Chips and algorithms have exploded, but you tell me that autonomous driving is going to be a failure?

Autonomous driving is undoubtedly the hottest technology in recent years. Using programs to let cars drive on a set route is a plot that only existed in early science fiction movies. With the rapid development of chip performance and the advancement of algorithms, do you think that autonomous driving is just around the corner? The facts tell us that autonomous driving may be far worse than you think.

"The flood flooded the Dragon King Temple"

01 Autonomous driving needs to be upgraded

According to the classification system proposed by the National Highway Traffic Safety Administration (NHTSA) and the Society of Automotive Engineers (SAE International), there are five levels of autonomous driving, which we generally refer to as L1 to L5, and L0 is fully manual driving.

Different levels of autonomous driving (Image courtesy of Xpeng Motors)

The realization of autonomous driving requires the combination of hardware and software. The hardware includes environmental perception tools such as cameras and radars, as well as chips for data processing; while the software is mainly algorithms that process and judge the collected information and output it to the car to accurately control speed and steering.

So the method is very simple. From the hardware perspective, add sensor equipment to improve equipment accuracy, while strengthening the processing capabilities of the chip to minimize power consumption. From the software perspective, conduct more road test analyses, continuously optimize the algorithm, and develop in the direction of autonomous driving.

02 Report good news instead of bad news

From the news media we have access to, many automobile manufacturers or solution providers are only reporting good news and not bad news: some claim that their technology has reached L3, and some say that autonomous driving can be achieved in 20XX.

Waymo's driverless car is avoiding pedestrians (Image from Waymo)

No matter how it is promoted, the cruel reality is that the vast majority of self-driving cars on the market currently remain at the L2 level, which means that the driver needs to be ready to control the vehicle at any time. When the car detects a situation where a driver is required, it will immediately return to let the driver take over its subsequent control. In the final analysis, it is still assisted driving.

The technological status of different manufacturers in autonomous driving (Image from Navigant)

When the recognition and algorithm do not meet the requirements, forcibly using autonomous driving will bring many problems. In addition, different road conditions such as night and rainy and snowy weather have more impacts on autonomous driving, which leads to the fact that we often see traffic accidents involving "autonomous driving" cars.

03 Hardware upgrade to overcome bottlenecks

To overcome this bottleneck, on the one hand, chip performance needs to be improved to bring about a significant increase in processing power, and on the other hand, hardware support is also needed. Many cars on the road are equipped with more than 20 lidars and ultrasonic radars to provide longer intervention-free mileage.

Uber's self-driving car overturned (Image from cnsteem)

In traffic accidents involving autonomous driving, a large proportion of them are caused by unpredictable actions of autonomous vehicles. Although these are not caused by algorithms, they also need to be continuously upgraded and optimized.

Uber's self-driving car caused the death of a pedestrian (Image from CNN)

In addition, high-precision maps are also the basis for realizing autonomous driving. With sufficiently accurate road and slope information, autonomous driving can better handle and respond to different problems, and can also better navigate normally in different weather conditions.

NVIDIA's autonomous vehicle development platform (Image courtesy of NVIDIA)

The problem that follows is that the result of relatively perfect autonomous driving is a sharp rise in costs. The good news is that consumer electronics chip giants such as Intel and Nvidia have also entered the autonomous driving chip industry, and many emerging brands have joined the market competition, which can make practical autonomous driving landed faster.

04 Autonomous driving is a promising future...or not?

However, the biggest problem for autonomous driving is not the complex road system, but other vehicles and pedestrians who do not obey traffic rules. Crossing the road randomly, making illegal U-turns, forcibly changing lanes, etc., will catch the autonomous driving off guard.

Google's self-driving car was hit by a car that ran a red light (Image from selfdriving)

For car manufacturers equipped with autonomous driving, difficulty in determining liability is also a problem. If the responsibility for traffic accidents caused by autonomous driving cannot be clearly defined, the manufacturer's strategy is not to open up too high a level of autonomous driving to drivers, or even to avoid vehicle defects, leading to adverse effects.

Autonomous driving is accomplished by a lot of software and hardware (picture from vision systems)

Another issue that cannot be ignored is that the autonomous driving program can be tampered with. Once used maliciously, it will cause serious consequences, which many users cannot afford.

Considering the current technology and road conditions, the better use scenario for autonomous driving is on closed/semi-closed roads with high-precision maps and a high coverage rate of autonomous driving, which can allow autonomous driving to better play its advantages.

As ordinary consumers, what we need to do is: recognize the connotation of autonomous driving when purchasing a car, choose the vehicle and configuration according to our actual needs, and don’t pay the "IQ tax" for flashy functions; when investing, we must avoid being misled by the manufacturer's gimmicks, so that the technology that should serve us becomes a tool for hype and cutting leeks.