4 ms·

> 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 outl

> 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. I did not say 80% of the road, but 80% of the travel. Most of the kms driven are on highways, ring roads, and main roads into larger cities. > 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. The car will pull over if you don't take over. I fail to see the issue in that. I'm advocating for a system that will autonomously drive where the infrastructure allows, not everywhere. If the system can handle driving at the highway for me for 2 hours, I'll be significantly more fit for driving the last 20 minutes. > 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 No you do not. They can even remove a small patch of tarmac, "shoot" and entire sewer line a long way and remove patch of tarmac at the destination. That's how they get optical fibers under roads and brick paving without digging it up. > 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 I never said it wasn't extra work. A cost benefit is needed, but the current AI approach will either cost many lives, or we agree on a better more expensive way. > If you don't have total coverage then the mesh network isn't going to be reliable. Again, a 100% solution is never happening. But by legislation we can increase adoption slowly, but steadily, until and past the point where we see the many benefits. In the EU all new cars must have DLR, AEB, TPM, ESP, ABS, seat belts, ect. - this would just be another thing new cars must have to ever be allowed to be sold on the market. It should not be up to the manufactures to agree, because they all want their solution, not an interoperable one.

hnlmorg 7y ago
> I did not say 80% of the road, but 80% of the travel. Most of the kms driven are on highways, ring roads, and main roads into larger cities. I don't think your experiences are typical for most people the world over. For example in the UK, we don't have that many motorways so most of the commutes by cars are along busy "normal" roads or dual carriageways (which are a little like motorways but with frequent junctions like roundabouts that have complex social interactions and thus might need AI to navigate safely anyway. But even there, most people people in the UK don't spend 80% of their drive on dual carriageways - let alone motorways. Instead we have "main roads" which join towns together. Commuter traffic in the towns in the Netherlands where I've been were more focused on public transport and bicycles than they are cars - though that's not to say they don't also have a heavy motor-vehicle traffic as well. I remember when I travelled around America thinking how much different attitudes to transport were to the UK and many European countries. I think that's in part because of how much farther away places are in the US. But I'm now drifting into speculation. My point was your observation about travel behaviours isn't globally representative. > The car will pull over if you don't take over. I fail to see the issue in that. I'm advocating for a system that will autonomously drive where the infrastructure allows, not everywhere. If the system can handle driving at the highway for me for 2 hours, I'll be significantly more fit for driving the last 20 minutes. Again, motorway driving is the easy part. It's where there is less to observe, low risk of people stepping out in front of you etc. It's where there are fewer road deaths too. And it's also the easiest part of driving to automate. So motorway driving isn't where we should be concentrating our efforts. > No you do not. They can even remove a small patch of tarmac, "shoot" and entire sewer line a long way and remove patch of tarmac at the destination. That's how they get optical fibers under roads and brick paving without digging it up. That's not how they do things here :) But in any case I did say that's still only a small part of the overall cost and orchestration needed in deploying cables. > the current AI approach will either cost many lives Will it though? That sounds like baseless speculation to me. Any who's to say your solution wouldn't also have glitches that result in a fatality? Ultimately you're still depending on software to make decisions and even robotic lawn mowers aren't without their own odd hiccup. > Again, a 100% solution is never happening. I know. That was my point. It's not going to happen and partial coverage isn't good enough. > But by legislation we can increase adoption slowly, but steadily, until and past the point where we see the many benefits. 50% coverage isn't enough. 90% coverage isn't even enough. 99% coverage isn't even enough but at least you're starting to get close. And how long will that take? Lots of people are still driving 20 year old cars every day. So lets say 10 years to design, develop. 5 years to then standardise the specification. 5 years of bureaucracy to get legislated then another 5 years for the legislation to become effective (because governments like to give a warning). They you need to wait 20 to 30 years until most people have bought a new car. That's 55 years at a minimum and you'd still not cover the thousands of people who like to drive classic cars. So you might as well double that figure. By which point AI has long since caught up with the ability to sense other cars on the road (because we've already mostly solved that specific part of the problem anyway) and at a fraction of the cost. I've already said that if we were to build infrastructure from scratch then your approach is definitely the way we should progress. But with the way our road infrastructures run at the moment: the vehicles on there, the hazards that can occur, the cost of maintaining the infrastructure and how that's even funded; well your solutions are simple not feasible for the real world. AI is one of those occasions where worse is better. I completely agree with you that your solutions are better - technically speaking - at solving specific problems related to driverless cars but it doesn't solve all the problems, it's more expensive to implement, can't be deployed everywhere, and yet it will still take longer to roll out than our current tact of developing AI. When you actually look at each problem in detail your proposals fall far short of being workable. These things aren't built without considerable thought put in them and I seem to recall that early prototypes of driverless cars did work on the principle of having smart cars that communicated with each other and smart roads too. Essentially everything you've been advocating. So it's not like you're suggesting anything that hasn't already been tried. But every lecture and documentary I've watched on this subject come back to the same final conclusion: for self-driving cars to become a thing they need to build the car to drive like a human would because changing the roads and the cars on it is actually a harder problem to solve.
- hvidgaard 7y ago
  I'm from Denmark, and my experience ranging from all the Nordic countries, Germany, Spain, France, Italy, and a bit in other countries but not enough to have any real experience. We could turn this system around. It would reliably cover the vast majority of semi transport if it just worked on major highways in Europe. Drive the truck to the highway on ramp, jump out and let it drive 1000km before someone picks it up and drive it "the last kilometer". I think your timeline is overly pessimistic. 55 years ago way 1964 - a lot have happened in that time frame, and we put a man on the moon and a rover on mars in far less time. It can happen significantly faster. If we decided to start developing now, I believe that within 5-10 years we would get near a specification and legislation. That leaves changing the cars and that only have to be done gradually. We know so little about this problem domain that it's best done gradually and carefully. I think we can agree to disagree. I'm not convinced that weak AI will be anything but an enhancement to the human driver. It's my belief that we don't even have a grasp of the AI domain to do this yet. Perhaps we will get it, perhaps we will not. Nothing that has happened in the AI space is new from the theoretical point of view. Neural networks was discovered in 1943, we just recently got the computational power to use it. But it doesn't scale and it's a black box approach. And it's easily fooled (see recent reports of a simply sticker on the road making a Tesla drive against traffic in the wrong lane). The strong AI field haven't seen any real progress or breakthrough in a long time, and people still argue whether it's even possible. To put it into numbers, a worm, which we have not successfully modeled the brain of yet, have around 300 neurons. A human have 10^11, each witch have a far a greater connective complexity . Strong AI is not coming in the next 100 years, if at all.

hnlmorg 7y ago

> I did not say 80% of the road, but 80% of the travel. Most of the kms driven are on highways, ring roads, and main roads into larger cities. I don't think your experiences are typical for most people the world over. For example in the UK, we don't have that many motorways so most of the commutes by cars are along busy "normal" roads or dual carriageways (which are a little like motorways but with frequent junctions like roundabouts that have complex social interactions and thus might need AI to navigate safely anyway. But even there, most people people in the UK don't spend 80% of their drive on dual carriageways - let alone motorways. Instead we have "main roads" which join towns together. Commuter traffic in the towns in the Netherlands where I've been were more focused on public transport and bicycles than they are cars - though that's not to say they don't also have a heavy motor-vehicle traffic as well. I remember when I travelled around America thinking how much different attitudes to transport were to the UK and many European countries. I think that's in part because of how much farther away places are in the US. But I'm now drifting into speculation. My point was your observation about travel behaviours isn't globally representative. > The car will pull over if you don't take over. I fail to see the issue in that. I'm advocating for a system that will autonomously drive where the infrastructure allows, not everywhere. If the system can handle driving at the highway for me for 2 hours, I'll be significantly more fit for driving the last 20 minutes. Again, motorway driving is the easy part. It's where there is less to observe, low risk of people stepping out in front of you etc. It's where there are fewer road deaths too. And it's also the easiest part of driving to automate. So motorway driving isn't where we should be concentrating our efforts. > No you do not. They can even remove a small patch of tarmac, "shoot" and entire sewer line a long way and remove patch of tarmac at the destination. That's how they get optical fibers under roads and brick paving without digging it up. That's not how they do things here :) But in any case I did say that's still only a small part of the overall cost and orchestration needed in deploying cables. > the current AI approach will either cost many lives Will it though? That sounds like baseless speculation to me. Any who's to say your solution wouldn't also have glitches that result in a fatality? Ultimately you're still depending on software to make decisions and even robotic lawn mowers aren't without their own odd hiccup. > Again, a 100% solution is never happening. I know. That was my point. It's not going to happen and partial coverage isn't good enough. > But by legislation we can increase adoption slowly, but steadily, until and past the point where we see the many benefits. 50% coverage isn't enough. 90% coverage isn't even enough. 99% coverage isn't even enough but at least you're starting to get close. And how long will that take? Lots of people are still driving 20 year old cars every day. So lets say 10 years to design, develop. 5 years to then standardise the specification. 5 years of bureaucracy to get legislated then another 5 years for the legislation to become effective (because governments like to give a warning). They you need to wait 20 to 30 years until most people have bought a new car. That's 55 years at a minimum and you'd still not cover the thousands of people who like to drive classic cars. So you might as well double that figure. By which point AI has long since caught up with the ability to sense other cars on the road (because we've already mostly solved that specific part of the problem anyway) and at a fraction of the cost. I've already said that if we were to build infrastructure from scratch then your approach is definitely the way we should progress. But with the way our road infrastructures run at the moment: the vehicles on there, the hazards that can occur, the cost of maintaining the infrastructure and how that's even funded; well your solutions are simple not feasible for the real world. AI is one of those occasions where worse is better. I completely agree with you that your solutions are better - technically speaking - at solving specific problems related to driverless cars but it doesn't solve all the problems, it's more expensive to implement, can't be deployed everywhere, and yet it will still take longer to roll out than our current tact of developing AI. When you actually look at each problem in detail your proposals fall far short of being workable. These things aren't built without considerable thought put in them and I seem to recall that early prototypes of driverless cars did work on the principle of having smart cars that communicated with each other and smart roads too. Essentially everything you've been advocating. So it's not like you're suggesting anything that hasn't already been tried. But every lecture and documentary I've watched on this subject come back to the same final conclusion: for self-driving cars to become a thing they need to build the car to drive like a human would because changing the roads and the cars on it is actually a harder problem to solve.

hvidgaard 7y ago
I'm from Denmark, and my experience ranging from all the Nordic countries, Germany, Spain, France, Italy, and a bit in other countries but not enough to have any real experience. We could turn this system around. It would reliably cover the vast majority of semi transport if it just worked on major highways in Europe. Drive the truck to the highway on ramp, jump out and let it drive 1000km before someone picks it up and drive it "the last kilometer". I think your timeline is overly pessimistic. 55 years ago way 1964 - a lot have happened in that time frame, and we put a man on the moon and a rover on mars in far less time. It can happen significantly faster. If we decided to start developing now, I believe that within 5-10 years we would get near a specification and legislation. That leaves changing the cars and that only have to be done gradually. We know so little about this problem domain that it's best done gradually and carefully. I think we can agree to disagree. I'm not convinced that weak AI will be anything but an enhancement to the human driver. It's my belief that we don't even have a grasp of the AI domain to do this yet. Perhaps we will get it, perhaps we will not. Nothing that has happened in the AI space is new from the theoretical point of view. Neural networks was discovered in 1943, we just recently got the computational power to use it. But it doesn't scale and it's a black box approach. And it's easily fooled (see recent reports of a simply sticker on the road making a Tesla drive against traffic in the wrong lane). The strong AI field haven't seen any real progress or breakthrough in a long time, and people still argue whether it's even possible. To put it into numbers, a worm, which we have not successfully modeled the brain of yet, have around 300 neurons. A human have 10^11, each witch have a far a greater connective complexity . Strong AI is not coming in the next 100 years, if at all.