Autopilot vs real world: what happens to infrastructure, morals and economy
In the previous series: the unsolvable problems of autopilot and bad sensors of collision avoidance systems .
Over the past month, little has changed. Tesla still get into accidents in autopilot mode, this time the first autopilot crash happened in China .
From the comments to the previous text, Elon Musk did not recognize that he needed to finish the algorithms. Therefore, I decided to add more cameras and radar to improve autopilot. Write to him someone.
')
This is what happened:
It is difficult not to notice the analogies with the May accident:
Guess the cause of accidents is easy. It is lack of information, namely:
Tesla did not understand.
While autopilots - a collection of crutches, with which help solve private driving situations. So, they operate a road strip as an inseparable whole. The lane is occupied or completely free; the obstacle cannot occupy a part of the lane to the left or to the right. The obstacle cannot occupy a part of the strip from above (see previous accident with a trailer). One of the many assumptions that narrow the piloting task and make it somehow solvable.
We will use another idea that hovers in our minds, we will make a road on which only autopilots go. Turn on their ITS (intelligent transport system). Let's go over the possible scenario, it seems it will work.
If the points?
1. The car can understand that its malfunction prevents movement. With the exception of the accident (when airbags are triggered), the person confirms the call to the operator. This is how ERA-Glonass or eCall is already working.
All is well, put a tick.
2. The car transmits its position and the fact of failure. Report what part of the road he took, while it is impossible. Navigation is not enough, and the car is not a material point, it can stand across the road or at an angle. Accurate information can be understood by the cameras or radars of the road infrastructure, but in this case they should be closed the entire road.
We do not tick the box, it’s still very far to complete coverage of even urban roads.
3. The signal from the car is received, the place of breakdown is plotted on the road grid. ITS sent information to other cars. The protocol of information exchange is, to issue information even in twenty different formats for different cars is not a problem.
If not yet decided, it will be. Check mark.
4. To receive information that there is no obstacle in a certain coordinate, but what to do to the others?
4a. Every car must be equipped with navigation. Just to match your route and problem on the way.
Not decided, navigation is equipped with less than 20% of all cars.
4b. The car does not understand which lane is moving; navigation accuracy is not enough. It is necessary to resort to indirect measurements (if on the left there are cars going towards you, you are in the extreme left lane), a reconciliation of beacons (they can stand on the separator) or a comparison of these objects with an ultra-precise map . Most likely we should expect a combination of all three options.
That doesn't work either.
4c. We rebuild into free lanes or get up when it is impossible to continue moving. The autopilot must understand whether he will be able to drive in free space or not. The infrastructure does not have the necessary information of the required accuracy.
Does not work!
As a result, the practical implementation of such an elementary scenario as a detour of a broken car leads to the need to completely hang the whole way with radars or cameras, additional navigation marks, not to mention the markup and all the rest.
And there is no problem. Just such a smart road autopilot needless!
Why is everyone investing in self-driving cars? Because autopilot does not need to build infrastructure. Everything is already built and it is public roads.
From a technology point of view, it is 100 times easier to automate trains, subways and trams, which have one degree of freedom within the rail. Works are underway, but even from the point of view of the use of collision avoidance systems, cars are ten years ahead of rail transport.
Dedicated roads are roads (the accents should be arranged according to taste) or simply impossible. Imagine a dedicated rail network to each entrance. Three-dimensional space is not enough.
Therefore, no one invests in the autopilot-only highway or dedicated lanes. These are useless projects from the point of view of economics. Speed ​​control, car counting, communication with each other? Another thing.
Dedicated lanes for autopilots do not wait, all the autopilot relish in self-sufficiency.
The idea to give the car more information about the road is good and technically feasible. Speaking of the V2X interface family, they usually mean V2I (machine-infrastructure) and V2V (machine-machine). There are other cases, such as V2B (car-bike) or V2P (car-pedestrian).
At first glance, everything is fine with the V2X. With the low level protocol agreed. Several private applications, such as fare, are also agreed. Anyone using transponders on the M4 and other toll routes use V2I.
Every automaker writes high-level solutions on its own and does not share it with others. Hence the incompatible standards that hinder the development of V2X. Around fools who do not understand the benefits of open standards? This is hardly the case. All use the same filling pistols, and tire sizes are also universal.
I suppose that automakers are afraid of “commoditization” of transport, that is, turning cars into nameless trucks that are chosen without regard to the brand. Then you have to compete only with the price that quickly burns out the market. The consumer will not be happy for long, the fall in profits should be followed by a sharp drop in development budgets. Machines will get worse.
How does it look in reality? The most famous competition is Euro Truck Platooning . It involves incompatible trucks that are already able to drive on public roads with an escort, with one driver for 4-5 trucks. For those who are interested in the topic, I recommend reading the latest report from the tests of 2016 . There are a lot of textures, interviews and everything that we love so much. For the rest of the summary: there are still too many problems, but they can be solved, especially if we recall that platooning is going to be launched in ... 2025. This is the question of the horizons of autopilot.
Technically, in order to give data to other machines, V2X are ready now . Transmit information about traffic lights, signs, speed limits - too. Like this:
V2X is already in use and will be used later. As an extension of information from the cards - already, as a way to exchange real-time information about cars - in the future.
The problem does not look large-scale, but it is impossible not to denote it. Using incorrect data will quickly lead to a crash.
Fake data from the infrastructure has already been discussed in the comments , it is possible. With better sensors of your own, here’s a fresh experience with Tesla’s autopilot disruption . The camera is burned out by diodes, but it copes. Radars are jammed with expensive special equipment, but only in statics. And the ultrasound is just crap :).
So, the car will be based on its own radar and cameras, which partly duplicate each other, take into account information from the V2X and make a decision. At the same time giving the impersonal data back to ITS so that it compares the information from the cars with its own. Good feedback channel.
What will be used to improve the accuracy and detail of the information?
Autopilot will always have its own sensors and a sophisticated decision-making system, taking into account the reliability of various sources.
Suppose technically we are already ready to auto-pilot plus or minus. And morally? I think not: the actions of autopilots began to be discussed publicly only a year or two ago.
There are two main topics for discussion and the loss of places by truckers is not included.
First, let's understand that autopilots will kill people in no indirect way, but directly in accidents. What kind of people? The KAMAZ questionnaire showed that there would be those who wanted to sentence a person in order to save the dog's life. Almost 3/4 of the respondents believe that the autopilot must kill its passenger, saving a group of people. Right? Exactly while when the respondent does not sit in the drone. At the cost of their lives, less than a quarter are ready to save others.
The flesh itself begins next. The remarkable MIT moral machine (pass the survey, this is not for long) shows that we are ready to delegate to the autopilot the choice of the victim, taking into account its value to society. Simply put, it is better to knock down an old criminal than a young prodigy. According to the link above, the detailed elaboration is more detailed, but taking into account some racial peculiarities of criminality, autopilot must kill Black five times more often than White. I would not want to explain it from the stage at the presentation of the new BMW model.
Secondly, the autopilot will still reduce the total loss of humanity from accidents. The number of victims will fall by 30-90 percent (yes, estimates vary). In any case, this is a lot, but the number of victims of accidents in developed countries is already halving in a decade, to which the autopilot has nothing to do.
Attention, question. How many accidents do we consider acceptable for the autopilot? There are no studies on this score, but the approach is likely to continue. “On average” people will be satisfied with even a 10 percent reduction in the number of victims, but only if the autopilot does not kill their relative. As soon as the account of autopilot victims goes up to hundreds and thousands (a matter of time), we are awaited by the discussion of a non-acidic degree.
Once touched the question of responsibility. Globally, the definition of a robo-personality that Europeans even propose to discuss will be able to respond to it. Only not right now, but with a horizon of 50 years . Autopilots did not go much further. The American NHTSA, as it were, allowed unmanned driving, but the same California did not refuse the desire to see driving the drone .
By the way, as an idea: the issue of responsibility can be translated into the money plane. By reducing the total number of accidents, insurance affiliates with automakers will be able to pay enormous compensation to the relatives of the victims, which will reduce the level of negativity.
Before we fully understand the moral and ethical implications of launching autopilot, we are still very far, well at least started to discuss.
The last topic is the most important one. All the same, the whole business is about money.
Autopilot has a positive economy, visible on a global scale. Large companies and funds are free to invest billions in R & D, because all this will pay off. Morgan Stanley believes that with 100 percent distribution, road autopilots will save the world about 5.6 trillion dollars annually. This is 5% of world GDP by the time, never a toy.
I will voice some of the economic consequences of autopilot.
Fuel companies will suffer. Within the United States, the savings will amount to $ 160 billion. Along with the spread of electric vehicles, this is an important factor in the pressure of oil prices and various energy superpowers. Yes, the sheikhs read this report of the Morgans.
Shipping will be cheaper and easier. In the cost of transportation, three quarters is the labor of the driver. Intermodal transportation will be simplified, hence attention to start-ups in this area. In the whole world, the cost of transportation will decrease approximately twice, and their profitability will increase. At the cost of releasing 90 percent of drivers, sorry. Some yogurt will be cheaper by a percentage. And in the price of sand or concrete the cost of delivery is more than 50%, that is, in this way all construction in the world will become cheaper.
There will be fewer accidents and medical bills. Savings are estimated at 180-190 billion dollars in the US alone. By the way, the GDP will fall precisely on this amount, and service companies and medical institutions will not receive the same money.
Automakers - in the current business model are more likely to lose, since cars need less. But they will provide more services. From here attention to any auto-startups, auto giants will make room. So that everyone understands, the car market is about $ 2 trillion annually, which still continues to grow.
A person will be released about 50 minutes of time per day. Most likely this will be used by the manufacturers of info-equipment and alcohol. In the press, of course, everyone writes about increasing productivity and over an hour a day, which can be spent on work.
The release of territories from roads and parking lots - and this land is mainly in cities. 5.7 billion square meters, it is difficult to count in money. Real estate should be cheaper.
The important problem of the last mile in the delivery of goods is solved. It is important and will affect an insane number of companies . For example:
We can wait for the distribution of new services, which we don’t even think about now. The simplest is the cheapest (around $ 1) delivery of croissants in the morning to thousands of homes.
Consensus on the future of automakers and insurance companies NO. It seems to be a good report Munich RE (one of the major reinsurance companies, 130 years, 280 billion euros of assets) which says a lot, but nothing is said. So I respect the experts in advance, who will write out in three lines what will happen to the market.
Over the past month, little has changed. Tesla still get into accidents in autopilot mode, this time the first autopilot crash happened in China .
From the comments to the previous text, Elon Musk did not recognize that he needed to finish the algorithms. Therefore, I decided to add more cameras and radar to improve autopilot. Write to him someone.
')
This is what happened:
It is difficult not to notice the analogies with the May accident:
Guess the cause of accidents is easy. It is lack of information, namely:
- The angle of the radar in the center of the car is too narrow to see cars at the edges of the strip.
- With the help of cameras it is impossible to understand the dimensions of a static obstacle and the dynamic corridor of a car.
Tesla did not understand.
While autopilots - a collection of crutches, with which help solve private driving situations. So, they operate a road strip as an inseparable whole. The lane is occupied or completely free; the obstacle cannot occupy a part of the lane to the left or to the right. The obstacle cannot occupy a part of the strip from above (see previous accident with a trailer). One of the many assumptions that narrow the piloting task and make it somehow solvable.
We will use another idea that hovers in our minds, we will make a road on which only autopilots go. Turn on their ITS (intelligent transport system). Let's go over the possible scenario, it seems it will work.
- The car understands that broke.
- The car informs the intellectual transport system that it has got in the left lane. Technical assistance has already left.
- ITS puts the mark of the accident in this place and recommends that those who are moving behind reorganize to the right.
- Cars took the information, rebuilt or went on a different route
If the points?
1. The car can understand that its malfunction prevents movement. With the exception of the accident (when airbags are triggered), the person confirms the call to the operator. This is how ERA-Glonass or eCall is already working.
All is well, put a tick.
2. The car transmits its position and the fact of failure. Report what part of the road he took, while it is impossible. Navigation is not enough, and the car is not a material point, it can stand across the road or at an angle. Accurate information can be understood by the cameras or radars of the road infrastructure, but in this case they should be closed the entire road.
We do not tick the box, it’s still very far to complete coverage of even urban roads.
3. The signal from the car is received, the place of breakdown is plotted on the road grid. ITS sent information to other cars. The protocol of information exchange is, to issue information even in twenty different formats for different cars is not a problem.
If not yet decided, it will be. Check mark.
4. To receive information that there is no obstacle in a certain coordinate, but what to do to the others?
4a. Every car must be equipped with navigation. Just to match your route and problem on the way.
Not decided, navigation is equipped with less than 20% of all cars.
4b. The car does not understand which lane is moving; navigation accuracy is not enough. It is necessary to resort to indirect measurements (if on the left there are cars going towards you, you are in the extreme left lane), a reconciliation of beacons (they can stand on the separator) or a comparison of these objects with an ultra-precise map . Most likely we should expect a combination of all three options.
That doesn't work either.
4c. We rebuild into free lanes or get up when it is impossible to continue moving. The autopilot must understand whether he will be able to drive in free space or not. The infrastructure does not have the necessary information of the required accuracy.
Does not work!
As a result, the practical implementation of such an elementary scenario as a detour of a broken car leads to the need to completely hang the whole way with radars or cameras, additional navigation marks, not to mention the markup and all the rest.
And there is no problem. Just such a smart road autopilot needless!
ROAD INFRASTRUCTURE
Why is everyone investing in self-driving cars? Because autopilot does not need to build infrastructure. Everything is already built and it is public roads.
From a technology point of view, it is 100 times easier to automate trains, subways and trams, which have one degree of freedom within the rail. Works are underway, but even from the point of view of the use of collision avoidance systems, cars are ten years ahead of rail transport.
Dedicated roads are roads (the accents should be arranged according to taste) or simply impossible. Imagine a dedicated rail network to each entrance. Three-dimensional space is not enough.
Therefore, no one invests in the autopilot-only highway or dedicated lanes. These are useless projects from the point of view of economics. Speed ​​control, car counting, communication with each other? Another thing.
Conclusion:
Dedicated lanes for autopilots do not wait, all the autopilot relish in self-sufficiency.
V2x
The idea to give the car more information about the road is good and technically feasible. Speaking of the V2X interface family, they usually mean V2I (machine-infrastructure) and V2V (machine-machine). There are other cases, such as V2B (car-bike) or V2P (car-pedestrian).
At first glance, everything is fine with the V2X. With the low level protocol agreed. Several private applications, such as fare, are also agreed. Anyone using transponders on the M4 and other toll routes use V2I.
Every automaker writes high-level solutions on its own and does not share it with others. Hence the incompatible standards that hinder the development of V2X. Around fools who do not understand the benefits of open standards? This is hardly the case. All use the same filling pistols, and tire sizes are also universal.
I suppose that automakers are afraid of “commoditization” of transport, that is, turning cars into nameless trucks that are chosen without regard to the brand. Then you have to compete only with the price that quickly burns out the market. The consumer will not be happy for long, the fall in profits should be followed by a sharp drop in development budgets. Machines will get worse.
How does it look in reality? The most famous competition is Euro Truck Platooning . It involves incompatible trucks that are already able to drive on public roads with an escort, with one driver for 4-5 trucks. For those who are interested in the topic, I recommend reading the latest report from the tests of 2016 . There are a lot of textures, interviews and everything that we love so much. For the rest of the summary: there are still too many problems, but they can be solved, especially if we recall that platooning is going to be launched in ... 2025. This is the question of the horizons of autopilot.
Technically, in order to give data to other machines, V2X are ready now . Transmit information about traffic lights, signs, speed limits - too. Like this:
Conclusion:
V2X is already in use and will be used later. As an extension of information from the cards - already, as a way to exchange real-time information about cars - in the future.
SECURITY OF DATA
The problem does not look large-scale, but it is impossible not to denote it. Using incorrect data will quickly lead to a crash.
Fake data from the infrastructure has already been discussed in the comments , it is possible. With better sensors of your own, here’s a fresh experience with Tesla’s autopilot disruption . The camera is burned out by diodes, but it copes. Radars are jammed with expensive special equipment, but only in statics. And the ultrasound is just crap :).
So, the car will be based on its own radar and cameras, which partly duplicate each other, take into account information from the V2X and make a decision. At the same time giving the impersonal data back to ITS so that it compares the information from the cars with its own. Good feedback channel.
What will be used to improve the accuracy and detail of the information?
- Data from the navigation of cars.
- Inertial sensors in the car and beacons on the road (one fruit company has patented them recently).
- Data from cameras and road radar.
Conclusion:
Autopilot will always have its own sensors and a sophisticated decision-making system, taking into account the reliability of various sources.
MORAL AND LEGAL ASPECTS
Suppose technically we are already ready to auto-pilot plus or minus. And morally? I think not: the actions of autopilots began to be discussed publicly only a year or two ago.
There are two main topics for discussion and the loss of places by truckers is not included.
First, let's understand that autopilots will kill people in no indirect way, but directly in accidents. What kind of people? The KAMAZ questionnaire showed that there would be those who wanted to sentence a person in order to save the dog's life. Almost 3/4 of the respondents believe that the autopilot must kill its passenger, saving a group of people. Right? Exactly while when the respondent does not sit in the drone. At the cost of their lives, less than a quarter are ready to save others.
The flesh itself begins next. The remarkable MIT moral machine (pass the survey, this is not for long) shows that we are ready to delegate to the autopilot the choice of the victim, taking into account its value to society. Simply put, it is better to knock down an old criminal than a young prodigy. According to the link above, the detailed elaboration is more detailed, but taking into account some racial peculiarities of criminality, autopilot must kill Black five times more often than White. I would not want to explain it from the stage at the presentation of the new BMW model.
Secondly, the autopilot will still reduce the total loss of humanity from accidents. The number of victims will fall by 30-90 percent (yes, estimates vary). In any case, this is a lot, but the number of victims of accidents in developed countries is already halving in a decade, to which the autopilot has nothing to do.
Attention, question. How many accidents do we consider acceptable for the autopilot? There are no studies on this score, but the approach is likely to continue. “On average” people will be satisfied with even a 10 percent reduction in the number of victims, but only if the autopilot does not kill their relative. As soon as the account of autopilot victims goes up to hundreds and thousands (a matter of time), we are awaited by the discussion of a non-acidic degree.
Once touched the question of responsibility. Globally, the definition of a robo-personality that Europeans even propose to discuss will be able to respond to it. Only not right now, but with a horizon of 50 years . Autopilots did not go much further. The American NHTSA, as it were, allowed unmanned driving, but the same California did not refuse the desire to see driving the drone .
By the way, as an idea: the issue of responsibility can be translated into the money plane. By reducing the total number of accidents, insurance affiliates with automakers will be able to pay enormous compensation to the relatives of the victims, which will reduce the level of negativity.
Conclusion:
Before we fully understand the moral and ethical implications of launching autopilot, we are still very far, well at least started to discuss.
ECONOMY
The last topic is the most important one. All the same, the whole business is about money.
Autopilot has a positive economy, visible on a global scale. Large companies and funds are free to invest billions in R & D, because all this will pay off. Morgan Stanley believes that with 100 percent distribution, road autopilots will save the world about 5.6 trillion dollars annually. This is 5% of world GDP by the time, never a toy.
I will voice some of the economic consequences of autopilot.
What everyone agrees on:
Fuel companies will suffer. Within the United States, the savings will amount to $ 160 billion. Along with the spread of electric vehicles, this is an important factor in the pressure of oil prices and various energy superpowers. Yes, the sheikhs read this report of the Morgans.
Shipping will be cheaper and easier. In the cost of transportation, three quarters is the labor of the driver. Intermodal transportation will be simplified, hence attention to start-ups in this area. In the whole world, the cost of transportation will decrease approximately twice, and their profitability will increase. At the cost of releasing 90 percent of drivers, sorry. Some yogurt will be cheaper by a percentage. And in the price of sand or concrete the cost of delivery is more than 50%, that is, in this way all construction in the world will become cheaper.
There will be fewer accidents and medical bills. Savings are estimated at 180-190 billion dollars in the US alone. By the way, the GDP will fall precisely on this amount, and service companies and medical institutions will not receive the same money.
Implicit moments:
Automakers - in the current business model are more likely to lose, since cars need less. But they will provide more services. From here attention to any auto-startups, auto giants will make room. So that everyone understands, the car market is about $ 2 trillion annually, which still continues to grow.
A person will be released about 50 minutes of time per day. Most likely this will be used by the manufacturers of info-equipment and alcohol. In the press, of course, everyone writes about increasing productivity and over an hour a day, which can be spent on work.
The release of territories from roads and parking lots - and this land is mainly in cities. 5.7 billion square meters, it is difficult to count in money. Real estate should be cheaper.
The important problem of the last mile in the delivery of goods is solved. It is important and will affect an insane number of companies . For example:
“Business model: 25-30% food commission and 2.5 euro for delivery from the client. It sounds profitable. But actually, no, because couriers need to pay at the rate of 15 euro / hour, which, with the current average check, should be more than 1.5 deliveries per hour.
Previous money from two rounds of investment with a total volume of 16 million euros ended, and the required “density” of deliveries did not accumulate. The third round of investment was promised, but after Dyudila, they recalled their offer back.
Two conclusions:
1. Logistics is able to devour all the money. “
We can wait for the distribution of new services, which we don’t even think about now. The simplest is the cheapest (around $ 1) delivery of croissants in the morning to thousands of homes.
Instead of output:
Consensus on the future of automakers and insurance companies NO. It seems to be a good report Munich RE (one of the major reinsurance companies, 130 years, 280 billion euros of assets) which says a lot, but nothing is said. So I respect the experts in advance, who will write out in three lines what will happen to the market.
And finally, we summarize.
- People are just beginning to think about how autopilots fit into the existing financial, legal and many other aspects of life. Nobody is in any hurry, we have 15-20 years.
- It is not in vain that an autopilot vehicle comes out so complicated - it is a programmed “Gordian knot” into which all the problems of the related markets have merged. For example, infrastructure.
- The effects of using autopilots are difficult to understand to the end, so much and differently affect autopilot on dozens of spheres. If my opinion is interesting - like other robotization, it brings humanity closer to unconditional income.
- Yes, and finally forget about the selected roads.
Source: https://habr.com/ru/post/396827/
All Articles