Smart driving is a major direction to confirm no doubt, but if there is no hardware, software, and other aspects of the match, everything is just empty words. Back-end supply market has become a hot sweet pastry. On the smart drive special seminar at the CCF-GAIR Conference, these back-end suppliers will tell us they are intelligent handling of what efforts have been made in the future?
(From left to right in the order given) telematics Industry Association (TIAA) Secretary-General Pang Chunlin, sagitar poly CEO Qiu Chunxin, Minieye CEO Liu Guoqing, Quanergy China General Manager Tang Jinsong, China Institute data principal data scientists Yin Xiangzhi and QNX company greater China General Manager Jay.
Pang Chunlin: introduction to your company.
Qiu Chunxin: sagitar poly is a focus on the laser-radar company. In 2014, we introduced the first LiDAR product, mainly used in the field of static mapping, modeling, at present we mainly focused on research and development and provide high performance, high production of laser radar unmanned.
Liu Guoqing: Minieye was established in April 2013, we focus on the perception of in-car technology, we hope that the contribution to the perception of automotive automation and unmanned.
Tang Jinsong: our company is Quanergy, based in the United States in Silicon Valley. Focus on the development of high performance, energy in the future of laser radar for mass production and related software, we cooperate with many manufacturers already have depth. LIDAR M8 many of our existing machinery is in use. We will be introducing a breakthrough solid state laser radar S3, will provide the industry with true energy production of high performance, high reliability laser radar.
Yin Xiangzhi: we are now artificial intelligence competition held in the contest we will provide 50,000 vehicle traveling data recorder experience, developing algorithms associated with smart drive contest.
Zhang Renjie: our operating system is widely used in the familiar variety of models, including BMW, Volkswagen and others, we have made many contributions in the field of new generation autopilot.
Pang Chunlin: about Tesla's in China and the United States to see how?
Qiu Chunxin: this reflects the deficiencies of the Tesla drive is currently, probably because it hasn't been used laser radar, so the perceived shortcomings that led to the judgment of white body.
Liu Guoqing Tesla for automated driving capability requirements definition has problems, you need to explicitly define automatic driving scenes, and constantly experiments to test these features. Another point is to have a mechanism, through multi-sensor fusion--not only has radar, laser radar, a problem, others can be connected.
Tang Jinsong: from a technical point of view, it is because their system in strong light and side detection is not reliable enough. Need to guarantee the safety and reliability of the technology in the future, is coupled with a laser radar using sensor fusion--the forth, fifth-level goals, set the automatic driving mechanism to ensure security of the system.
Yin Xiangzhi: this incident highlights is at present the biggest obstacle AI: machines can only recognize what we taught and can't understand and judge the additional events. So algorithms need to identify towards understanding and awareness, so as to better prevent accidents from occurring.
Zhang Renjie: now our system security is divided into two types: functional security and system security. From the functional safety, Tesla in-vehicle systems that are not currently and system security, there is no backup, so this system is both a very weak.
Pang Chunlin: smart driving the car does need a black box in the future? Judging the signal instruction trace sources and what is the device? Moschino iPad 4 case
Qiu Chunxin now found to be judged by means of vehicle traveling data recorder. Installed on the driverless car is relatively large in the black box, when the accident happened, it can record from a context-aware information, DMU, motion control, and define responsibilities. In the era of unmanned, LiDAR is not only in the car, also like street lights, surveillance cameras installed around the road, traffic conditions are clearly present, as a result, finds it much easier.
Liu Guoqing: I think this is necessary. There are two thoughts: one is how to tamper data? Is this information going to upload?
Tang Jinsong: possible directions for the future were, each sensor at run time are recorded, and stored in the cloud. Role in the black box on the one hand, on the other hand is also a large part of the data. Did not rule out the possibilities of using black box on some record. In this way, intelligent sensing devices is very important, our solid state laser radar for the future, has great advantages and potential in this area.
Zhang Renjie: I think a black box approach to mend the way and cannot alter the fragile nature of the system. Automatic driving system from a security point of view into security area and not a security domain, how to do the isolation of these fields are very important issues, to addressing safety concerns at its source. There is a power problem, black box will have their own power, and when recording, when not recording, or SD failure ... ... Fraught with uncertainty. Now more popular software-defined vehicle, the software's growing role in the system, we can be more independent hardware functionality is implemented through software.
Pang Chunlin: currently on the self-driving car, do you think the sensor should be attached to an interface or multiple interfaces? Moschino iPad Case
Qiu Chunxin: external sensor input data directly into the car, is very big, I feel in this process is the front of the camera, laser radar have a certain data-processing capacity, provided to the information necessary for decision making the car itself. These simplified and effective data and information can be entered through an interface.
Liu Guoqing: I think they can share one channel, but not packaged with all of the information. On the mosaic of information, I think it is to think about, because the data is particularly important for unmanned, whether developing or testing. When we have clear requirements definition will need to do a lot of testing, must have an information feedback to help us to form a closed loop iterations, gradually improving feature.
Tang Jinsong: the future of computing platform, can do is to be effective and safe, reducing the computational burden, now it's hard to say what structure in the middle. May need to be involved, to find one of the best platforms, including GPU manufacturers.
Yin Xiangzhi: sensor of the unit as a whole has a lot of professional and complex algorithms, through a platform that is difficult to solve. For a variety of occasions has a different algorithm, if you really do not encourage innovation model. I think after the data is processed through a standard API will be transmitted to the central system can, but needs are defined around them with related companies involved in developing standards. Transmission accuracy depends on the performance of the company.
Zhang Renjie: I am more in favour of the integration trend, because the cart is a very powerful computing platform of the future, a variety of sensors, cameras and laser radar are there I/O line cards, all manufacturers follow the I/O line cards, the whole system becomes very simple. In addition to using such a framework can easily increase its computing power. For example the processor die, you can upgrade without redundant backup, backup I can do part of the GPU ... ... I/O line cards that can be used in several similar functionality to solve part of the problem, and continuous improvement.
Pang Chunlin: what about future cost trends in the development of automotive sensors and systems? Second is a practical subjects has any good suggestions, you can push this thing?
Qiu Chunxin: automatic driving more beginner's mind is to reduce the accident rate, which was unchanged. Sagitar poly laser radar are doing, now I can better answer your question, the current price is not your main reason. The main question then is how to meet the requirements of the depot, how to make devices more stable behind the other to say. Perception of unmanned sensors, more expensive is the radar, everyone is concerned, I believe that the price will come down soon.
Liu Guoqing: sensor main determining factor is the price of its production, cost reduction is just a matter of time. Reality promotion, I recommend using the camera, there are a few benefits: one is a lower-cost, more friendly; there is it on a new car has a big market, it's in the stock market, but also to improve traffic safety.
Tang Jinsong: laser radar to automatically while driving up to 30 cm to 200 meters, the amount of data a second over point cloud data, is a very complex technology development and research costs are huge. Traditional Lidar is spinning, is a combination, each factory-a is to be debugged, costs are more difficult to come down. Future of our solid state laser radar, no mechanical parts, fully realized by semiconductor and software. Through this technological breakthrough and repercussions, on the premise of future mass production, the price is down.
Yin Xiangzhi: I think the commercial is an effective approach. Business buyout even subsidized way may be more effective.
Zhang Renjie: we do really need 16 line LiDAR? Single camera + 4 lines of laser radar must be cheap, do better than single camera, why don't we make some low-end attempts? We always think very well in theory, actually we went too fast. Laser radar may wish to present selected high prices of low-end configurations. For China, the current mainstream economic models of the basic price of around 200,000, looks more like no candidate; but for commercial vehicles, 1.5 million cars to increase a $ 8000 laser radar is no problem. So can lead to promote in commercial vehicles. When the market reaches a price down, the last civilian cars are also popular.
No comments:
Post a Comment