Shulou(Shulou.com)06/03 Report--

As more and more autopilot manufacturers begin to enter the field testing stage, the safety of autopilot has attracted more and more attention. After two years of desensitization therapy, we have gained some immunity to autopilot accidents without making a fuss. The solutions to safety problems in the industry are also gradually falling into systematization.

Just yesterday, 11 companies, including Aptiv, Continental, BMW, Audi, Durham, Intel, Baidu and others, issued a 157-page "White Paper on Safety in Autopilot" in an attempt to explain and regulate various safety issues in autopilot.

Since last year, Apple and Waymo have successively launched self-driving safety reports with similar concepts. What is the difference in this safety white paper, and where are the safety regulations for self-driving now?

Is it the human fault or the car's fault? The cause of an unexpected accident

Before we discuss the safety regulations of autopilot, let's take a look at: what are the problems with autopilot at present?

When autopilot really starts the field road test, many of the safety questions are imagined or predicted by people in an ideal environment. We imagine whether self-driving cars will be hacked and whether self-driving technology will not be able to cope with emergencies.

But in the process of continuous testing, it is found that there are frequent errors in the "coordination" between humans and autopilot before hackers, viruses and emergencies occur.

Since the Delphi road test accident in 2014, the California vehicle Administration has been continuously recording and updating the cause of self-driving car test accidents. In 2018, the media sorted out a total of 51 self-driving accidents recorded at that time. It turns out that only 14% of the accidents are self-driving cars, and most of the rest are ordinary cars. In 14% of these accidents, human safety officers are involved, that is to say, even human intervention has not been able to reverse the situation. And four of the 51 accidents were due to "improper intervention of security officers". In other words, the safety officer intervened in driving out of distrust of autopilot, resulting in a car accident.

In one case, the Waymo was "jammed" by a car during the test, and the security guard changed the lane to the right in order to avoid the stopper. As a result, the speed of the car coming from the right rear was very fast, which caused a rear-end collision to the Waymo.

In fact, from these cases, we can find that the safety regulation of autopilot is moving from "technical safety" to "program safety". In L3 and L4 autopilot, which has relatively strong applicability and high landing, it is not only the technology that poses the safety problem, but also the human safety officer's understanding of the autopilot system and the degree of cooperation between the two sides.

What did we read in the white paper on autopilot safety?

This is also reflected in this white paper on autopilot safety. The self-driving safety white paper divides the safety regulation of self-driving cars into three parts: research and development, testing and operation. Twelve general principles of autopilot are put forward, including safety operation, operation design, active switching of autopilot system / safety officer, safety, user responsibility, active switching of vehicle, interdependence between safety officer and autopilot system, safety assessment, data recording, passive safety, traffic behavior and safety layer.

Compared with the previous autopilot safety reports from Apple and waymo, which highlighted technology development and testing specifications, and the U.S. Department of Transportation's "Autopilot system 2.0: safety Vision" from the perspective of government legislation. In this white paper on autopilot safety, we can see the following differences.

1. Clear division of rights and responsibilities of people and vehicles.

The white paper devotes a lot of space to describing "what cars should do" and "what people should do" in different levels of L1-L5 autopilot systems, and when human drivers should take over vehicle control-if human drivers operate the car in other states, the autopilot system can be exempted from liability in the event of a safety accident.

2. the design requirements of the vehicle itself.

Perhaps it is because many car mainframe manufacturers have participated in the formulation of the white paper, which also puts forward suggestions on the safety design of the vehicle itself. For example, in the future, with the combination of V2X technology, where will the information appear in the car is the safest? Is it on the screen of the car entertainment system, or is it the dashboard?

3. Explanation of the behavior of autopilot system

The white paper also specially emphasizes the "interpretability" of the behavior of the self-driving system, which does not refer to the interpretability of the black box, but to enable pedestrians, bicycles and drivers on the road to know what self-driving cars want to do. In some cases, the driving habits of the autopilot system are different from those of human beings, just as in previous accidents, when someone is jammed in front, many human drivers subconsciously choose to change lanes, but the autopilot system does not necessarily do so. When autopilot has the potential to break the hidden rules of driving in the human world, it is important to let all people on the road know what they think.

Why did technology alliances evolve into security alliances?

From this joint release of the White Paper, we can also see an interesting phenomenon.

The release of this white paper is not too much to say that it has gathered half of the autopilot, especially to firmly grasp the mainframe manufacturers, which is the final link facing the market. Although the significance of the white paper itself is to provide advice and guidance to the market rather than to establish barriers to entry to screen out other suppliers / manufacturers, we might as well consider how much expert advice will be drawn from such a white paper when government departments intervene in the safety regulation of autopilot in the future. For example, when the United States Transportation Administration formulated "Autopilot system 2.0: safety Vision", it referred to the autopilot safety report put forward by Waymo.

At the same time, many of the manufacturers who have jointly issued the white paper have begun to promote public education on self-driving safety, and the methodology is naturally the same as that in the white paper. If these guidelines become both policy considerations and public awareness, they still invisibly set a threshold for entry for autopilot applications.

In particular, we can see in the white paper release list, Intel, Durham, BMW …... In fact, these names are often combined a few years ago, working together to develop a self-driving technology platform, in opposition to manufacturers such as Waymo.

Now this kind of technology alliance has been transformed into a "security standards alliance", which has bought more time for the factions of BMW and Intel. If we can grasp the security gates, especially the safety gates of field testing, the technology of other manufacturers can only be "besieged" in the laboratory.

Of course, the publication of this white paper does not play such a strategic role. But what is certain is that in the future, self-driving manufacturers and car manufacturers will have more and more participation in the formulation of self-driving safety standards.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.