Thursday, August 4, 2016

In the eyes of a longtime user of Tesla Autopilot

Lei feng's network (search for "Lei feng's network" public attention) by writer Xia Heng, Xiaopeng cars founder and CEO.

In the eyes of a longtime user of Tesla Autopilot

Soon in the United States there was a murder case, an ex-American poster player named Joshua Brown hit the road. Puzzling is that after this seemingly ordinary murder has aroused world-wide attention. What reason? Let us go out to the crime scene.

Crime time: the afternoon of May 7, 2016

Place: United States, Florida, highway

People: Joshua Brown,18 wheel of large trucks, and its Autopilot – Tesla (autonomous) systems

This is the world's first automatic driving accident deaths, the accident directly push the current hot autonomous vehicle technology to the forefront of public opinion.

Many people cannot help but ask, Tesla did not claim automatically driving? The responsibility of the accident should be Tesla Autopilot? In this regard, the United States highest auto safety regulator--NHTSA (National Highway Traffic Safety Administration, United States Highway Traffic Safety Administration) is also investigating when Tesla Autopilot systems work, but it certainly was not responsible for monitoring driver for which you want to.

Starbucks iPhone 6 Plus Case

Long-term users of the author as Tesla's Autopilot system, since Tesla launched the autopilot functions, each version has first-hand experience. On Tesla's autopilot system security, I would like to talk about my own views.

In the eyes of a longtime user of Tesla Autopilot

| Tesla's automatic drive hint

1) Tesla's manual. Tesla's manual automatic driving accessibility has a huge red exclamation mark warning. For example:

Lane assist: Lane assist provides guidance only, do not replace direct Visual observation. Drivers have the responsibility to remain vigilant, pay attention to travel lanes, and always look out for other road users. Otherwise, you may cause serious damage or casualties.

Active cruise control: active cruise control is designed for comfort and convenience, not part of collision warning and avoidance system. Drivers have the responsibility to remain vigilant, safe driving, and command vehicles. Be sure to look ahead and be ready to take corrective measures. Otherwise, you may cause serious damage or casualties.

Collision-assisted: front collision warning provides guidance only, cannot replace the focus on driving and accurate judgment. Be sure to observe the road while driving, do not rely on front collision warning on the possibility of a collision warning. Depend on warn before collision warning of potential collisions may result in serious personal injury or death.

2) Tesla instruments. Tesla offers is automated driver assistance technology, rather than drone technology, Tesla is still require human supervision. Automatic driving modes that can also be opened from the Model s show reminds language after confirmation: "always hold on to the steering wheel, ready to take over. If the system detects that your hands are not placed on the steering wheel, it will even issue a warning, and gradually reduce the speed until your hands back on the wheel again. "

3) Tesla constantly reminding drivers to take over. When Tesla opened automatically after driving, even when good driving condition, will jump out of the screen automatically after about 15 minutes, and sounded a beep alarm, holding the steering wheel. Otherwise, the vehicle will slow down to stop. This initiative is also in order to prevent the dangers of driver let up.

In the eyes of a longtime user of Tesla Autopilot

Tesla's autopilot function alerts the user to hold the steering wheel

| Repeat cases restore the truth

Let Tesla companies and the news media to disclose information restored to the scene of the accident.

On May 7, the pleasant climate of the Florida Sun is shining, a Model s-driving in a flat on the highway. Dashboards display current conditions with Autopilot, owner of Joshua Brown launched his favorite autopilot functions, handed over control of the car, vehicle is in good condition, everything is alright. In the vehicle interior, are periodically collected information of multiple sensors to perceive the surrounding environment, the intelligent control system based on environment parameters to make decisions, control vehicles. Installed in front of the front of the millimeter-wave radar to monitor the position and velocity of an object, Mobileye Q3 camera on the windshield is staring straight ahead, judge vehicle around and obstacles. At this time, around the corner in front of a white trailer body is crossing the road. Only more than 100 meters from the Tesla. Sunlight is too strong, white body makes it difficult to distinguish the front is a car or a cloud in the sky.

"Well, is the car? Does not seem to move, some high, seemingly not a car? Hi, camera, what do you think? " Radar asks.

"In front of cars? Well, can't you see, quite open in front, clouds. "Camera said.

"It should be hung a large sign in front, how do I position a bit high? Well, no traffic in front, you can and move on. "

In this way, it fooled the Tesla's Autopilot system.

After a brief few seconds, tragedy struck. Model s into the bottom of the trailer, its front windshield and trailer bottom impact, the pilot died on the spot. Starbucks iPhone 6 Plus Case

Due to the traffic accident, these are based on Tesla extract information of vehicle records results of restoring Tesla's official statement: "under the intense sun, the pilot and the autopilot had failed to notice trailer white body and therefore did not start brake system in a timely manner. Because the trailer is crossing the road, and tall, this special case led Model s from the bottom through the trailer, its front windshield and trailer bottom impact. "

So in the case of Tesla's statement is true, we can be sure that the above is true after the accident. To sum up, the cause of the accident are two aspects: one is the perception module failure aboard Tesla, master dereliction of a second driver.

Sensing module failures, mainly consists of two aspects:

First of all, false detection of millimeter-wave radar trailer body signs. According to the description of Elon Musk, millimeter-wave radar to estimate a huge obstacle ahead, due to the large reflective area trailer and high ground clearance vehicle, from the perspective of millimeter-wave radar, it's more like a hang over a road traffic sign.

Excessive body and large reflective area is understandable, but location is detected flying over a road there are questions. Millimeter-wave radar is used to detect obstacles on the road information, not hanging over a road traffic sign information, simply by using millimeter-wave radar is difficult to determine what object is in front, can only judge the obstacle's shape, size, and location. Height of the trailer there is no road sign high, so when the millimeter wave radar detects obstacles when judging traffic signs, freak.

In the eyes of a longtime user of Tesla Autopilot

Accident scene and accident

Second, the visual perception module failure. Tesla is the Mobileye EyeQ3 camera sensors aboard, some have speculated because at the blue sky and white clouds, camera body defined as white clouds of the Visual system, although this is only speculation. We can confirm in accordance with Tesla's statements, white body errors of judgment of the Mobileye EyeQ3 is crossing the road, failing to identify obstacles.

According to these two synthetic judgments, many coincidences are brought together, resulting in the accident. Deeper technical reasons, may be radar modules with the visual perception of lack of fusion algorithm for systems validation, bug or algorithm.

From the driver's point of view, Tesla drive demand drivers remain vigilant and ready to take over. Tesla in its own manuals, disclaimers, as well as driving constantly warn. But according to reports, when Brown is likely to see the Harry Potter movie DVD in the car, rather than observing the road.

If this is true (and only has died Mr Brown knew what he was doing), in this case, the main cause of accidents can be categorized as Mr Brown relies too heavily on Tesla's autopilot system, control of dereliction of duty.

Tesla's Autopilot system does not detect the truck is problems existing technology is not mature, but Tesla has repeatedly reminded warned drivers to pay close attention to road conditions, ready to take over the steering wheel. Reliance and the excessive use of this feature for this potential risk did, even if only one out of 10,000 chance, also at the expense of living life.

| The status of automatic drive technology

NHTSA for unmanned technology is divided into 5 stages:

Level 0 is the 100% control over drivers and vehicles without any security system auxiliary equipment;

Level 1 are vehicles with single or multiple electronic control system of independent functions, such as automatic emergency braking system; Starbucks iPhone 6 Plus Case

Level 2 is to have at least two controls can be automated, such as combining active distance control cruise control and lane support systems; 

Level 3 is the vehicle with automatic Dodge obstacles, self guided, active control function, but motorists have the right to control;

Level 4 is a vehicle automated driving, users shall be given the relevant information only, such as destinations, paths, and so on, means that the driver can engage in other activities in the car, such as the Internet Office and entertainment or rest.

In the view of the industry, automatic driving technology development is a gradual process, the ultimate stage of development to the drone. Now, the so-called automatic driving more carts and enterprises is to join the Intelligent Assistant system, so that cars can complete certain types of "autopilot". Seen from the concept of strict, automatic and unmanned cover content and boundaries is significantly different. Under this classification, currently worldwide automated driving technology has not yet reached the 5th stage unmanned level, Level 4 provides fully automated driving capability.

Google last year began in California for the first time outside of Austin to test their research and development of self-driving cars, it only tests on public roads in the Valley before. This year, the test site increased again, Kirkland, Washington, Phoenix, Arizona has become a testing ground. Used to test vehicles were equipped with specially trained drivers, and is not open to the public. Google patents in recent years, including switching unmanned status, reminded passengers, gestures to control the car and so on. We can see that the Google environment testing autonomous vehicle technology is more roads, also undertakes a number of automatic service for research, but still failed to reach Level 4 for commercial civilian level, is only on the premise of someone monitoring for unmanned test to advance unmanned technology services.

In addition to Google, at present, most other OEM's about skill level is Level 2 stage. Tesla is also at a stage of transition from Level 2 to Level 3. In other words, as promoted by Tesla, it provides autonomous vehicle technology is the need for manual control, driving the main responsible party is the person, the so-called "Autopilot system" provides more accessibility features.

| Automatic driving posture--alert

It cannot be denied that power of Tesla's Autopilot system, brings impressive user experience and sense of fashion technology, reducing the burden on drivers, partially release the pilot pressure to address complex road conditions, but also to some extent promote driving safety. However, the autonomous vehicle technology is continually improving, still cannot completely replace human dominance, we still cannot totally trust, drivers should always pay attention to the road.

Automatic driving systems on different roads in the country, is not the same as. More on the open road, lane lines clearly visible between train less, which uses automatic driving systems with slow driving, was able to identify the current driveway, a smooth turn and the speed to keep distance, driving feel quite comfortable. But in urban rapid transit sections especially highways, it feels different, attention needs to be highly concentrated, ready to take over the steering wheel and the brake pedal, to cope with possible problems. For example, Tesla also does not recognize the traffic signal, when it encounters a crossroad, will not take the initiative and stop, requiring artificial active braking. At highway speeds, forcibly and in case individual drivers in the front line, even with turn signal reminder that Tesla was also unable to identify and respond effectively.

In General, at higher speeds, hard drive easy to stress, when driving at low speed, anticipate longer, can more easily observe the surrounding roads condition, much easier. Through the use of, I personally prefer to use the low-speed automatic features.

Especially when the traffic, stop-and-go rhythm, will make drivers feel irritable, bored or tired. If the autopilot function through the effective integration of the various sensors and artificial intelligence judgments, achieves low speed walking traffic jam, the driver's Liberation, for drivers, it must be a rare benefit.

National conditions, China and Europe are quite different. Europe road covers large ratio, wider cars less, more cross-city driving, speed naturally faster. While domestic vehicle density, especially at rush hour, City Road on the bus more than low speed driving is the norm.

Based on the above two points, China's developers should focus on low-speed automatic drive, such as automatic parking, low-speed vehicles, and so on. For OEMs, developed autopilot functions correctly, did a safe, extreme vehicles, first driving assist function, starting from low-speed auxiliary.

Automated driver assistance functions, Xiaopeng automotive solutions at this stage on the one hand is to be active cruise, lane keeping, automatic, change lanes, automatic car or overtaking a range of intelligent and automatic parking feature, and from low-speed automatic on the one hand, as well as holographic warning function, automatically at low speed driver assistance technology research and development. Due to low speed, risk control, we believe that the best starting points is to enter the autonomous areas.

Human development has always been a process of self-liberation. Also confirm the history of human science and technology, the development of any new technology is advancing despite twists and turns, the first ship sunk, space for the first time, the first nuclear ... ... Are very difficult to stop the human passion for technology exploration. We think more persistent, automatic is the direction we move, will not change just because of an accident.

Lei Feng network Note: the writer Xia Heng, article the first cloud. Please contact the authorized and keep the source and author, no deletion of content.

No comments:

Post a Comment