10 ms·

Disagree entirely. Yes, the number of cars on the road and trips driven will increase, but this will not overshaddow the traffic advantage from SDCs. All in al

Disagree entirely. Yes, the number of cars on the road and trips driven will increase, but this will not overshaddow the traffic advantage from SDCs. All in all SDCs should increase road bandwidth about 5x by my estimates. SDCs can drive much, much closer together. They will also never be the cause of phantom traffic jams. For every human that switches to SDC, phantom traffic jams get less likely - a main cause of traffic - which again increases road capacity. We don't have to wait until 100% SDCs to start increasing road capacity either. SDC "convoys" travel as a single unit, and interact cleanly with human drivers. Regardless of SDCs, when road capacity is reached, traffic will happen. But SDCs have the potential to greatly increase road capacity.

saosebastiao 9y ago
> With only self driving cars on the road we have the ability to fix #2, and thus greatly increase road capacity and reduce traffic. Parity would be reached at, say, 4x more cars. This is fantasy land. This idea has been perpetuated by a few layman websites and extremely simplistic closed-loop simulations, but have never been validated by the more sophisticated traffic simulation systems used by traffic engineers. In fact, quite the opposite. Bad reaction times contribute to waves, but waves are just annoying and feel like the cause of congestion to the driver. According to the traffic engineers I know: 1) You can't ever fully get rid of traffic waves because reaction time is only a fraction of the speed mismatch correction problem, physics still dominates. 2) You can't ever fully get rid of speed mismatches. Even a plastic bag flying on the roadway is enough to cause them. 3) Traffic waves only feel like congestion to those who experience them first hand. Minimizing traffic waves doesn't decrease travel times significantly, it merely moves congestion to harder bottlenecks like traffic signals, highway on and off ramps, or merge points. 4) Braess' paradox only manifests itself when agents are acting rationally. As rationality increases, whether through better mapping and routing systems like Waze or by algorithmic rationality of SDCs, more bottlenecks appear that didn't previously manifest themselves under a lower rationality state. Unless you can somehow centrally control SDCs (which is unlikely IMO), this will only become more prevalent.
bryanlarsen 9y ago
5x seems extreme, but I'll certainly grant you 2x on the freeways. But if you put twice, or worse 5x, as many cars on the freeways, that'll just jam up the urban core even worse than it already is. And self-driving cars are probably less efficient than human-driven cars in the urban core simply because they're much less aggressive in their interactions with pedestrians and cyclists. This is a good thing, but it will increase congestion. Efficiency improvements can only add increase capacity by a fixed percentage. The massive improvements in cost ($0 in labour!) and convenience will cause usage to increase by orders of magnitude, IMO.
jcranmer 9y ago
> SDCs can drive much, much closer together. No they can't, not if you want them to be safe. The biggest problem we have with human drivers, most people will say, is that humans can't drive safely. And one of the biggest reasons that is is because we drive way too close to each other. In an urban environment, it is practically impossible to maintain safe following distance: someone will slip in and take that spot instead. Sure you might say that they can communicate with each other--they can't, not in any reasonably close timeframe for infrastructure (say, the next 30 years) where cars have to assume that human drivers are still on the road. A worse effect is that self-driving cars are more likely to strictly stick to rules of the road and safe driving than humans are: so longer following distances, slower speeds, strict right-of-way rules when trying to make a left onto a jammed road... I don't see how they're going to be better.
- adrianN 9y ago
  But about half of the necessary safety margin comes from reaction times. Computers tend to be much better than humans, so that autocars can get closer without compromising safety.
- randyrand 9y ago
  SDCs can form SDC only convoys. But maintain their distance from human drivers.

saosebastiao 9y ago

> With only self driving cars on the road we have the ability to fix #2, and thus greatly increase road capacity and reduce traffic. Parity would be reached at, say, 4x more cars. This is fantasy land. This idea has been perpetuated by a few layman websites and extremely simplistic closed-loop simulations, but have never been validated by the more sophisticated traffic simulation systems used by traffic engineers. In fact, quite the opposite. Bad reaction times contribute to waves, but waves are just annoying and feel like the cause of congestion to the driver. According to the traffic engineers I know: 1) You can't ever fully get rid of traffic waves because reaction time is only a fraction of the speed mismatch correction problem, physics still dominates. 2) You can't ever fully get rid of speed mismatches. Even a plastic bag flying on the roadway is enough to cause them. 3) Traffic waves only feel like congestion to those who experience them first hand. Minimizing traffic waves doesn't decrease travel times significantly, it merely moves congestion to harder bottlenecks like traffic signals, highway on and off ramps, or merge points. 4) Braess' paradox only manifests itself when agents are acting rationally. As rationality increases, whether through better mapping and routing systems like Waze or by algorithmic rationality of SDCs, more bottlenecks appear that didn't previously manifest themselves under a lower rationality state. Unless you can somehow centrally control SDCs (which is unlikely IMO), this will only become more prevalent.

bryanlarsen 9y ago

5x seems extreme, but I'll certainly grant you 2x on the freeways. But if you put twice, or worse 5x, as many cars on the freeways, that'll just jam up the urban core even worse than it already is. And self-driving cars are probably less efficient than human-driven cars in the urban core simply because they're much less aggressive in their interactions with pedestrians and cyclists. This is a good thing, but it will increase congestion. Efficiency improvements can only add increase capacity by a fixed percentage. The massive improvements in cost ($0 in labour!) and convenience will cause usage to increase by orders of magnitude, IMO.

jcranmer 9y ago

> SDCs can drive much, much closer together. No they can't, not if you want them to be safe. The biggest problem we have with human drivers, most people will say, is that humans can't drive safely. And one of the biggest reasons that is is because we drive way too close to each other. In an urban environment, it is practically impossible to maintain safe following distance: someone will slip in and take that spot instead. Sure you might say that they can communicate with each other--they can't, not in any reasonably close timeframe for infrastructure (say, the next 30 years) where cars have to assume that human drivers are still on the road. A worse effect is that self-driving cars are more likely to strictly stick to rules of the road and safe driving than humans are: so longer following distances, slower speeds, strict right-of-way rules when trying to make a left onto a jammed road... I don't see how they're going to be better.

adrianN 9y ago
But about half of the necessary safety margin comes from reaction times. Computers tend to be much better than humans, so that autocars can get closer without compromising safety.
randyrand 9y ago
SDCs can form SDC only convoys. But maintain their distance from human drivers.