Article May 17, 2019.
A few years ago, the auto industry boldly predicted the release of fully unmanned vehicles of level 5 in 2020 or 2021, but this task turned out to be much more complicated than they thought.
The recently popular concept of unmanned vehicles that you can ride anywhere and anytime (or cars that you can sleep in the backseat while riding) slowed down a bit, as carmakers recognized that the development of full-fledged unmanned control technology turned out to be more complicated. than expected.
Questions about the future of technology reached full public comment in April 2019, when the CEO of Ford Motor Co. Jim Hackett acknowledged what was already painfully apparent to most of the engineering community. “We have overestimated the prospect of autonomous cars,” Hackett quote, which circulated through numerous news channels. “The scope of such vehicles will be quite narrow, the geofence restriction will come into play.”
However, Hackett was not the first to make such a statement. The automotive industry hinted at this for months before Ford announced. In November 2018, for example, John Krafchik, CEO of Waymo, Google’s drones division at Google, was even more outspoken than Hackett. “It’s very, very complicated,” Krafchik said during a live technology conference. “You don’t know what knowledge you are missing until you try to do something.”
Further, Krafchik said that the car industry may not be able to create a car that can drive independently at any time of the year, in any weather and under any conditions. “The condition of autonomy will always impose restrictions,” he added.
The comments by Krafchik and Hackett confirmed what many industry analysts have been saying for more than two years. “I agree with John Krafchik’s comment,” said Sam Abuelsamide, chief analyst at Navigant Research, a publication that publishes an extensive annual assessment of the robotic vehicle industry. “There is no guarantee that in the foreseeable future we will ever have automated vehicles capable of working anytime, anywhere.”
Fifth level dilemma
Such statements, of course, contrast sharply with earlier statements. Just three years ago, many OEMs, fueled by advances in robotics, boldly foresaw a day in the near future when drivers would not be needed. Ford, for example, predicted that this would happen as early as 2021. “There will be no steering in the cabin,” said former senior director Mark Fields in 2016. “There will be no gas and brake pedals and, of course, the driver will not be needed.”
And not only Ford made such statements. Honda has publicly discussed the possibility of cars without drivers appearing on the streets of Tokyo for the 2020 Summer Olympics. Volvo, Hyundai, Daimler, Tesla, Fiat Chrysler, Renault-Nissan and others planned to produce such cars in the range from 2018 to 2025. Some talked about models with disabilities: for example, first only highway driving will be available, and then city functionality will appear. But the message was essentially the same: the future is on our doorstep.
In 2016, Ford boldly predicted that in 2021 it would have cars without a steering wheel, gas pedal, or brake.
Of course, today this agenda is not much different from what it was in the past. A future with unmanned vehicles is still expected, but the tone of rhetoric is softening. Most OEMs are now more openly saying that the path to full autonomy in cars will be a sequence of small, gradual steps. Automatic emergency braking will appear first. Then a robotic package delivery and an unmanned taxi in areas within the geo-fencing. Drivers will first sit at the helm and then disappear. A high degree of automation – the so-called 4th level – will appear only in certain places. But the Holy Grail of unmanned driving – the full 5th level in all cars, providing driving anywhere and anytime – is now recognized as more difficult.
There are many reasons for this complexity. The first is the weather. Industry insiders say it is no coincidence that the most prominent standalone testing programs are located in California, Arizona, and Nevada, and not in Maine or Minnesota.
“Driving on a snowy road is difficult for a variety of reasons,” said Stuart Sellars, CEO of LiDar Group for Analog Devices, Inc. “Most sensors used for unmanned driving rely on the line of sight.” You use cameras, lidar or radar, and snow is essentially an obstacle. It prevents these sensors from receiving feedback. “
And the thing is not only that snowflakes in the air block the return signal. Snow also tends to accumulate on the side of the road and in the middle of the road, blocking road markings, which are so important for the automatic recognition of lanes.
Moreover, the issue is not only in the snow. Different regions present different weather problems. “If you are going to the northeast, then you are faced with ice, and heavy rain, and hail, so you have to solve completely different problems,” Sellars said. “So, yes, it may take longer than people expected.”
Perhaps the biggest technical hurdle is the transformation of the human mind into AI. Intelligence, which allows a person to drive a car, is taken for granted in many ways, and reproducing it is more time-consuming than engineers assumed.
“If you think about it, when you drive along the road, you are dealing with hundreds of different situations at every mile you have traveled,” Sellars said. “You see things and intuitively understand how to respond to them.”
And although these situations may seem simple for human drivers, they are not so simple for cars. For example, when a cardboard box is brought out onto the road 200 yards in front of the wind, human drivers quickly determine whether they should move it or go around it. For a computer, this is not so simple. Is this a piece of metal? Heavy or light? Does the car even know that a heavy piece of metal does not fly out of the wind through the roadway? All of these issues are very difficult for AI.
Most of these problems need to be solved using tests – either driving physical miles or performing software modeling. Both approaches have their place, mainly because software modeling cannot foresee every chance. For example, when a car arrives at a four-way stop at the same time as another vehicle, a dilemma arises for the car. Human drivers can make a gesture or look into the eyes of another driver, but microcontrollers do not know all this. Some developers are currently teaching their vehicles to move forward a little, monitoring another vehicle for implied consent, but such situations are not simple and, as a rule, cannot be modeled today.
Simulation system manufacturers are working on this, and are successfully expanding the number of tests that can be performed in software. Today, according to experts, there are two ways to simulate: firstly, recording real events and playing them in software, and secondly, increasing the number of tests to include situations that were not recorded in them.
“We believe that both are needed,” said Vinci Jeanne, automotive industry manager at MathWorks, a software product called Automated Driving Toolbox. “The human imagination is limited, and there are always real cases that you cannot imagine. Therefore, you should be able to take a certain amount of data from the reproduction and copy it into the simulation environment so that you can conduct tests in the spirit of “What if?”.
For both suppliers and OEMs, such procedures represent a brave new world of testing and inspection. Suppliers claim that this process is a deviation from all test procedures used before unmanned vehicles. According to them, unmanned vehicles are no longer just providing a part that meets the prescribed specifications. Suppliers should now help their customers understand the development of sensors and algorithms, in the context of their use, and not in terms of simple specifications.
“It’s not just supplying an airbag sensor that needs to meet specifications,” Sellars said. “With an unmanned vehicle in real-world conditions, you need to think about all use cases. This is the biggest problem, and you can only solve it with a huge number of tests. “
Indeed, countless hours of testing are required. Most of this is because engineers “don’t know what knowledge they lack,” said Waymo Krafchik. They need more hours of testing in order to take into account use cases that they cannot imagine. As a result, most experts estimate that the number of test miles should be measured in billions. Toyota, for example, has publicly stated that it needs 8.8 billion test miles for the safe introduction of self-propelled vehicles.
However, whatever this figure may be, almost everyone agrees that a large number of physical tests are still inevitable. “There are certain situations that we cannot simulate because it is related to human behavior,” Sellars said. “So the number of physical miles traveled should be a huge part of the process.”
Automakers go into the shadows
The big problem in all this for automakers is money. Manufacturers spend huge sums of money in their autonomous driving system development programs, and are constantly looking for investors to attract more. GM Cruise LLC, for example, recently announced a $ 1.15 billion equity investment from a group of institutional investors. The new funding brought the company a staggering $ 19 billion – about a third of the total cost of General Motors Corp. Cruise plans to use the money to double staff and triple the size of office space in San Francisco.
In 2016, General Motors invested about $ 600 million in a Cruise Automation robotics expertise.
However, such figures do not appear not only with GM. Most of the industry spends huge amounts at the same time. Ford, for example, has invested $ 1 billion in Argo AI; Toyota has invested $ 1 billion in Toyota Research Institute; GM has invested $ 500 million in Lyft, Inc .; Volvo entered into a joint venture with Uber Technologies Inc. and invested $ 300 million, while Intel is said to have spent $ 15.3 billion on the acquisition of Mobileye.
Automakers say this situation is not like anything they saw before. “This is the most intensive engineering work that has ever been undertaken,” said one of the car managers in an informal conversation with Design News. “And for that you need a lot of the best engineers in the world.” I’m not talking about dozens or hundreds of engineers. I’m talking about thousands. It’s about billions of dollars. “
That’s why some companies are now moving away from short-term forecasts, said the chief executive. They see the number of tests needed, engineering requirements and cost, and they wonder how long it will take.
“The automotive industry gets scared as if you’re getting ready for something, and then you have to go on stage and really do it,” said Mike Ramsey, senior director and automotive analyst at Gartner, Inc. “And then you suddenly realize:“ Maybe I’m not as ready as I thought. ”
However, not all automakers are pushing the deadlines. CEO of Tesla Inc. Elon Musk has maintained the belief that his company will create a fully autonomous car in 2020. “I think it’s safe to go to bed in the car and wake up at your destination by the end of next year,” he said in a February podcast. Most recently, he confirmed this statement, saying that in 2020 he plans to bring more than a million robotic taxis to the roads. The key point, he said, is that Tesla can test its autonomous driving technology more efficiently, as it accumulates “100 times more miles per day than everyone else.”
Privately, most engineers doubt Mask’s claims. However, they prefer not to speak out and remain in the shadows. Some hint at the emergence of a full 5th level in the late 2020s or early 2030s. But overall, automakers say they’re no longer predicting.
“From the very beginning, we knew it was hard,” one engineer told us. “That’s all we can say.”
However, almost every automaker and supplier is moving forward at full speed. “This is inevitable,” Sellars told us. “It will happen. The only question is how much time will pass before we can enter the dealership and buy a level 5 car. ”
We are perhaps the most powerful competence center in Russia for the development of automotive electronics in Russia. Now we are actively growing and we have opened many vacancies (about 30, including in the regions), such as a software engineer, design engineer, lead development engineer (DSP programmer), etc.
We have many interesting challenges from automakers and concerns driving the industry. If you want to grow as a specialist and learn from the best, we will be glad to see you in our team. We are also ready to share expertise, the most important thing that happens in automotive. Ask us any questions, we will answer, we will discuss.
Read more useful articles:
- [Прогноз] Transport of the future (short-term, medium-term, long-term horizons)
- The best materials for hacking cars with DEF CON 2019-2020
- [Прогноз] Motornet – a data exchange network for robotic vehicles
- Companies spend $ 16 billion on drones to capture 8 trillion market
- Cameras or lasers
- Autonomous cars on open source
- McKinsey: Rethinking Software and Electronics Architecture in Automotive
- Another OS war is already under the hood of cars
- Program code in the car
- In a modern car, there are more lines of code than …