Editor's Note: The author of this article is the co-founder of Shenzhen Polar Perspective Technology Co., Ltd., a columnist for Lei Feng Network (search for “Lei Feng Net†public number) , and walks in the double abyss of computer vision industry and artificial intelligence academic circles.
The author of this paper believes that the safety issues that affect autopilot can actually be attributed to two major issues: the question of cybersecurity and whether the level of intelligence is adequate.
In May of this year, a deadly accident occurred in Tesla's self-driving car. For a time, the safety of driverlessness was pushed to the cusp.
All along, although all autonomous driving development companies are constantly making drive test data for automated vehicles, the safety issue of autonomous driving has been a key point in the industry's internal and external awareness debates, and it is also the ability of self-driving cars to get on the road smoothly and smoothly. The key to passengers and users.
Uber Autonomous Car (pictured from yifan, taken at 2016.8.25, United States)
The problem of automatic driving safety can actually be attributed to two major problems :
One is network security . Like all smart devices, smart cars are bound to have the risk of being invaded and then manipulated. Once they are controlled by an intrusion, the consequences are far more serious than the intrusion of any current smart device.
Second is the question of whether the level of intelligence is sufficient and whether the level of intelligence in road planning, positioning, target identification, judgment, etc. is sufficient. Once the level of intelligence is not enough, the identified pedestrians are wrongly identified and the proceeding route planning is wrong. And so on, the consequences are unimaginable.
The problem of unmanned or self-driving cars in cybersecurity is arguably inevitable. Just like all smart devices, as long as the device has an IP address, it can intrude into the car via IP to control the car's engine and brake system.
Take this Jeep Cherokee as an example, the vehicle is connected to Chrysler's network center via IP. Chrysler released a software patch to fix the system's vulnerabilities before it was discovered by a hacker. The main problem is that most software engineers who develop systems do not consider security issues.
There is another real example, Chrysler spent a huge sum of money to return automakers to repair the loopholes. Because once one of the cars is invaded, the same method can almost invade all the cars. Therefore, compared to such a security crisis, the huge amount of money is also necessary.
If it is a smart refrigerator, then the refrigerator may be invaded only if the food deteriorates and the ice melts, but if the car is invaded, it will have a fatal traffic accident.
At the latest international hacking conference, there were hackers who threatened to have a popular electric vehicle capable of remotely controlling the device’s autopilot system, allowing the vehicle to operate according to the hackers’ own wishes (such as steering), and changing the radar detection distance value (etc. Detected have been hit).
Do not want to encounter network security problems, the first thought is of course isolated network, so for security reasons, the car's key systems such as the engine and brake system are best isolated from the network, but in reality, because of cost considerations, Therefore, some networked entertainment systems will be connected to the critical system of the car on the same analysis computer, resulting in all systems being connected.
If it is to be networked, in order to prevent malicious intrusion control, it will limit the need to control the system through wired connections.
In addition, in order to ensure the security of the system, it is necessary to strictly control the network authorization, code protection and the use of third-party technology provider's audit. For example, the third-party technology provider will provide the corresponding sample code (Sample Code), and these codes are used without considering any security issues, so for safety reasons, try not to directly use the test Code. Another example is the use of different network authorizations between different systems, which is also commonly used to improve security.
Intelligent misjudgment problemIn addition to network security issues, the most compelling driverless safety issue is the accuracy of intelligent judgments. In May of this year, Tesla’s fatal accident occurred in self-driving cars. After investigation, he discovered one of them. The reason is that the car judges the front of a white carriage to be a sky. As shown in the figure, the “artificial intelligence security†problem that has attracted much attention in the field of artificial intelligence, how can we obtain safe artificial intelligence to avoid similar accidents? Not happening.
Tesla Car Accident
From the software algorithm level, currently in the world, the safe AI organization dedicated to safety is best known as OpenAI, and they are committed to making artificial intelligence not to make fatal mistakes.
In human eyes, some subtle changes in the image may not affect us to judge a thing, but in the eyes of the machine, it may be completely considered as fooling samples, so as to make different responses. Behavior, this situation is particularly prone to training the target through deep learning and video data under the monocular camera.
Tesla's accident not only exposed the shortcomings of autopilot technology in software algorithms, but also exposed shortcomings in hardware technology.
From the hardware device level, monocular cameras like the Model S have flaws in the recognition of three-dimensional and large-area planar objects, limitations in the identification interval of millimeter-wave radars, and trade-offs and redundant judgments for comprehensive scenarios in extreme cases. .
Therefore, the accident will promote the hardware and algorithms in the field of automatic driving. The comprehensive adoption of binocular/wide-angle cameras, millimeter-wave radar, and laser radar will hopefully usher in ice-breaking. Especially for the previous high-cost lidar + multi-camera program will bring impetus. In addition, Tesla's accident can be avoided if it adopts a multi-detector redundant judgment mode.
Therefore, for safety reasons, NHTSA (United States Highway Safety Administration) may also urge the relevant departments to impose higher thresholds and regulations on hardware configuration and system requirements. This means that for the industry, it means that the value-added of bicycles Promote.
Although auto-pilot safety hazards still exist, and there are inevitable risks, but one thing can be confirmed is that compared to the current 94% of traffic accidents caused by human factors, continuous research and development of autopilot is inevitable. It is of great significance to human security.
Related reading: Another tragedy, China's Tesla autopilot accident occurred
Glass fiber sleeving /Silicone fiberglass tube
Glass fiber sleeving, High Temperature Insulating Sleeve, Silicone fiberglass tube is made of braided e-grade glass fiber, then coated with silicone resin. Silicone Coated Braid Fiberglass Sleeving, Silicon Rubber Coated Fiber Glass Sleeving is widely used to protect electric cables,bundles and wires.Silicone Rubber Sleeving, Heat Resistant Fiberglass Sleeve, Silicone Fiberglass Sleeving, Fiberglass Silicon Rubber Insulation Sleeve it`s outstanding electrical insulation,self-extinguish,water & oil proof,resist to most acid and alkali,and anti-aging,which make this kind of sleevings to be a very good choice for electrical area using.
Product Characters:
|
Thermal Class |
H |
|
Size/inner diameter: |
0.5mm-50mm |
|
Colour: |
Black, red & blue, others on request
|
|
Working temperature: |
-60 degrees C -180 degrees C |
|
Rate voltage: |
1.5kv--7kv |
Structure:
Application:
Packing and Delivery:
|
PART NO. |
ID/mm |
WALL THICK/mm |
||
|
Nominal |
Tolerance |
Min. |
Max. |
|
|
GWIT-005 |
0.5 |
±0.15 |
0.35 |
±0.65 |
|
GWIT-008 |
0.8 |
±0.20 |
0.35 |
±0.65 |
|
GWIT-010 |
1.0 |
±0.20 |
0.35 |
±0.65 |
|
GWIT-015 |
1.5 |
±0.20 |
0.35 |
±0.65 |
|
GWIT-020 |
2 |
±0.20 |
0.35 |
±0.65 |
|
GWIT-025 |
2.5 |
±0.20 |
0.35 |
±0.65 |
|
GWIT-030 |
3 |
±0.25 |
0.35 |
±0.65 |
|
GWIT-035 |
3.5 |
±0.25 |
0.35 |
±0.65 |
|
GWIT-040 |
4 |
±0.25 |
0.35 |
±0.65 |
|
GWIT-045- |
4.5 |
±0.25 |
0.35 |
±0.65 |
|
GWIT-050 |
5 |
±0.25 |
0.35 |
±0.65 |
|
GWIT-060 |
6 |
±0.25 |
0.35 |
±0.65 |
|
GWIT-070 |
7 |
±0.25 |
0.35 |
±0.65 |
|
GWIT-080 |
8 |
±0.30 |
0.5 |
±0.90 |
|
GWIT-090 |
9 |
±0.30 |
0.5 |
±0.90 |
|
GWIT-100 |
10 |
±0.30 |
0.5 |
±0.90 |
|
GWIT-120 |
12 |
±0.30 |
0.5 |
±0.90 |
|
GWIT-140 |
14 |
±0.50 |
0.5 |
±1.00 |
|
GWIT-160 |
16 |
±0.50 |
0.5 |
±1.00 |
|
GWIT-180 |
18 |
±0.50 |
0.5 |
±1.00 |
|
GWIT-200 |
20 |
±0.50 |
0.5 |
±1.00 |
|
GWIT-220 |
22 |
±0.60 |
0.5 |
±1.50 |
|
GWIT-240 |
24 |
±0.60 |
0.5 |
±1.50 |
|
GWIT-250 |
25 |
±0.60 |
0.5 |
±1.50 |
|
GWIT-300 |
30 |
±0.60 |
1 |
±2.00 |
|
GWIT-350 |
35 |
±0.60 |
1 |
±2.00 |
|
GWIT-400 |
40 |
±0.60 |
1 |
±2.00 |
|
GWIT-450 |
45 |
±0.60 |
1 |
±2.00 |
|
GWIT-500 |
50 |
±0.60 |
1 |
±2.00 |
Glass fiber sleeving,Silicone fiberglass tube​,Silicone Fiberglass Sleeving,Silicon Rubber Coated Fiber Glass Sleeving, Heat Resistant Fiberglass Sleeve
KEYUACE Materials Co., Ltd. , https://www.insulationtubing.com