As more and more self-driving vehicles go on the road this year, more people are beginning to pay attention to these cars, and what's certain is that they are no longer fresh products.
However, these self-driving vehicles also revealed some basic weaknesses when conducting public driving tests. Above all, although they have the ability to strictly abide by the rules of the road traffic, these self-driving cars show their understanding of the human driving behaviors of sharing roads with them Talent is not high, perhaps as Mike Demler, senior analyst at Linley Group, puts it: 'How can you program a robot to have human knowledge?'
Undoubtedly, in 2017 we saw several highly publicized, non-fatal, autonomous vehicle accidents.
A consumer survey conducted in 2016 by AgeLab and the New England Motor Press Association at the Massachusetts Institute of Technology (MIT) revealed that about 3,000 respondents were asked about the presence of autonomous vehicles When not interested, nearly half (48%) said they would not want to buy a car that did not have to drive at all; those interviewees did not like the feeling of loss of control nor did they trust technology and considered unsafe driving vehicles.
In short, most non-engineered consumers have doubts or mistrust of autopilot technology, but technology vendors and car makers have not shied away from them, apparently not paying enough attention to consumer acceptance and not expecting consumers to buy Large numbers of autonomous vehicles - at least for a short period of time - have been set aside by automakers for the first time to put the issue of 'human-machine trust' aside for future processing.
The team is the focus!
And now, the next big thing is the 'fleet' that owns and is responsible for running these self-driving vehicles. They see the self-driving commercial model with great potential. This year, manufacturers such as Waymo, Uber, GM, Ford and others , Made it clear that they would use the fleet service as the main market for autonomous vehicles.
GM's smart phone application Cruise (source: Cruise)
Phil Magney, Founder and Principal Consultant at Vision Systems Intelligence (VSI Labs), a technology consultancy, said: 'We've seen a step-by-step approach to advanced Level 2 autonomous driving applications such as Advanced Driver Assistance Systems (ADAS) and other sensors this year Progress, but Level 4 self-driving? They are progressing faster than industry expected a year ago. "
Egil Juliussen, director of market research at IHS Markit and director of market research for ADAS, agrees. "Waymo, in particular, has quietly been ahead of everyone else in the (unmanned aerial vehicle) market." He pointed out , Waymo has enough confidence that fully automated vehicles will start driving on the open road in Phoenix, Arizona, USA, and there is no need for safe human driving in the car, which is one of the major milestones for driving in 2017 .
Let's work together to review several important developments in autonomous vehicles in 2017 ...
Really 'no one' driving!
The self-driving car without a human pilot in charge of safety monitoring on the car has been tested on the open road─an John Krafcik of Waymo CEO recently revealed that the company has been in Arizona since mid-October with its autopilot Multi-purpose RV, for road testing without human driving.
And Waymo is boldly planning an 'early bird experience' program that invites the general public to experience these unmanned vehicles for free; the latest move from Waymo shows confidence in the company's progress in hardware and software technology and shows that the company is paying for it Fleet service, strong interest in competing with corporate services such as Uber and Lyft.
2. When driving by car human encounter
According to Linler Group's Demler, a very important lesson learned from driving in 2017 is proof that 'security is the single most significant challenge; it is not enough to just estimate how many accidents a technology can avoid.'
In 2017, we saw several self-driving accidents that took place on the open road. Although none of them were casualties but vehicle damage, and in most cases the authorities concerned found that negligence was a human pilot driving a car collided with a car, there were Several incidents have begun to allow experts to examine 'what they did and did not do about autonomous vehicles'; a case of a Uber-affiliated automatic piloted car accident in Tempe, Arizona.
A Uber autospoke Volvo SUV overturned in a collision in Arizona, USA (source: ABC 15)
According to Demler, Uber's accident is a daily occurrence and not a special case: "Automated driving is technically compliant with traffic rules when it's time to respond and act on the fly; 'In short he thinks Uber Self-driving programming does not include the defensive driving skills that all good people drive.
A similar case of a Navya autopilot bus collision in Las Vegas, United States, followed a slowly reversing cargo truck and then stopped 'very patiently' until it was The truck hit the truck. The police decided that the responsibility for the accident lies with the truck. However, experts later questioned why the car was parked in motion and why it was not honked to remind the truck to drive?
Demler says it now seems that self-driving cars are exempt from traffic rules on all books, but "we need to consider everyday situations to develop standards of driving behavior, especially for those traffic rules that may not be defined, but humans can use common sense Evaluate the situation with safe driving (in most cases); 'But:' How can a robot be programmed to have human knowledge? 'This question has not been answered in 2017.
3. War between Intel and Nvidia warmed
In 2017, the industry saw the swordsmanship of the performance of its own self-driving AI processors between Intel (ie Mobileye) and Nvidia, the two major suppliers of autonomous vehicle platforms. In the coming 2018 International Consumption During the Electronics Show (CES), Intel expects to announce an autopilot platform that combines EyeQ 5 with a low-power Atom chip and other hardware features including I / O and Ethernet links.
Intel CEO Brian Krzanich at a Los Angeles auto show introduced Mobileye EyeQ5 processor and Nvidia competing products Xavier performance comparison (source: Intel)
Juliussen of IHS Markit said although he was not sure how the war between the two would take place, he saw some advantages with Nvidia's DrivePX platform. He pointed out that Nvidia, with its DrivePX platform, has been more deeply AI-oriented Driving the vehicle developer camp, and once you have invested heavily in a platform, especially software programmers and system designers have built the relevant knowledge, it is difficult to switch to another platform, unless there is a very good reason.
In the meantime, Nvidia was also the first to build a fast-growing ecosystem centered on DrivePX, with several major announcements by Toyota Motor Co. in the year of 2017 to launch its advanced autopilot system using the DrivePX platform, which Intel The spokesman said they are also already establishing an ecosystem and there are several important partners, but they can not make public the list.
4. Baidu platform want to be 'self-driving vehicle Android'
China became one of the focal points of the autopilot battlefield in 2017 because Baidu's presence in the self-driving (and AI) segment seems to be increasing only as Baidu unveiled plans for the open-source autopilot platform Appollo this summer So far, 73 companies have joined the program, including China's leading OEMs such as Chery, Changan Automobile, Great Wall Motor Co., Ltd., and a number of US technology companies and German automotive players such as Nvidia, Intel, Microsoft, Ford, Delphi, Continental Bosch, Daimler, Velodyne and TomTom.
Not long ago, NXP announced its participation in the Apollo platform of Baidu, which will provide experience in the development of semiconductor technologies including millimeter-wave radar, V2X, safety and connectivity, and automotive technologies. It is worth noting that the Apollo platform is not just intended to grow in China The car industry, TomTom and Microsoft to join the platform is to provide services outside the Chinese market.
Baidu is ready to open source self-driving vehicle platform Kill the Quartet (Source: Baidu)
Another aspect of the Apollo project is a $ 1.5 billion fund set up by Baidu and the Yangtze River Industry Fund that will invest in over 100 autopilot development projects over the next three years.
Egil Juliussen, director of market research at IHS Markit and director of market research at ADAS, said Baidu's Apollo project is a direct reference to Android's script for providing autonomous vehicles with a cloud-based service, an open source software platform, and localization Sensor fusion and hardware reference designs and more. With Apollo's hardware reference design platform, partners are sure to start developing their own applications very quickly.
And because Baidu's reference design platform is currently based on Nvidia's DrivePX, Juliussen said: 'You might think that Nvidia has an advantage, but that does not mean that Baidu's platform will not import other solutions because Intel is also a member of the Apollo program one. '
Auto-driving vehicle industry related vertical and horizontal manufacturers (Source: EE Times)
In 2017, the autonomous vehicle industry proved to be an intricate network of partnerships, with Nvidia, Intel and Waymo playing a pivotal role and each of them building a stronger partnership than the others; however, Baidu embraced the open source platform Strategy that could change the market situation, Baidu will launch Apollo 2.0 at the International Consumer Electronics Show 2018 (CES 2018) in January, claiming to allow self-driving cars to run on simple urban roads.
5. Intel / Mobileye Establishes 'Safety Model' for Autonomous Vehicles
To Mobileye, a subsidiary of Intel, published a technical paper in October detailing how to use a formal mathematical formula to ensure that drivers drive on a cautious and responsible road without causing traffic accidents; the paper states that driving By following a set of rules that have been clearly defined in advance by industry and government authorities, they will only operate within a 'safe' framework and will not be the cause of the accident.
For this paper, Mike Demler, a senior analyst with The Linley Group, a market research firm, thinks there is "misleading" suspicion that it should not be done in such a way that 'our autopilot technology will not cause a traffic accident and we can tell you why.' Won the public's trust; but EE Times interviewed experts in the field, they are praising the company to persevere in the face of the most difficult issue in the field of self-driving.
However, academics also caution that the merit of formula methods and mathematical proofs is that they can be proved to be correct in principle. The disadvantage is that they always require basic assumptions, and those assumptions may not be true in the real world; Therefore, the first thing the industry needs to discuss is not just defining what kind of solution is appropriate to the system, but also whether such a solution is reasonable for the real world.
6. Waymo amazing simulation of autopilot mileage
Waymo published a safety report in October as a virtual roadmap for autonomous vehicle developers, with details detailing how far the company is already ahead of the industry─Waymo mentions how Design their autonomous driving hardware and software, and how to test the vehicle.
Juliussen of IHS Markit, after reading the Waymo report, said that the company's program, which competes with other self-driving vehicles the most, is that they design their own sensor system from a software perspective, and after eight years of autopilot software development, Waymo has a higher horizons than the other rivals, which mimics Apple's ability to work closely with software, which most traditional car makers lack and can not replicate.
Even more noteworthy, Waymo differentiated itself with its own test methods, and the company's safety report states that they have tested 8.5 million miles on actual roads; however, Juliussen notes that road test miles are not Instead, the actual road test results into the simulation environment, so that self-driving software in which the number of miles.
Waymo said in its safety report that the company turned the most challenging conditions that auto-driving vehicles encounter on real roads into virtual scenarios that allow autopilot software to practice in a simulated environment, with any changes in the autopilot software or Updates will be deployed on the fleet only after rigorous testing in a simulated environment Waymo ran 2.5 billion miles of simulated autopilot in 2016, which is 8 million miles a day, up to 1,000 a day by 2017 Miles.
How to operate an automated driving simulation environment (source: Waymo)
Juliussen thinks it's impressive: 'Because in the simulation environment, they are specifically testing the most difficult conditions;' The Democrats at The Linley Group also believe that it is impractical for machines to learn common ground by accumulating more actual driving miles Rather, it requires a great deal of simulation testing in real-world cases and the development of autonomous vehicles based on general safety standards: 'But to realize the latter, the industry still has a long way to go.'
Overall, Juliussen believes Waymo is the No. 1 player in the industry for progress in autonomous vehicle development, with GM second (GM acquired Cruise in 2016 for autonomous vehicle software developer) and Ford is slightly behind Far third place (the depot acquired Argo, developer of self-driving AI solution in February 2017).
7. Level 4 Self-driving can not run too far - is still limited by geographical and weather conditions
Although Waymo's technology for autonomous vehicles advanced ahead of schedule, the company's development efforts are still classified as Level 4 self-driving rather than Level 5. According to the SAE definition, Level 4 self-driving vehicles are ' Limit ", the driver may not need to pay attention to the driving safety of the vehicle, for example, in the array of traffic jams or a designated area, but otherwise, the driver will safely terminate the journey if the driver can not take control of it .
Waymo's report made it clear that the company's autonomous vehicles only operate on 'operational design domains', and their design conditions include the geographical environment, the type of road, speed limits, weather, operating hours, and local Traffic regulations; Waymo admits that such a designed field could be "very restricted", such as a single fixed route with low speed limits or a private park (such as the Science Park) that needs to be covered by daylight and weather conditions .
Waymo's self-driving vehicles can only travel in 'operational design' (source: Waymo)
Waymo explained in his report that their self-driving passengers can not choose destinations outside their approved geographic boundaries and that their self-driving software does not create routes other than 'geo-fencing'; if their co-driver's customer requests Outside of that line, Waymo sent a human-driven taxi instead of driving.
Of course, Waymo will continue to expand its operating range and actively test various bad weather conditions, such as varying degrees of snowy days; it is understood that the company began cold weather conditions testing since 2012 and announced that its self- The road to Michigan, U.S. state opens this winter, Jon Krafcik, Waymo chief executive officer, said: 'Our self-driving car will have more practical driving experience on snow, slippery and icy roads.'
Compile: Judith Cheng