3 ms·

A Better Solution for Autonomous Cars

9 points by Rohanarun 11y ago

Rohanarun 11y ago
Would appreciate any feedback on how to improve the simulation for testing. For now turning off traffic lights makes the efficiency go up by 10 cars/minute. It would be interesting to test real city models, and different traffic algorithms
- aidenn0 11y ago
  I don't understand how this is hacking-resistant; if the car is relying on external information, then any flaw in its authentication of the external information can be exploited, regardless of any changes deployed by the city.
  - Rohanarun 11y ago
    If a car starts moving independent of a city's models, it would easy to identify as a possible hack in real time. If a car gets hacked independently, there may not be much evidence of how it happened. The idea is a city based system can send help faster. A completely independent system can be hacked in multiple different ways to trick the controls, such as through the model, but a city based system would also have minimal interfaces in the actual car.
    - aidenn0 11y ago
      That makes sense, but I wonder if the city could respond in time to prevent casualties, given the close proximity of autos and pedestrians in most cities.
      - Rohanarun 11y ago
        I agree any real car hack will likely not end well. This solution also introduces the possibility of a new vector for state sponsored attacks, which should be concerning to some. The utilitarian decision is still to upgrade everyone, because the people who would hack cars to kill would likely do it another way. This may become the new improved way to "cut brakes." A state mandated system could upgrade hacking charges to serious felonies, which deters most except state-sponsored and psychopaths. I have a feeling that real car hacks will be as rare as poison Halloween candy Also in a situation where someone is hacked, having a dense recorded model of all the roads in a city would be very handy for investigation, since the hacker would likely have to be near the car.
kstenerud 11y ago
So a centralized control model that becomes a single point of failure? No, thanks. Centralized, monolithic models fail. They're slow to upgrade, slow to adapt, don't fit the edge cases, and suffer from our inability to completely model a problem with 100% accuracy. And that's not even counting the up-front infrastructure costs that would NEVER make it past any political process. And the 15 year standardization process for the technology.
- Rohanarun 11y ago
  That's a good point, thanks I'll add it to the post under the top arguments. My answer would be that the current upgrade speed is to buy a new car, and how many people do you expect will do that in 15 years? This alternative approach brings up the lower bar, but still allows people to buy special cars like Teslas, if you want. So you can get the independent car if it's important to you, but you would also get the benefit of every other car on the road being safer. This solution is more geared towards optimizing in the numbers in car crashes
  - kstenerud 11y ago
    Yes, that's entirely the problem: Seeking an optimal solution when a less-than-optimal solution gives us "good enough" workability while avoiding all of the heavy costs and obsolescence risk of a centralized system. You're pushing for a technical solution that ignores political problems. The political effort to even get this to the planning stage is already astronomical. Then you get into the endless committees, finding a mayor with the cojones to confront his constituency with the up-front costs, somehow keeping everything compatible between each and every city you do this in (because each will be done by different contractors, each with their own ballooning expenses - 10x cost is not uncommon), and hoping to god the other cities don't get cold feet after watching the vanguard get eaten alive come election day for blowing through $5 billion or more (did I mention ballooning costs?). And then of course there's coordinating with the car companies... The fact is, autonomous cars as they're being done now will win by default, because besides the already huge political momentum they have, they also involve little-to-no cost for municipalities. Private companies pick up the R&D tab.

9 points by Rohanarun 11y ago

Would appreciate any feedback on how to improve the simulation for testing. For now turning off traffic lights makes the efficiency go up by 10 cars/minute. It would be interesting to test real city models, and different traffic algorithms

aidenn0 11y ago
I don't understand how this is hacking-resistant; if the car is relying on external information, then any flaw in its authentication of the external information can be exploited, regardless of any changes deployed by the city.
- Rohanarun 11y ago
  If a car starts moving independent of a city's models, it would easy to identify as a possible hack in real time. If a car gets hacked independently, there may not be much evidence of how it happened. The idea is a city based system can send help faster. A completely independent system can be hacked in multiple different ways to trick the controls, such as through the model, but a city based system would also have minimal interfaces in the actual car.
  - aidenn0 11y ago
    That makes sense, but I wonder if the city could respond in time to prevent casualties, given the close proximity of autos and pedestrians in most cities.
    - Rohanarun 11y ago
      I agree any real car hack will likely not end well. This solution also introduces the possibility of a new vector for state sponsored attacks, which should be concerning to some. The utilitarian decision is still to upgrade everyone, because the people who would hack cars to kill would likely do it another way. This may become the new improved way to "cut brakes." A state mandated system could upgrade hacking charges to serious felonies, which deters most except state-sponsored and psychopaths. I have a feeling that real car hacks will be as rare as poison Halloween candy Also in a situation where someone is hacked, having a dense recorded model of all the roads in a city would be very handy for investigation, since the hacker would likely have to be near the car.

kstenerud 11y ago

So a centralized control model that becomes a single point of failure? No, thanks. Centralized, monolithic models fail. They're slow to upgrade, slow to adapt, don't fit the edge cases, and suffer from our inability to completely model a problem with 100% accuracy. And that's not even counting the up-front infrastructure costs that would NEVER make it past any political process. And the 15 year standardization process for the technology.

Rohanarun 11y ago
That's a good point, thanks I'll add it to the post under the top arguments. My answer would be that the current upgrade speed is to buy a new car, and how many people do you expect will do that in 15 years? This alternative approach brings up the lower bar, but still allows people to buy special cars like Teslas, if you want. So you can get the independent car if it's important to you, but you would also get the benefit of every other car on the road being safer. This solution is more geared towards optimizing in the numbers in car crashes
- kstenerud 11y ago
  Yes, that's entirely the problem: Seeking an optimal solution when a less-than-optimal solution gives us "good enough" workability while avoiding all of the heavy costs and obsolescence risk of a centralized system. You're pushing for a technical solution that ignores political problems. The political effort to even get this to the planning stage is already astronomical. Then you get into the endless committees, finding a mayor with the cojones to confront his constituency with the up-front costs, somehow keeping everything compatible between each and every city you do this in (because each will be done by different contractors, each with their own ballooning expenses - 10x cost is not uncommon), and hoping to god the other cities don't get cold feet after watching the vanguard get eaten alive come election day for blowing through $5 billion or more (did I mention ballooning costs?). And then of course there's coordinating with the car companies... The fact is, autonomous cars as they're being done now will win by default, because besides the already huge political momentum they have, they also involve little-to-no cost for municipalities. Private companies pick up the R&D tab.