9 ms·

I don't know what to believe. A few days ago I spent some time checking publicly available data and the result is that the current state of art is extremely far

I don't know what to believe. A few days ago I spent some time checking publicly available data and the result is that the current state of art is extremely far from an autonomous car that, could like used to drive a taxi in a busy city like Rome. Not far as "still some effort needed" but potentially "not clear when this could be available" far. Now there are those claims made by multiple companies... Go figure.

gervase 9y ago
Based on my own observations, there's probably a big difference between operating as a taxi in Berlin and Rome, and another between operating in Rome and Mumbai. This only factors in other human drivers and infrastructure design/signage, not weather conditions (a different can of worms). I suspect that we'll see a progression like: 1) Well-regulated urban and highway environments 2) Well-regulated rural environments 3) Moderately regulated urban and rural environments 4) Poorly regulated rural environments 5) Poorly regulated urban environments And I think we'll probably be stuck on #1 for a while. I would also venture that most companies claiming to "be there" or "almost there" are referring to #1 as well. I also think that it's possible that politicians may try to legislate a transition up that list as that first case starts taking off (greater traffic enforcement, better signage/infrastructure, more rigorous driver training, etc).
- antirez 9y ago
  My feeling is that under special situations 1 degradates to 5, so to have good security anyway you have to solve problem 5.
  - mulmen 9y ago
    What are the special situations where 1 gives way to 5? The well-regulated part would imply to me that the possibility of slipping to another situation are eliminated or not allowed.
    - cmrdporcupine 9y ago
      Blizzard. Heavy rainfall or wind. Sudden crowd / riot / chaotic event. Pretty sure you can't legislate those out of existence. And sadly I think the fact that these vehicles are being developed and tested in Silicon Valley is either a cause or an symptom of their probably inability to deal with those kinds of conditions.
      - guelo 9y ago
        Also, accidents, road spills, random obstacles, construction, animals.
        mulmen 9y ago
        Road spills, random obstacles and animals all have to be dealt with for any of this to work at all. Construction can be considered in the regulations and self-driving cars can just be restricted from auto-drive mode in construction areas.
      - seanmcdirmid 9y ago
        Erm, china has more than a dozen well funded self driving efforts, some of them by very big companies (baidu). Those are being tested in places like Beijing, you never hear of them because you aren't looking outside of the USA.
      - mulmen 9y ago
        I wasn't suggesting legislating the conditions out of existence. I was thinking that kind of situation could be considered in the regulation. Much like VFR flight conditions a self-driving car would not operate in heavy rain/wind or low temperatures.
- alkonaut 9y ago
  Agree. And "will regulated" also probably includes a climate similar to what's found in e.g California.
1_2__4 9y ago
Maybe stop believing what companies say and start looking for third-party evidence that they're telling the truth
- revelation 9y ago
  That gets boring fast, if you go by the evidence the only player in the self-driving space is Google. Ironically they are the most restrained in public statements while there is a long list of pretenders announcing they'll ship in 2020 and already discussing far-fetched (entirely) philosophical scenarios, like the whole abortion of an discussion we had on "should the car hit pedestrians or save its driver". The whole conversation is unfortunately poisoned by endless wishful thinking instead of discussing hard evidence. I mean there is still a large proportion of people on here who think Tesla will somehow leapfrog Google because they have all those cars out there streaming the raw video data from their umpteen cameras back to Tesla HQ, utilizing an imaginary mobile communications standard that leapfrogs LTE by two orders of magnitude that they invented alongside their LIDAR-less L5 autonomous car software in record time.
  - cmrdporcupine 9y ago
    Thank you, this is exactly spot on. My personal skepticism on this subject has made me very unpopular at parties (well I don't actually go to parties, but you know) where lots of techy types go. Also earned me lots of downvotes here.
  - bobsil1 9y ago
    They're bulk uploads over home WiFi, and Tesla hired plenty of ML names like Andrej Karpathy. Having said that, AGI is the bottleneck for full self-driving.
cropsieboss 9y ago
The thing is, for car driving you need combinatorial search combined with a good model (statistical) of the world. Research I've seen does not incorporate any kind of meaningful search. Company that has all the roads and surroundings scanned and knows exactly where the car is at all times is at a huge advantage. Partially observed world gets your head hurting. Not modelling other cars practically makes everything useless. If you can't hallucinate everything around you as close as possible, the problem gets very very difficult. I'm pretty sure there are companies close to solving the problem, as they have more data and some also know practically every road in the modern world. Modelling and tracking other cars is a bigger issue, including pedestrians.
- dclowd9901 9y ago
  > Modelling and tracking other cars is a bigger issue Aren't there very reliable ways to model other cars? Something as big as a car can't move very unpredictably in a millisecond-to-millisecond basis. And at the end of the day, all you need to do is be able to model the behavior of your car. If car in front of me starts to swerve over on a collision course with my car, my car can note the probability of a collision a long (computer-long) time before it really needs to do anything about it. Knowing full-well its own braking, swerving and acceleration capabilities at any given speed, it should be able to know when it _must_ take action to avoid the actions of another, and it doesn't even have to rely on reaction time. No, I think the real problem here is for cars to do things like understand _context_ of a situation, like, for instance, how can it tell that there's a bus in front of me turning around that's going to be clear of my way in a moment vs. a tree that's been felled? That's something that computers are and will probably continue to be terrible at until they can closely mimic the capabilities of the human brain.
- bitL 9y ago
  > Modelling and tracking other cars is a bigger issue, including pedestrians. Unscented Kalman filters and particle filters are used for that - they are pretty good if you sample environment with >10Hz rate. Most of what you said has been taken car of pretty well, those aren't really hot issues anymore. Also, the thinking these days is that humans (even mediocre) can do a good job driving using eyes only and more robust camera solutions based on deep learning are here. Their problem is inference speed as HW is not fast enough to allow FHD or 4K input feed to be processed realtime, so imagery has to be downscaled significantly, leading to errors. Similarly, tracking other cars only depends on how big supercomputer can you fit into your trunk.
return0 9y ago
OTOH it is easy to teach human drivers to behave carefully and differently when they encounter a self driving car. They could even require special marking or color on the cars. People learn to be careful with trains, trams etc so i don't see why they won't learn to behave more rationally when a SD car is passing by.
- falcolas 9y ago
  Trains and trams are fairly easy to predict, though. They can only really go in two directions, and change speeds very gradually. And for all that we've "learned to be careful" around them, they still rack up 2-3 thousand crashes every year, even with extensive signing and signaling.
simonebrunozzi 9y ago
You and I are both Italian (good to see you, by the way) - and we both know that every self-driving car would totally fail in Rome :)

gervase 9y ago

Based on my own observations, there's probably a big difference between operating as a taxi in Berlin and Rome, and another between operating in Rome and Mumbai. This only factors in other human drivers and infrastructure design/signage, not weather conditions (a different can of worms). I suspect that we'll see a progression like: 1) Well-regulated urban and highway environments 2) Well-regulated rural environments 3) Moderately regulated urban and rural environments 4) Poorly regulated rural environments 5) Poorly regulated urban environments And I think we'll probably be stuck on #1 for a while. I would also venture that most companies claiming to "be there" or "almost there" are referring to #1 as well. I also think that it's possible that politicians may try to legislate a transition up that list as that first case starts taking off (greater traffic enforcement, better signage/infrastructure, more rigorous driver training, etc).

antirez 9y ago
My feeling is that under special situations 1 degradates to 5, so to have good security anyway you have to solve problem 5.
- mulmen 9y ago
  What are the special situations where 1 gives way to 5? The well-regulated part would imply to me that the possibility of slipping to another situation are eliminated or not allowed.
  - cmrdporcupine 9y ago
    Blizzard. Heavy rainfall or wind. Sudden crowd / riot / chaotic event. Pretty sure you can't legislate those out of existence. And sadly I think the fact that these vehicles are being developed and tested in Silicon Valley is either a cause or an symptom of their probably inability to deal with those kinds of conditions.
    - guelo 9y ago
      Also, accidents, road spills, random obstacles, construction, animals.
      - mulmen 9y ago
        Road spills, random obstacles and animals all have to be dealt with for any of this to work at all. Construction can be considered in the regulations and self-driving cars can just be restricted from auto-drive mode in construction areas.
    - seanmcdirmid 9y ago
      Erm, china has more than a dozen well funded self driving efforts, some of them by very big companies (baidu). Those are being tested in places like Beijing, you never hear of them because you aren't looking outside of the USA.
    - mulmen 9y ago
      I wasn't suggesting legislating the conditions out of existence. I was thinking that kind of situation could be considered in the regulation. Much like VFR flight conditions a self-driving car would not operate in heavy rain/wind or low temperatures.
alkonaut 9y ago
Agree. And "will regulated" also probably includes a climate similar to what's found in e.g California.

1_2__4 9y ago

Maybe stop believing what companies say and start looking for third-party evidence that they're telling the truth

revelation 9y ago
That gets boring fast, if you go by the evidence the only player in the self-driving space is Google. Ironically they are the most restrained in public statements while there is a long list of pretenders announcing they'll ship in 2020 and already discussing far-fetched (entirely) philosophical scenarios, like the whole abortion of an discussion we had on "should the car hit pedestrians or save its driver". The whole conversation is unfortunately poisoned by endless wishful thinking instead of discussing hard evidence. I mean there is still a large proportion of people on here who think Tesla will somehow leapfrog Google because they have all those cars out there streaming the raw video data from their umpteen cameras back to Tesla HQ, utilizing an imaginary mobile communications standard that leapfrogs LTE by two orders of magnitude that they invented alongside their LIDAR-less L5 autonomous car software in record time.
- cmrdporcupine 9y ago
  Thank you, this is exactly spot on. My personal skepticism on this subject has made me very unpopular at parties (well I don't actually go to parties, but you know) where lots of techy types go. Also earned me lots of downvotes here.
- bobsil1 9y ago
  They're bulk uploads over home WiFi, and Tesla hired plenty of ML names like Andrej Karpathy. Having said that, AGI is the bottleneck for full self-driving.

cropsieboss 9y ago

The thing is, for car driving you need combinatorial search combined with a good model (statistical) of the world. Research I've seen does not incorporate any kind of meaningful search. Company that has all the roads and surroundings scanned and knows exactly where the car is at all times is at a huge advantage. Partially observed world gets your head hurting. Not modelling other cars practically makes everything useless. If you can't hallucinate everything around you as close as possible, the problem gets very very difficult. I'm pretty sure there are companies close to solving the problem, as they have more data and some also know practically every road in the modern world. Modelling and tracking other cars is a bigger issue, including pedestrians.

dclowd9901 9y ago
> Modelling and tracking other cars is a bigger issue Aren't there very reliable ways to model other cars? Something as big as a car can't move very unpredictably in a millisecond-to-millisecond basis. And at the end of the day, all you need to do is be able to model the behavior of your car. If car in front of me starts to swerve over on a collision course with my car, my car can note the probability of a collision a long (computer-long) time before it really needs to do anything about it. Knowing full-well its own braking, swerving and acceleration capabilities at any given speed, it should be able to know when it _must_ take action to avoid the actions of another, and it doesn't even have to rely on reaction time. No, I think the real problem here is for cars to do things like understand _context_ of a situation, like, for instance, how can it tell that there's a bus in front of me turning around that's going to be clear of my way in a moment vs. a tree that's been felled? That's something that computers are and will probably continue to be terrible at until they can closely mimic the capabilities of the human brain.
bitL 9y ago
> Modelling and tracking other cars is a bigger issue, including pedestrians. Unscented Kalman filters and particle filters are used for that - they are pretty good if you sample environment with >10Hz rate. Most of what you said has been taken car of pretty well, those aren't really hot issues anymore. Also, the thinking these days is that humans (even mediocre) can do a good job driving using eyes only and more robust camera solutions based on deep learning are here. Their problem is inference speed as HW is not fast enough to allow FHD or 4K input feed to be processed realtime, so imagery has to be downscaled significantly, leading to errors. Similarly, tracking other cars only depends on how big supercomputer can you fit into your trunk.

return0 9y ago

OTOH it is easy to teach human drivers to behave carefully and differently when they encounter a self driving car. They could even require special marking or color on the cars. People learn to be careful with trains, trams etc so i don't see why they won't learn to behave more rationally when a SD car is passing by.

falcolas 9y ago
Trains and trams are fairly easy to predict, though. They can only really go in two directions, and change speeds very gradually. And for all that we've "learned to be careful" around them, they still rack up 2-3 thousand crashes every year, even with extensive signing and signaling.

simonebrunozzi 9y ago

You and I are both Italian (good to see you, by the way) - and we both know that every self-driving car would totally fail in Rome :)