positions that contain the keyword “deep learning”, especially in the field of autonomous driving, there will be opportunities.
Question Three
Q: It seems that studying unmanned driving is more suitable for computer majors. Are students majoring in vehicle engineering not suitable for working on unmanned driving? Do they have to change majors when they apply for graduate school?
Note: No HTML tags in this question, only translate the content.A:
Q4: My son is currently studying at Jida Automotive and plans to study in the United States after graduation. If he wants to work on autonomous driving in the future, which research direction should he choose?
A4: You need to research universities in the United States that are strong in autonomous driving, or check which universities have high scores on the KITTI dataset. You need a solid programming foundation in university and learn as much computer knowledge as possible. As for which research direction to choose, there is no single answer. There is demand for talents in various research directions. You can share my answer “What technologies are involved in autonomous driving cars?” with him, and he should have his own ideas.
Q5: Hello, can I ask you a question? I am a graduating undergraduate student who has never been exposed to autonomous driving before. I plan to take the postgraduate entrance examination in 2018 and have only basic knowledge of Python. According to what you said, what else do I need to overcome in order to study autonomous driving? I look forward to your advice. Thank you.
A5: You can do research on the Changshu China Intelligent Vehicle Future Challenge to see which universities’ teams have won the championship in previous years. Find a teacher from that team and try to get in touch with them. This is the first step. Before your graduate school interview, you can read books on autonomous driving such as popular science book “The First Book of Autonomous Driving Technology”. As for specific research, you should wait until you receive the offer from the graduate school.
Q6: What technologies are involved in autonomous driving cars? It was mentioned that “features such as Tesla’s AutoPilot do not even need to use laser sensors.” But isn’t Tesla’s fatal accident caused by mistaking the reflection of a white truck in the strong sunlight for a highway sign? Therefore, in addition to lidar, I think even if driving only on highways, sensors such as sonar and ultrasonic should be added to achieve true autonomous driving, as autonomous driving without safety considerations is rogue.
A6:A: The terms of Autopilot clearly state that the cruise control function is only applicable on highways and passengers must keep their eyes on the road and be ready to take control of the car at any time. From a legal standpoint, Autopilot is simply a lane-keeping feature for a single scenario. However, Tesla has exaggerated and promoted it as self-driving, leading to many people not being cautious enough when using it. Tesla’s accidents are caused by using Autopilot in areas where it is not recommended according to the terms.
Another important reason is that the Autopilot algorithm relies mainly on vision and is supplemented by mmWave. For driving comfort, Tesla filters out all static obstacles detected by mmWave radar (for example, overhead gantries and signs above the road on the highway, which could otherwise be classified as static obstacles), resulting in accidents when trucks across the road cannot be detected by vision and static obstacle data from mmWave radar is not utilized.
At present, without lidar sensors, it is difficult to achieve complex autonomous driving, which we agree with. However, for a single scenario such as highways, the algorithm performs well enough without the use of lidar, as laser sensors are too costly.
Postscript
That’s all for the first Q&A session on autonomous driving technology. If you have any further questions about autonomous driving technology, please leave a comment in the comment section below, and your question might be answered in the upcoming episodes.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.