Comment by mjevans | Hacker News Reader

Comment by mjevans | Hacker News Reader

Parent Full thread

This isn't what they studied, but a simple thought experiment shows a limited use case where traffic absolutely can be helped by speeding things up.

Let's imagine a world where self driving cars are required for freeway driving; this allows them to respond in a predictable and timely manor with respect to directives to avoid lanes (and not jump the line) as well as to go at an exact speed.

An event (accident, whatever) obstructs One of Five possible lanes for use (including the HOV lane) on a section normally rated at 60 MPH.

This reduces the capacity of the road by 1/5th, or alternately, one lane of flow should be re-distributed across 4 lanes (1/4th per lane).

Idealized solution: A temporary bypass zone is communicated to the self-driving cars. There is a lead up zone where the speed is expected to ramp up to the new target, followed by a merge zone where the lanes reduce and the flow is re-shaped to balance traffic evenly. The cars would then pass the incident zone in this higher speed lane-compressed flow, before expanding out and then slowing back down to normal speed.

Due to the unexpected nature of the incident there has also been a buildup of volume before the incident, so to alleviate that slightly higher flow will be allowed to drain the high pressure area.

In the above hypothetical case, for 4 out of 5 lanes nornally at 60MPH, I would prefer to imagine: (1 + 1/4) * 60MPH => 75, but round up another 5 MPH for the bypass flow for a total of 80MPH in the corrected flow area.

Many sections of freeway could likely support that safely ; they're often engineered very well but have conservative speed limits set for political and driver quality reasons. However some sections would 'cap out' at a lower actually safe speed; in those sections the above procedure could help minimize the traffic flow impact.

However... CGP Grey really nailed the real problem: https://www.youtube.com/watch?v=iHzzSao6ypE

> Idealized solution: A temporary bypass zone is communicated to the self-driving cars.

There’s so many great opportunities like this once the cars are self driving and connected. For example, a ten lane highway without dividers! 8v2 split depending on the time of day and traffic patterns. How many times have you been stuck in a lane with the other lanes in the other direction completely open?

Also, drafting. With human drivers it's not feasible to have cars travel 6 inches behind each other, because human reaction time is shitty enough that everybody would crash. With computer control you could easily have cars match speeds, then slowly get closer until they latch together and form a train. This'd both dramatically increase the carrying capacity of roads (with the default 1-car-length-per-10mph-of-speed rule, 7/8 of a highway is empty space), and it would improve fuel economy by reducing cross-sectional area and wind resistance.

We have a similar thing going on here in Vancouver. The computer controlling the cars centrally manages their location and velocity, as to ensure that there's a breadth of passenger coverage around the metropolitan area. It's also elevated to help streamline trip times too.

This is still a terrible idea. What happens when the car in front of you has an axel failure?

Oops, my car's instant reaction time is worth nothing because the brakes cannot physically prevent my imminent death.

...which can be mitigated with a different high-tech solution: accurate sensors. If some critical component is near imminent failure, the car shouldn't allow itself to be included in a high-velocity train.

But I fear we have a trust problem there: we wouldn't trust the industry to be accurate in their sensor assessments, since the prevailing assumption (I'm looking at you, HP) is that these sensors are used to maximize revenue instead reliability.

High quality brakes on all four wheels should be able to stop a car faster than an axle failure.

It seems like this is the exact same problem that was solved (to our satisfaction, at least) by trains over a century ago. If you're riding on the light rail and the train car ahead of you has an axle failure, what happens? We don't think very much about this case, because in practice it basically never happens.

It works because they're physically connected already, which would be really really difficult with cars (standardizing the connectors, etc etc). Agreeing on a standard protocol for communication is a hard enough problem to start with.

Nobody who's looked at the economics of self-driving cars realistically thinks private ownership is an option: if you're putting that much capital investment in a vehicle that could operate 12-16 hours/day, your amortized costs go up by a factor of 6-8 if it's only operating for 2 hours/day. If we get self-driving cars it'll be in the form of ridesharing or rental services.

Agreeing on a standard protocol for both connectors and communications is a relatively easy problem if you have an industry with a dozen or so operators rather than 300 million private owners. It also solves the other major reason trains don't regularly crash, which is regular maintenance.

Nobody who's looked at the economics of self-driving cars realistically thinks private ownership is an option: if you're putting that much capital investment in a vehicle that could operate 12-16 hours/day, your amortized costs go up by a factor of 6-8 if it's only operating for 2 hours/day. If we get self-driving cars it'll be in the form of ridesharing or rental services.

This same reasoning applies to non-self driving cars, but somehow we do have a few privately owned cars around. The only way self-driving would make it any different would be if self-driving cars were so much more expensive that only very few would be able to afford it. If you however try to estimate the shape of the demand curve, you'll note that even today that are plenty of cars sold at $100k mark, so unless self-driving cars are many hundreds of thousands of dollars each, you'll still see privately owned self-driving cars.

More importantly, once the technology is out in the wild, many manufacturers will copy it, and it will push down the markup for self-driving capability rather low. You can sell a self-driving Toyota Corolla for $100k if you're the only manufacturer of self-driving cars. Once Nissan and Ford have their own equivalent technology, you can no longer do that.

When I think of the economics of self driving cars I pull out of my keister a sub $10k price for that option and then estimate the monthly amortization cost at 'under $200/mo'

At that point I start worrying that self driving cars externalize the cost of congestion off the driver to the public at large. Everyone has to deal with congestion except the people inside their self driving cars yapping on their phones.

Then nobody intelligent has looked at the idea. There isn't much gain from car sharing: more then half of the cars in the fleet will be used for one trip in the morning and one in the afternoon - rush hour is much busier than any other time of the day. Even if a car could be used for a second trip, it needs to get to the next rider, which means travel back out to the suburbs empty - adding wear and tear to the car and burning fuel (CO2)

Some people will car share, but most will decide that the convince of having their gold clubs in the trunk is worth having their own for the little cost difference it will be.

Nonsense, if self driving is a $30k option on a camry, that makes it a 60k car to amortize. How many $60k+ BMWs sold last year?

Probably a small fraction of the total sales, given that the average new car price was about 3 times less (and many people buy second-hand cars). Therefore, private self-driving cars will only be available to a very small fraction of the population, and the rest of the population will have to use car-sharing.

You could open up lanes a quarter mile at a time even depending on traffic demands.

Eg. Between exit 1 and exit 2, 8 northbound lanes are open. Between 2 and 3 we have 6 northbound lanes.

It's would be pretty disconcerting to travel in your self driving car and head full speed into oncoming traffic only to shift lanes a few moments before collision though.

They had this in Montreal back in the day (i.e. the 1960s, when the risk of a head on crash was one you just accepted by driving).

There would be a bunch of lanes undivided from each other and over the lanes would be lights telling you which lanes were open for your direction of travel.

Movable barriers work too but are far more expensive.

Seattle has this with a physically separated highway. It’s closed in one direction and switches directions around noon.

Fun to wait at the entrance with a motorcycle. As soon as it’s open you have 5 miles where it’s physically impossible for another car to merge on.

Yes, nothing sounds more utopian to me than a world covered in 10-lane highways...

Hate to break it to you but the world is already covered in 10 lane highways.

I'm sure we'll start seeing autonomous driving only lanes in a few years.

Also introduce effective information to commuters to reduce flow: many drivers can vary their timing (come into work 2 hours later), or path (delivery schedule; go to a different office), or just cancel (go shopping tomorrow instead).

Reply on news.ycombinator.com