Technology and Trust: The Old Driver Explains the Four Problems Concerning Autopilot

Lei Feng network (search "Lei Feng network" public number concerned) : This article by the author of the Green Water, autopilot industry practitioners.

On one day of a certain year in a certain year, the term “driver” will have a completely different meaning (it is not an old driver). Imagine that a car became a feast for the flow. The guests exchanged their joys and entanglements, and they felt that they had been on the move ... and stopped. It took decades. Well, what is it that stands in front of us, leading to the rush of artificial intelligence? Today, far from the future of automatic driving, is it still a long way off? I am afraid that only one practitioner can understand this important level. Thus, we summarized the following challenges from the technical level (car issues) and the psychological level (human issues).

1, the perception of the environment

A long-standing misconception about Google's self-driving car (hereinafter referred to as G-car) may be that it decides on the basis of the situation it sees in real time. But in fact, every step it took was engineer-washed - they would go back and forth several times on the target route, and draw a detailed 3-D map to teach G-car. Obviously this is a very time-consuming project, so why should we do this?

The answer is to save computing time and energy to deal with emergencies in pedestrians, traffic, and other kinds of driving in the field. However, Google’s 3-D map currently covers only a limited number of experimental fields, such as Jingshan, California. Until the day when the autopilot hits the road, the map must not only be covered in large areas, but also updated at any time. This workload is self-evident.

However, the plan is always out of step with the change. The Oxford University researchers conducted a one-year record of a six-mile street feature in England - the conclusion is that the street scene is always changing, and a single turn to the road sign has moved three times. If you take into account the changes in the weather, the difficulty is even greater.

For example, in Hokkaido, Japan, which is often covered by heavy snow, to indicate to the driver the width of the road surface, the small arrow pointing to the shoulder in the figure below can be seen everywhere. If you still have an impression on the movie “Worry Evening,” you may remember that the smiling girl (Shu Qi) once told you what they were doing, but what if your car did not understand it?

For this issue, Google said it was confident because as more and more autonomous vehicles got on the road, there was a steady flow of surface information back to the drawing team. Of course, this process will be much faster if the individual company's autonomous driving R&D team can share information.

From another perspective, drawing ahead is not the only way out. Iron Man's Tesla is more inclined to rely on real-time processing of image and sensor information. However, at present, Tesla is mainly a driver-assisted driver, and has not yet adopted the G-car so aggressive drone. In the long run, it is still not known which way is more stable.

2, understanding of intention

As a driverless car, it's not enough to just look around and you need to know what is happening: A plastic bag wrapped with your dream of getting rich in the air is flying in the wind and you need to dodge it; When a pedestrian looks around while talking, does she decide to cross the road? The gesture of the traffic police reveals the rhythm of the mechanical dance. What is this ghost? These conditions place very high demands on computer vision.

Just like a bad-tempered schoolmaster rants when he explains his exercises: Why is it so simple to understand? This is actually a dilemma that is often encountered when humans understand machine vision or general speaking machine learning. That is, the machine is good at dealing with difficult problems we think of such as playing chess, but more often it faces like a nerd. Real life is at a loss, especially when it comes to extrapolating the decisions of others.

As it was on Google's Aventure on February 14th this year, it was naive to think that the bus coming from behind would give way to it, and this timeward way to go is:

The good friend Tesla couldn't bear to watch G-car alone, so it was just last month and it crashed. The reason was that a moving car in front of it changed course in order to avoid a small truck parked on the road, and Tesla, who was standing straight behind, immediately ran into the minivan. It seemed that the front car did not understand it.

It is true that similar human drivers are unavoidable. It is not a requirement that a self-driving car be seamless in its development. The general demand of the audience is: Better than the human driver. But only this point, it is not accurate.

It must be noted that Google Motors has accumulated more than 1.3 million miles of driving experience since 2009, compared with the fact that American drivers have traveled an average of more than 100 million miles in serious traffic accidents that caused deaths. A sample of such a small amount of autopilot is far from enough to drive traditional measurement methods to give a reliable estimate of its statistical safety performance. In short, it still has a long way to go.

At the same time, it is also necessary to develop new methods for detecting the safety performance of the automatic driving technology. Otherwise, at this speed, experiments will never come to a conclusion.

3, the "people - car interface" control

This kind of good medicine for autopilot is mostly used by electric vehicles. This is due to the fact that the motor is easy to control precisely. On the other hand, it is also a high degree of coincidence between the two audiences and it is easy to promote. As long as the previous two issues are handled well and pedestrians fail to put on a variety of classic moral dilemmas (such as the trolleyy dilemma) to make things difficult for them, “driving” this thing is not difficult in itself.

After the popularity and networking of automatic driving, it will be a gratifying scene. Line optimization, car sharing, and intelligent transportation are no longer dreams. Not to mention the once-scale effect has formed, and pedestrians and road traffic regulators are bound to make adjustments accordingly. Such an autonomous driving friendly traffic environment will in turn Relieve its difficulties in perception and understanding and form positive feedback.

Prior to this, even the most advanced self-driving cars would occasionally be trapped in environments beyond their intelligence. By then, it will depend on human beings to take over. This led to the problem of autopilot or, more generally speaking, assisted driving: handover .

How does it shake the spare tire driver from his sleep (below, the Tesla driver is asleep when he is cruising automatically), pull it back from the movie back to reality, or interrupt his call and get his attention? To know that a person who is immersed in the road chase film and stares at the serial crash shot, it is not unusual to miss any voice light tips.

Eventually people find that it's still awkward to do - vibrate - just 0.8 seconds allows a passenger who is focused on returning mail to switch to "driver" mode. At the same time, what should the vehicle tell him about the location of the crisis?

Previous studies have shown that "the belt with a tactile feel, the person's perception of the vibration in its horizontal direction can be accurate to 10 degrees," so you can stimulate the corresponding body parts according to the spatial orientation that people need to see. However, even though the vibration warning can effectively achieve this, letting it work with other complicated signals such as lights, buzzers, and voices may make people feel at a loss: the brain is overloaded by the “alarm flood” and the monstrosity is... ...

In order to avoid the above situation. Reasonable advice is that different warning modes are used at different stages of the human takeover process. For example, using vibration to arouse people's attention, and then use a visual method to inform the dangerous position, so that at the same time take advantage of both: the tactile sensation makes people react quickly, and the visual accurately informs the situation to be dealt with.

But I think that Imagine you are in the car, eating hot pot singing songs, and suddenly the love of the car is shaking your heart a tight, this time, whether it is you want to deal with irrelevant troubles, or give you a choice of life and death This is not a good experience, nor is it the original intention of automatic driving.

However, optimistically, technical issues have been improving. As long as its growth rate has not decreased, it will only be a matter of time before a fixed target is reached. Another problem is how to make people accept this new thing?

If you think of the solution to the problem of the car, you must be able to respond to the human camp. The first thing to face is, Are we ready? Technology giants and motor vehicle manufacturers have spent billions of dollars on autonomous driving. They desperately hope that the public will trust such cars.

However, according to a survey conducted by the University of Michigan Transportation Research Institute, the issue of trust in self-driving cars is still a driver's heart disease: Three-quarters of U.S. drivers are afraid or worried about riding a self-driving car.

When asked if they are willing to buy or rent a self-driving car, their answer is not optimistic. More than one-third said: None of the doors.

The reason is that safety issues are still the most concerned. The good news is that this doubt will be gradually dispelled. Volkswagen had previously done a study called the "Wizard of Oz," which allowed volunteers to ride cars driven by human drivers, but at the same time made them mistakenly thinking that it was a machine-controlled, self-driving car. Although tense at the start, passengers were basically able to relax their journey of anxiety after ten minutes.

Similar research on trust, Virginia Institute of Transport Studies also has a copy. They recruited dozens of volunteers, the experiment requires the measured personnel to remain vigilant, observe the road and surrounding conditions in order to control the car if necessary. However, the vast majority of drivers said that they quickly relaxed, and their hands and feet were out of control. (Oh, see the sleeping student in front of him. Is it easy to relax and be vigilant? Is it good or bad?)

The lack of human confidence in autonomous driving comes from a skepticism about the technology itself. Another deeper fear is directed at the moral principles of those who master the technology.

In the eyes of a car radar, every pedestrian on the road is a floating "survival probability," or "impact resistance index." When it is not for it, it has to use these numbers to arbitrarily decide to "sacrifice" one of them. Which one is even more. So, assuming we are passengers in a self-driving car, what are the ethical guidelines for your car? Is the following left or right?

We hope that all the cars on the road will have a noble sentiment that can be self-sacrifice when necessary, in addition to the one I have on my own. (nani?) In fact, most people have expressed their unwillingness to buy vehicles that have been pre-written in the “Home for the Small” program. What about you?

One side is the ever-changing technology, and the other is the people who are guilty of consternation. The legislative process caught in it is called a stumble. However, that is another story. I will have the chance to talk to you about the U.S. legislation for autonomous cars.

On the other hand, the progress of industrialization is one step at a time. Which one of these minefields has not stepped on. What we have learned is that there will always be methods to solve. If the intelligence of the machine encounters a bottleneck, it will not be possible for people to change their thinking.

First of all, an endless stream of assisted driving technologies has accumulated extensive experience while cultivating the masses, and piloted fully automated special zones in small areas to bring points to the surface, until the time is ripe to form a united front and strive for full victory .

An interesting story tells us that on every 911 anniversary, the World Trade Center site will erect two pillars of light up to several kilometers, clearly visible over a hundred kilometers away, and for the migrating birds. Their biological navigation system mistakenly believes that this is a natural light source, so that it can't escape from its circling. Later, at the suggestion of ornithologists, the light beam will be shut down at regular intervals for 20 minutes and fly.

Our relationship with the machine is also a period of learning: the machine learning to drive, and humans, learning to walk with the machine, there must be two outstretched hands before they can be friendly together.

PP Self Closing Sleeve

PP Self-Wrapping Sleeve,PP Self Closing Wrap,PP Split Sleeving,PP Self Closing Sleeve

Dongguan Liansi Electronics Co.,Ltd , https://www.liansielectronics.com