1) Even with robots there will be unknowns which the robots will have to deal with which will still require a margin of safety. Likely that margin of safety will be lower with robots, but definitely will not approach 100% capacity. (things like tire blow-outs, deer, even sudden rain etc).
2) Road wear will still need to be considered in any cost benefit analysis. Presumably road wear is less of a factor than jamming cars on the road, but I don't know that for sure and someone should do the math on that. There is also a point where if you exceed a certain capacity percentage, anytime road work needs to be done it would dramatically impact traffic since there is less of buffer to absorb the lack of lanes.
3) As we increase capacity, we may run in to additional problems that we cannot imagine. Bridges for example could be an issue as the amount of weight would be increased dramatically, even more remote a possibility would be damage from additional vibrations etc. Those are just ideas that come to mind, but there are probably many other risks we are not considering.
Overall, I am 100% for robotic cars, robotic passenger jets, etc. I just wanted to brainstorm a few future problems that may need to be considered as we move forward. The last thing I want to see is a 5 year setback.
Then again, the engineers at google and elsewhere have probably considered all these things which just seem new to me as a casual outside observer.
I think your broad point is correct; no system can be perfect. But most of your criticisms have straightforward rebuttals. Robots, being robots, are going to be inherently safer at the stuff you mention (tire failure, sudden obstacles, weather) simply because those things are "dangerous" precisely because they are unforseen and require the driver to take action quickly. Robots are faster.
Road wear is a cost, but it scales at worst linearly with traffic. I can't imagine anyone crying because we can now build ($$$$) half as many lanes of highways but much repave ($) them twice as often.
And the bridges things is just wrong. Bridges are obviuosly designed for maximal loads, which in this case would be a bumper-to-bumper traffic jam of even more weight than a packed stream of robots. And that happens routinely.
By broad point was not really clear, but could simply be stated as:
We will make a lot of assumptions about what is a linear expense, linear threat etc. without really knowing what will be linear until we pack an extremely tight grouping of cars on the road.
Taking my point further:
For road work, expense may go from linear to nonlinear because of something that seems completely counterintuitive. Imagine every car now has variation in the exact placement within the lanes. Highways have marks where the majority of cars drive, but variation in drivers accounts for variation in path. If every robot executed their driving in the exact 1 foot per tire space, it is likely that 1 foot per tire space would require a greater than linear road work. This could be solved by programing an automatic variation in all cars or possibly changing how roads are built so that they require only cement in those 1 foot areas.
Now, an example of a time when engineers thought they had it right, but didn’t think of everything:
Engineering has come a long way since the Tacoma bridge incedent, but we cannot underestimate what assumptions were previously made that will be incorrect when we start driving far more vehicles on roads. In the Tacoma bridge edge case, sustained winds caused flutter (self-feeding vibration) which caused the bridge to fail.
While I do not suggest this is a likely case, the fact is, we are probably not considering everything and there will be issues we do not foresee, but we should try to eliminate as many edge cases as possible.
To your point about robots being faster: Yes, I agree they are faster, but even the fastest robot will need some space to interact. In a situation where a truck on the highway has a blowout which causes large debris to separate from the truck, the robots still need some room to slow down. In the extreme example where cars are 1 foot away from each other, an accident like that would have tremendous negative consequences that even robots would have a tough time dealing with.
We will have self-driving cars, and they will likely be much safer than human operated cars, but we are going to also have setbacks and many of them can be avoided by looking to the future and trying to solve the problems early rather than later.
We don't want to mis-underestimate the complexity driving a car. The dynamics of obstacle avoidance, for example. Have you ever been hit by a deer?. Had a blowout (not a flat) on an SUV, say in a decreasing radius, off camber turn? Can you solve for some of these considerations? certainly. But you would need extra-ordinary technology. The dataset for camber-level 3D GPS maps, for example = large. What about a sinkhole or collapse of roadway al-la Embarcadero freeway? Or the Collapse of the Bay Bridge? What about hydro-planing after a rainstorm in CA? Lastly, ABS can be dangerous on snow, dirt and other imperfect road surfaces. How do you feather the brakes? Just some edge cases. On the cost side, the interstate system is ~complete. So, you wont be saving money by avoiding future construction, in proportion. You will just be speeding up the depreciation of the existing capital infrastructure. So, again, that's not a clear cut case there.
I'm not following those edge cases at all. Those all sound pretty fatal to humans too, or invoke driver skills ("feather the brakes", seriously?) that real-world drivers don't have. The point here isn't to be perfect, it's to be better. And that is, frankly, a very low bar.
Also: I'm very amused by your assertion that the interstate highway system (probably 10% of which is under expansion at any given time in any give metro area, and the remainder of which is rapidly approaching capacity and/or experiencing routine unplanned congestion) is "done".
Understood. So I will explain a bit more in depth. But, the things I laid out are pretty basic. The dynamic responses of a vehicle under motion works something like this.
(1) Balance. The car is never level. As you accelerate, there is inertia. The drive train and the chassis do not move in unison. They are connected by "Springs". This is like an airplane: attitude. So, the forces on the car and the road are different, but related. And there is a lag.
(2) Dynamics. Accelerate, nose up + ass down. Decelerate, nose down+ass up. The road is not straight? Similar for a turn (tilt left, tilt right). Now, combine. What happens? There is a dynamic weighting applied to all 4 corners. Go into a rt turn? Weight front left. Unweight rear right. etc. So, the point is that the friction is changing in each corner as you drive. The friction must exceed the energy of the car, or you will slide like on ice, etc. But, for the reason you brake in a turn, this is not always guaranteed, etc.
(3) Topology. Now, add some complexity: The road is not flat. This changes the calculation of the weighting for each tire. A tire on a 45degree incline is not holding as much friction as on a 2 degree one. This is like standing on a sand-slope, etc. Just basic idea. Now, The traction is a function of the weighting of each tire, plus the relative position of the tire to the road surface.
(4) Environment. Also, you have to consider a few other things: Do you know the co-efficient of friction of the road surface? Clean? Dry? Wet? WIth autumn leaves? What tyres do you have? What is the friction curve with relation to the rubber type and the ambient air-temperature? Oh, and by the way where is the tread depth? Today? Yesterday? These are all things that go through the mind of a trained driver and are not un-common knowledge (think: f1, rally). Same thing with left-foot braking, not using ABS, non-abs brakes on snow/ice/dirt/ etc.
(5) Complex System. If you cannot predict any part of this, you run the risk that the inertia of the car from speed etc >friction => loss of traction, accident, crash etc. This means: If you don't know the "camber" of the turn, your math is a problem. You might know that you are turning right at x degrees, but what will the turn be in 100m? constant radius? or not? What if you change lanes in the turn? etc. Now, you can topo-map solve this at high enough resolution. Eg., something akin to a race-track, in a video game. Go to Laguna seca, and put that in a computer. Problem here, though, is scale: like 1 inch topo variances or something, but that data set for CA state? Is huge. You don't have it. How would you even get it? What is your other option? Terrain acquiring radar? That might work. There is probably something in 60ghz and up that in theory might work. But you are line of sight constrained and now how far out can you look? 10m? What is the system's reaction time? How fast are you?
(6) Road hazards. Similar problem here. If there are abrupt changes. How do you acquire them? A sinkhole. A pot-hole (might break your wheel, etc) These are things not on a map-set. A live person just drives around them. But, what about a freeway? 6 Lanes, everyone is packed in like sardines. Does the guy in lane 6 know the guy in lane 2 is going to swerve? If he doesn't swerve (an breaks his wheel) what happens? The bridge failures are an example similar. If you are driving, you just look and see: no road. But, what if you are on auto-pilot? Who tells the computer there is no road? Same thing flash flood. Even worse, is hydroplaning risk (like 1/4 inch of standing water, say). If you are not driving are you paying attention? If you don't feel the steering wheel, how do you know? Can you put that into the computer?
That being said, they are doing amazing things with traction control systems on motorcycles right now. So some of this may get figured out. But you will not see a guy with all of this tech riding no hands/no brakes, etc. This is strictly in addition to user input and continual monitoring of the controls.
____________________________
The Interstate System
Do you have experience actually navigating? Like, say long stretches accross the USA? At continental scale? The Interstate systems is what it is. Its "90%" done. There is not 1000 mile segments being built. Max is 100 miles, and even that is incredibly difficult (permits, etc). Also, those are not "capacity" related build outs, you are talking only a couple of missing geographic links. New lanes/etc are a different animal (thats maintenance, for the most-part). Put another way, US is not going to rip out 2/3 of the roads and put 3x the traffic on the smaller footprint, to get to a break-even case. Now, you want to increase by 278% more cars/hours on the system? OK fine, but it will cost you. Its basic math, no? Also, the economics of construction don't work out that you save money by only "fixing" half the lanes or whatever. Lastly, consider the purpose: more capacity? History shows, no matter how many lanes you put down, the 405 fwy will be in Jam at 4-6pm in LA. Its not just rubbernecking. It is a larger social issue: people will procrastinate. Now they are in a "hurry". Oops. [As the saying goes, sometimes there is no engineering around stupid =D.]
Other Safety cases.
The point of these cases is not that they are per-se fatal. It is that they require real-time rapid-terrain-acquisition to avoid making the problem worse. Deer is similar. I doubt a basic Lidar is going to acquire a deer in the brush that jumps out in the road. It might, with thermal imaging, make it possible to see hiding off to the side. But, the calculation would need to be predictive in a what which is pretty amazing.
If automatic car implies networked car, routes that are under construction could easily be converted into toll roads (presuming there is a toll that will offset the congestion). I guess lots of people wouldn't like it, but lots of people already don't like traffic jams.
> Bridges for example could be an issue as the amount of weight would be increased dramatically
If that was a large concern, then wouldn't we see a lot of bridge failures during bumper-to-bumper traffic due to accidents and/or rush hour? I would hope that bridges are designed to handle the weight of cars being packed that closely together, seeing how often it actually happens in the real world.
You have to pay workers more at night which will increase costs in areas that currently use day workers. That is just what I can think of offhand. Likely there are no fullproof solutions to this problem, but rather OK solutions that will work in some cases and not in others.
Then again, the future probably has all road work done by robots too so this will be less and less of an issue.
1) Even with robots there will be unknowns which the robots will have to deal with which will still require a margin of safety. Likely that margin of safety will be lower with robots, but definitely will not approach 100% capacity. (things like tire blow-outs, deer, even sudden rain etc).
2) Road wear will still need to be considered in any cost benefit analysis. Presumably road wear is less of a factor than jamming cars on the road, but I don't know that for sure and someone should do the math on that. There is also a point where if you exceed a certain capacity percentage, anytime road work needs to be done it would dramatically impact traffic since there is less of buffer to absorb the lack of lanes.
3) As we increase capacity, we may run in to additional problems that we cannot imagine. Bridges for example could be an issue as the amount of weight would be increased dramatically, even more remote a possibility would be damage from additional vibrations etc. Those are just ideas that come to mind, but there are probably many other risks we are not considering.
Overall, I am 100% for robotic cars, robotic passenger jets, etc. I just wanted to brainstorm a few future problems that may need to be considered as we move forward. The last thing I want to see is a 5 year setback.
Then again, the engineers at google and elsewhere have probably considered all these things which just seem new to me as a casual outside observer.