Tesla Rashomon: How far is it from true autonomous driving?

The automobile industry revolution is happening again a hundred years later. Who will take the lead?

Change Maker

No.20

A 315 gala has put Tesla in a Rashomon again. Tesla is the only one to become the focus of public opinion by failing to make the list. But this is not the first time Tesla has encountered a public opinion crisis. Compared with the past "fatal accidents", the "brake problem" is not even a crisis. The first of the five fatal accidents involving Tesla's autonomous driving occurred in China. On January 20, 2016, a serious rear-end collision occurred on the Beijing-Hong Kong-Macao Expressway in Handan, Hebei Province. A white Tesla crashed into a road sweeper operating in front, and the driver was unfortunately killed. In 2017, Tesla admitted that the vehicle was in autonomous driving at the time of the incident. 

Four months after the Handan accident, a similar fatal car accident occurred in Florida, USA, where a Tesla Model S crashed in autopilot mode, killing the occupants. Tesla explained that the system did not "blind" the white truck in the face of a bright sky, and further stated: "Autopilot cannot prevent all accidents - it is impossible to set such a standard, but it can minimize the occurrence of accidents. It will undoubtedly make the world's car owners, pedestrians and cyclists safer."

In January 2017, the National Transportation Safety Board (NTSB) of the United States issued a report on this, finding that Tesla was not at fault. The investigation was more favorable to Tesla. After installing the Tesla autopilot system, the crash rate dropped by 40%. People who support autopilot believe that 37,000 people died in car accidents in the United States in 2016, of which 94% were due to driver errors. Humans are so likely to make mistakes, why not replace them with more reliable robots? Tesla becoming the center of public opinion on 315 in China does not mean that Chinese people really don’t like Tesla or autopilot. Perhaps they are just disgusted by the arrogance of the company. In terms of attitude towards autopilot, Chinese people may be more tolerant and optimistic than Americans. After the fatal Tesla accident in California in 2018, iResearch’s public opinion report showed that American experts generally expressed dissatisfaction with Tesla. Chinese experts, on the other hand, expressed more expectations for the further maturity of autopilot, and Chinese netizens were generally optimistic about the future of autopilot. A public opinion survey by Cisco Systems also showed that China, Brazil and India are the countries most willing to trust automation technology. 

According to US media reports, when Trump met with corporate CEOs, he told Musk that he "prefers traditional cars". When talking about autonomous driving at the Bedminster Golf Club, Trump also said: "It's really cool, but I will never use an autonomous car... I don't trust a computer to drive for me." This is probably the attitude of most Americans. According to survey data in 2017, 71% of American drivers are concerned about autonomous vehicles. As of 2020, this concern still exists on a large scale. According to the PAVE report of the American Association for Autonomous Vehicle Education, 48% of Americans said they would never ride in a self-driving taxi Robotaxi. 

In the American TV series "Upload", the protagonist's self-driving car "cannot recognize" the truck parked in front of it. The protagonist tried hard to switch to manual control, but the car refused to return the "steering wheel", and the protagonist eventually died in a car accident.  

It is not certain whether Upload was inspired by Tesla's fatal accident, but the protagonist's death reflects the concerns of humans about autonomous driving: machine failure, AI deliberate "murder", and hacker intrusion. Before AI autonomously controls cars, the current distrust mainly comes from "machine failure". 

In the early days of autonomous driving, such concerns were not without reason. In March 2018, an Uber self-driving car hit a woman crossing the road during road testing in Tempe, Arizona, killing her. 5.6 seconds before the accident, Uber's self-driving car had detected the woman, but the system mistook her for a car. 5.2 seconds before the accident, the system classified her as "other", thinking she was a stationary object. The decision-making time was missed by wavering between "car" and "other".  

1. Milestone 2021: A batch of L5 models emerge

John Markov wrote in his A Brief History of Artificial Intelligence: "The DARPA competition marked the dividing line between two worlds. In one world, robots were seen as toys or playthings of researchers; in the other, people began to accept the fact that robots could move freely in the world." 

In the last century, Japan's University of Tsukuba, the United States' Carnegie Mellon University and Germany's Federal University all explored autonomous driving to some extent. But the real breakthrough was made by DARPA (Defense Advanced Research Projects Agency). The US Congress instructed the military to design autonomous vehicles, but the military worked for several years without any progress. Tony Tesse, then director of DARPA, took a different approach and held the famous urban challenge DARPA Grand Challenge for hackers, scientists and companies in society. 

The millions of dollars in prize money and priceless prestige attracted dozens of teams from academia and industry. Stanford University professor Bastien Thrun organized a team of car company engineers and student hackers, and the following year completed the unmanned driving that DARPA had not been able to develop for several years. (See the previous article "The History of American Technology "Going to Sea") 

Autonomous driving is the epitome of artificial intelligence technology, with three system units: a perception unit that includes sensors and environmental algorithms, a decision unit that includes decision models and decision results, and a control unit that includes control interfaces and control behaviors. Specifically, it requires a large number of software and hardware integration technologies such as perception ranging, lane keeping, stereo vision, high-precision maps, inertial navigation, vehicle-road collaboration, road condition decision-making, and real-time analysis. 

Basic conceptual model of autonomous driving system In 2009, Google launched the autonomous driving car project Chauffeur under the leadership of Thrun. Google's move triggered a chain reaction. Major automakers such as BMW, Mercedes-Benz, Volvo, and Audi quickly mobilized their forces to develop autonomous driving systems. Uber, Apple, and Nvidia also joined the game. In an inexplicable panic, Intel acquired Mobileye, an Israeli computer vision system manufacturer, for a huge sum of US$15.3 billion. 

Mobileye is indeed an alternative to autonomous driving. Unlike Trond’s precise thinking path, Mobileye’s founder Amron Shashuva has brought another possibility for autonomous driving. Trond mainly relies on the increasing performance of computers to improve physical recognition capabilities, while Shashuva uses biological methods to study computer vision. John Markov commented that the experience of Google’s autonomous driving is a sense of separation between people and machines, while Mobileye allows passengers to keenly feel the presence of machine assistance. 

There is no doubt that humans have begun the chapter of autonomous driving, but there is still a long and difficult road to go before we can achieve the autonomous driving we imagine.

In 2016, the Society of Automotive Engineers (SAE) updated the definition of autonomous driving classification, dividing automation into 6 levels.

SAE defines 6 levels of autonomous driving technology

my country's standards are basically consistent with SAE standards. The only difference is that we call L0 "emergency assistance". The driver can take control of driving independently, the car system can sense the environment and provide alarms, assistance or brief intervention in driving, thus achieving separation from non-driving automation functions. 

Simply put, L0-L2 is controlled by real drivers. L1 is the "hands-on" level, L2 can be appropriately "hands-off"; L3-L4 real drivers and machines jointly control; L3 needs to "keep your eyes open"; L4 can be "basically assured"; L5 can choose to be completely controlled by the machine. From a commercial perspective, L5 is a long-term goal, L3 has limited applications, L2 and L4 are the best applications, and L2 is currently in large-scale commercial use. In 2020, the chairman of the US NTSB publicly stated that there are no autonomous driving cars above the L3 level in the US market. However, on March 5, 2021, Honda began to provide a limited edition rental service of 100 Legend Hybrid EX sedans in Japan. These sedans are said to be equipped with L3 autonomous driving equipment that has been certified by the Japanese government. 

The biggest challenge facing L2 autonomous driving at present is the distraction of real drivers, that is, how to enable distracted drivers to switch to operation immediately in an emergency. Many fatal accidents of Tesla are also caused by this. Automakers have given various solutions. Lexus and Mercedes-Benz monitor the position of the driver's eyes and head to determine whether they are dozing or distracted. Audi uses two cameras to detect whether the driver is concentrating, otherwise the autonomous driving system will terminate. 

In 2017, Audi launched the world's first L3 car A8, and released the L4 concept car Elaine and the L5 concept car Audi Aicon.

In 2018, Toyota launched the autonomous driving travel platform e-Palette, and plans to expand the scope of application of advanced autonomous driving to ordinary roads from 2025 to 2029. In 2019, GM began mass production of the L4 autonomous driving car Cruise AV. 2021 will be a milestone year for autonomous driving. This year, Ford plans to launch an L4 autonomous driving car without accelerator pedals and steering wheels, and no control of the vehicle is required in specific areas. Daimler plans to test L4/5 autonomous driving cars. BMW plans to start testing L5 autonomous driving cars, and Volkswagen also plans to launch the L5 autonomous driving car Sedric.