3 ms·

While we may get to the point where sensors and a form of AI can reliably drive a car, we're so far from it that we don't even know if we're heading towards the

While we may get to the point where sensors and a form of AI can reliably drive a car, we're so far from it that we don't even know if we're heading towards the goal post. What we need, if we want to solve the partial self driving problem right now, is not fully autonomous systems. We could start by agreeing on a way for cars to sense the road - think in the way robotic lawn mowers work where you have to bury cables. Then implement that in highways, ring roads and other long stretches of road. Then we can agree on a way for cars to communicate and exchange relevant information. Then we can equip cars with hardware like the Tesla have, and then use it to collect data and compare the "auto pilot" with the known working system of following a signal in the road. We still need LIDAR to reliably stop for other cars, but that is a mostly solved problem today.

hnlmorg 7y ago
The problem we have right now is that: 1. laying cables wouldn't work because there are a great many historic towns and cities (eg Amsterdam) and digging up cobble streets to lay cables for self-driving vehicles simple isn't going happen when those cities have so many bigger problems for autonomous cars such as trams, cycle paths, pedestrian zones and tourists what will just walk out in front of everyone in an unpredictable haze. 2. you have districts that are so under funded that the roads aren't even maintained in their current state. 3. So lets say you scape redeveloping the roads and have radio sensors on lap posts - then you have the problem that many rural areas in even richer countries like the UK don't even have street lamps. And what about countries like Australia and America that can have long stretches for tens or even hundreds of miles without a single house, let alone street lights to put sensors on. I agree with you that normally this kind of problem requires a technical approach but the problem we have is that redeveloping our roads would be such a monumental overhaul that it would actually be a bigger job than developing the AI to handle our existing infrastructure. Plus redeveloping our roads only solves part of the problem - you still have the issue of other cars on the road, pedestrians, cyclists, unexpected hazards like fallen trees, etc. You could fix other vehicles on the road by allowing all vehicles to communicated with each other - like a mesh network - but then you have the problem of doing that securely (so you couldn't just "virtually ram" another car off the road by telling it there's an obstruction that doesn't exist) and you again have the problem that it would be more expensive and take longer to retrofit every single vehicle with that tech than it would be to write AI. And you still wouldn't have addressed the problem of kids, animals and fallen trees. AI might seem like a naff workaround and maybe if / when we colonise a new planet then we might be able to build roads with sensors inside tubes that restrict access (I guess a little like an underground rail network) but for the time being AI is our best bet for autonomous vehicles.
- hvidgaard 7y ago
  I specifically said that it should be done on high volume, maintained roads, such as highways and ring roads. It doesn't have to happen overnight, but it could happen gradually. It would serve other purposes too, like guidance for snow plows. It's a 80/20 solution - cover 80% of the travel for 20% of the cost. Solving the problem on highways and high volume roads without cyclists are significantly easier, and would still give a tremendous benefit. Regarding laying the cables in existing roads. We're already pretty good at "shooting" cables under existing roads with pneumatic solutions, and I'm confident that we could do so for rural roads at a modest price. But that's a different talk. If we want to develop an "AI" for this, it's not going to be ready in 10 year, nor 20 - we do not have the processing power, or understanding of the domain to do it at this point, and we're moving rather slowly. Some of the more practical problems such as secure communication are 100% solved, and reliable car, pedestrian, cycle detecting are problems that are, fairly well understood at this point, and we certainly can make solutions that rival even alert and good drivers. Things like this move slowly - it's not going to be retrofitted, but agreeing on a standard and new vehicles can implement in gradually.
  - hnlmorg 7y ago
    > I specifically said that it should be done on high volume, maintained roads, such as highways and ring roads. It doesn't have to happen overnight, but it could happen gradually. It would serve other purposes too, like guidance for snow plows. It's a 80/20 solution - cover 80% of the travel for 20% of the cost. Solving the problem on highways and high volume roads without cyclists are significantly easier, and would still give a tremendous benefit. I've travelled by road a fair bit around different countries and I assure you that a lot of high volume maintained roads still fall into the categories I outlined. Moreover, a lot of high volume roads aren't maintained. I honestly don't think you would even get remotely close to covering 80% of the roads for 20% of the cost. Maybe if you isolate specific districts like Silicon Valley then your figure might be accurate. But for "normal" towns in Europe that wouldn't be the case and, as has been discussed elsewhere on HN, having some roads updated and others not would actually be more dangerous than having none redeveloped because humans would demonstrate the same complacency across all the roads once they're used to autonomy on the new roads. Also highways aren't the problem - they're significantly easier to determine in code and see fewer deaths per year. It's the smaller streets in towns that are the problem. Both from the perspective of statistical deaths and also from the perspective of the complexity of AI required to navigate them. They're what really need guide wires the most. However going back to highways: even on motorways you still have the problem of motorcyclists weaving in and out of traffic. So I'd argue that you haven't entirely eliminated the "cyclists" problem - you just now have faster and even deadlier ones. I should say that I do agree with you that there would be tremendous benefit if it were feasible. What I disagree with is just how feasible it is. > Regarding laying the cables in existing roads. We're already pretty good at "shooting" cables under existing roads with pneumatic solutions, and I'm confident that we could do so for rural roads at a modest price. But that's a different talk. Shooting the cables is the easy if you have pavements with cables already installed. If you don't then you need to dig up the road to lay them. And that's still only half the story. There's a whole load of costs involved in the planning stage; scheduling the work around busy periods (which sometimes means paying for night shift work) and organising the road closures. And it's the maintenance of that infrastructure after the work has happened. Remember that a lot of busy streets aren't even maintained currently. And it's doing that work sympathetically around historical locations (where applicable). There's a significant amount of more cost and work involved than I feel you're giving credit for. > If we want to develop an "AI" for this, it's not going to be ready in 10 year, nor 20 - we do not have the processing power, or understanding of the domain to do it at this point, and we're moving rather slowly. I completely agree that we're decades from a solution but I disagree that we're moving rather slowly. 10 years ago the stuff that are in consumer vehicles - even some entry level cars - would have been considered Sci-Fi just 10 years ago. But even if we are 50 years away, I can't see cables laid in twice that time. I mean half the towns in England and America don't even have broadband, let alone fibre, and you have consumers begging to pay for it. Who's going to foot the bill for redeveloping all our roads? Musk certainly wont. > Things like this move slowly - it's not going to be retrofitted, but agreeing on a standard and new vehicles can implement in gradually. The problem is you need it implemented on every vehicle for it to work. It's like anti-vaxers argument that "I don't need to get vaccinated when everyone else is" but it actually doesn't need to take a large percentage of unvaxinated people for an outbreak to still occur. Equally you only need one vehicle without a mesh network installed to case an accident that could kill several people - including those in autonomous vehicles. If you don't have total coverage then the mesh network isn't going to be reliable.

hnlmorg 7y ago

The problem we have right now is that: 1. laying cables wouldn't work because there are a great many historic towns and cities (eg Amsterdam) and digging up cobble streets to lay cables for self-driving vehicles simple isn't going happen when those cities have so many bigger problems for autonomous cars such as trams, cycle paths, pedestrian zones and tourists what will just walk out in front of everyone in an unpredictable haze. 2. you have districts that are so under funded that the roads aren't even maintained in their current state. 3. So lets say you scape redeveloping the roads and have radio sensors on lap posts - then you have the problem that many rural areas in even richer countries like the UK don't even have street lamps. And what about countries like Australia and America that can have long stretches for tens or even hundreds of miles without a single house, let alone street lights to put sensors on. I agree with you that normally this kind of problem requires a technical approach but the problem we have is that redeveloping our roads would be such a monumental overhaul that it would actually be a bigger job than developing the AI to handle our existing infrastructure. Plus redeveloping our roads only solves part of the problem - you still have the issue of other cars on the road, pedestrians, cyclists, unexpected hazards like fallen trees, etc. You could fix other vehicles on the road by allowing all vehicles to communicated with each other - like a mesh network - but then you have the problem of doing that securely (so you couldn't just "virtually ram" another car off the road by telling it there's an obstruction that doesn't exist) and you again have the problem that it would be more expensive and take longer to retrofit every single vehicle with that tech than it would be to write AI. And you still wouldn't have addressed the problem of kids, animals and fallen trees. AI might seem like a naff workaround and maybe if / when we colonise a new planet then we might be able to build roads with sensors inside tubes that restrict access (I guess a little like an underground rail network) but for the time being AI is our best bet for autonomous vehicles.

hvidgaard 7y ago
I specifically said that it should be done on high volume, maintained roads, such as highways and ring roads. It doesn't have to happen overnight, but it could happen gradually. It would serve other purposes too, like guidance for snow plows. It's a 80/20 solution - cover 80% of the travel for 20% of the cost. Solving the problem on highways and high volume roads without cyclists are significantly easier, and would still give a tremendous benefit. Regarding laying the cables in existing roads. We're already pretty good at "shooting" cables under existing roads with pneumatic solutions, and I'm confident that we could do so for rural roads at a modest price. But that's a different talk. If we want to develop an "AI" for this, it's not going to be ready in 10 year, nor 20 - we do not have the processing power, or understanding of the domain to do it at this point, and we're moving rather slowly. Some of the more practical problems such as secure communication are 100% solved, and reliable car, pedestrian, cycle detecting are problems that are, fairly well understood at this point, and we certainly can make solutions that rival even alert and good drivers. Things like this move slowly - it's not going to be retrofitted, but agreeing on a standard and new vehicles can implement in gradually.
- hnlmorg 7y ago
  > I specifically said that it should be done on high volume, maintained roads, such as highways and ring roads. It doesn't have to happen overnight, but it could happen gradually. It would serve other purposes too, like guidance for snow plows. It's a 80/20 solution - cover 80% of the travel for 20% of the cost. Solving the problem on highways and high volume roads without cyclists are significantly easier, and would still give a tremendous benefit. I've travelled by road a fair bit around different countries and I assure you that a lot of high volume maintained roads still fall into the categories I outlined. Moreover, a lot of high volume roads aren't maintained. I honestly don't think you would even get remotely close to covering 80% of the roads for 20% of the cost. Maybe if you isolate specific districts like Silicon Valley then your figure might be accurate. But for "normal" towns in Europe that wouldn't be the case and, as has been discussed elsewhere on HN, having some roads updated and others not would actually be more dangerous than having none redeveloped because humans would demonstrate the same complacency across all the roads once they're used to autonomy on the new roads. Also highways aren't the problem - they're significantly easier to determine in code and see fewer deaths per year. It's the smaller streets in towns that are the problem. Both from the perspective of statistical deaths and also from the perspective of the complexity of AI required to navigate them. They're what really need guide wires the most. However going back to highways: even on motorways you still have the problem of motorcyclists weaving in and out of traffic. So I'd argue that you haven't entirely eliminated the "cyclists" problem - you just now have faster and even deadlier ones. I should say that I do agree with you that there would be tremendous benefit if it were feasible. What I disagree with is just how feasible it is. > Regarding laying the cables in existing roads. We're already pretty good at "shooting" cables under existing roads with pneumatic solutions, and I'm confident that we could do so for rural roads at a modest price. But that's a different talk. Shooting the cables is the easy if you have pavements with cables already installed. If you don't then you need to dig up the road to lay them. And that's still only half the story. There's a whole load of costs involved in the planning stage; scheduling the work around busy periods (which sometimes means paying for night shift work) and organising the road closures. And it's the maintenance of that infrastructure after the work has happened. Remember that a lot of busy streets aren't even maintained currently. And it's doing that work sympathetically around historical locations (where applicable). There's a significant amount of more cost and work involved than I feel you're giving credit for. > If we want to develop an "AI" for this, it's not going to be ready in 10 year, nor 20 - we do not have the processing power, or understanding of the domain to do it at this point, and we're moving rather slowly. I completely agree that we're decades from a solution but I disagree that we're moving rather slowly. 10 years ago the stuff that are in consumer vehicles - even some entry level cars - would have been considered Sci-Fi just 10 years ago. But even if we are 50 years away, I can't see cables laid in twice that time. I mean half the towns in England and America don't even have broadband, let alone fibre, and you have consumers begging to pay for it. Who's going to foot the bill for redeveloping all our roads? Musk certainly wont. > Things like this move slowly - it's not going to be retrofitted, but agreeing on a standard and new vehicles can implement in gradually. The problem is you need it implemented on every vehicle for it to work. It's like anti-vaxers argument that "I don't need to get vaccinated when everyone else is" but it actually doesn't need to take a large percentage of unvaxinated people for an outbreak to still occur. Equally you only need one vehicle without a mesh network installed to case an accident that could kill several people - including those in autonomous vehicles. If you don't have total coverage then the mesh network isn't going to be reliable.