The biggest uncertainty for driverless cars in China is the complex traffic environment and unpredictable driving habits, argues Wang Fei in his guest contribution. Wang Fei is current editor in chief of the technology media company PingWest, an observer of new technologies, new hardware and new model areas in the Internet science and technology industry, and a student of the technology media industry.
We often think of the advanced technology of automated driving in terms of: Humans are not perfect and are prone to mistakes and biased judgments, so we’ve created a variety of technologies to help. That’s why our drivers need various auxiliary systems to help them improve their driving safety; the highest level of the auxiliary driving system is to be completely driverless. We know that driverless cars are basically mobile computers with the functions of visual perception, algorithmic decision-making and mechanical control.
Based on the above three points, we can think of the future of driverless cars as the continuous compression of technology costs, the gradual improvement of visual accuracy with optimized algorithms, and final products with driverless technology for consumers, introduced to the market on a step-by–step basis after facing a series of problems. There is no question about the path of all the car / internet tech companies in general.
We can't stop Moore's law of rapid development, and this rapid development will continue to benefit us. The hardware and the components in our automobile supply chain have also rapidly reduced our costs. Our relevant laws, too, will move forward very quickly when the technology on the market is fulfilled. If this is the case, is there no specific problem in developing driverless technology in China, just like in other countries' markets?
I think the biggest uncertainty for driverless cars in China is the complex traffic environment and unpredictable driving habits. We know that in some other countries, driving at a speed of 200KM/H or more on the highways is permitted, while China's transportation environment is full of low-speed accidents. Traffic accidents in China are disparate, but some of the main reasons include drivers not complying with the traffic regulations or each element in the traffic environment does not abide by the regulations.
Automated driving is a complex set of technologies that combines computer decision-making, network technology and the automotive industry. When we use some of the algorithms instead of the brakes and accelerators and even the steering wheel, it should essentially be a process of machine automation. We are familiar with machine automation, often within a very standard process, operating automatically with a set of complete and normative standards of conduct – for example, in the field of industrial manufacturing, it has replaced our manual labor with increased speed and accuracy.
But I think automated driving in China's transportation environment (or anywhere in the world) plays another role – along with the driver's cognition. To a large extent, it is the driver – or the driver’s thoughts – that should be replaced by the driverless car. The most sophisticated drivers in China gained their knowledge and experience by having fender-benders, collisions or accidents. This is actually about cognition. For example, at a corner where accidents are frequent, an experienced driver would anticipate the road conditions ahead, since he or she has encountered a bike suddenly crossing the road there once before. But the driverless car can’t anticipate based on experience, and it's hard for it to take early action on the next similar occasion after having had an accident. There is only a pile of code in its head, and the programmer tells it to "always check the situation in front, and if a person suddenly appears, please brake the car."
Of course, sometimes it may not be able to detect the sudden appearance of a person, and it will be too late to stop the car. In general, driverless car technology should replace drivers at the cognitive level to take on China's traffic environment. There, the traffic environment is relatively complex – there are not only new drivers, but also drivers with different driving styles and impaired drivers – at the cognitive level, the driverless car is currently a total novice driver.
People can easily judge, without really thinking about it, whether the driver in front of them is a novice, simply because the driver does not drive straight and then slows down; people can sum up driving skills in a relatively complicated traffic environment. Decisions made by a car require more precise perceptions and more complex algorithmic decision-making, but situations that go beyond algorithmic decision-making occur constantly.
People will avoid Wudaokou with its complex road conditions during rush hours, but the driverless cars may move on according to the high-precision navigation system and have to stop every now and then – a man crossing the road, stop; express tricycle passing through the narrow space, stop. People like to change lanes and pass cars driving slowly in front of them on the highway with different speed ranges, while driverless cars are more likely to speed up or slow down in the same lane. Of course, maneuver code for passing can also be written into the driverless car system with a fixed preset condition, but the situation can be more complicated when it is faced with a driver who constantly changes lanes and passes others.
If all the cars on the road were driverless cars, it would be better. Imagine that drivers with different styles and driverless cars were on the road at the same time – the driver on the left side of the road likes accelerating and passing, and the novice driver on the right side of the road stops his car from time to time; then someone who doesn’t even know how to drive may decide, "I should stay away from them" earlier than the driverless cars with high-level driving skills. In fact, that is using a cognitive level of dimension reduction to compare things on the perception level. This is in the most complex traffic conditions, and also the traffic conditions with Chinese characteristics.
Automakers and industry participants believe that driverless cars will first complete tests on specific roads or in specific environments, such as dedicated venues, with driverless bus routes and so on. From the point of view of a driverless industry researcher, it's more like automating machines under the Internet of Things. Taking automated driving from the perception level to the cognitive level is like overturning our way of thinking to deliberate about the whole thing again. It is just like adding a brain to the four mechanical, automated wheels and the brain is not for processing sensory information, but for anticipating the road traffic and for adding knowledge. I think such a brain is really needed in the traffic environment with Chinese characteristics.
Whatever the industry, at the cognitive level of the computers, more long-term research is still required, and that is why today's hottest concepts are always about artificial intelligence. For driverless cars, we should either try to drive in all kinds of road conditions, and then the programmer adds each situation as a piece of code after each accident, thus creating a collection of code that will handle all of the situations, or we should make the driverless car a cognitive car brain which will learn the driving styles and decisions of Chinese drivers and anticipate the road conditions in advance.
Most automakers in China are not aggressive, and their auxiliary driving systems are just upgrades of the pre-crash systems. Even when there are radical manufacturers who want to test completely driverless cars, they mostly choose California's Silicon Valley, where the roads look much cleaner. In China, the road conditions are more complicated than in other countries – the uncertainty of the traffic environment, unpredictable driving habits, and the existence of traffic elements that others don’t have – this is the automated driving road with Chinese characteristics.