Hacker News Reader

Comment by bankcust08385 | Hacker News Reader

Self-driving cars able to communicate intent and negotiate could be extremely efficient by reducing collisions and traffic.

From a standstill, all vehicles waiting could accelerate simultaneously rather than create pressure waves due to human reaction times.

With fully-autonomous coordination, might also be possible to do away with traffic lights and other control elements to negotiate scheduling of vehicles moving across each other so they cross intersections using precisely-allocated time slots without stopping.

View on HN

This is a terrible world for pedestrians, bicycles, or anything other than an autonomous vehicle.

Not necessarily? This depends on how the system is set up.

For example, cars could share the positions of pedestrians and bikes with each other to ensure that even cars with no direct line of sight are aware of them, making the roads safer for everyone.

Likewise, if traffic lights are integrated into the system, the waiting times could be much shorter as cars can dynamically slow down to allow pedestrians to cross, wihtout being contrained by fixed time blocks of green/red.

We already have systems that permit pedestrians to cross with priority over cars; they don't need lights.

They don't scale to really busy streets, and one of the failure modes would be perpetually blocked vehicles.

And this still leaves other road users that aren't autonomous cars up in the air.

Why, could integrate the beacons into signal pedestrian crossings forcing the cars to stop (whereas currently an absent minded driver might go through it), also doesn't stop development of other safety systems like object detection.

If you want to avoid traffic 'waves', so all vehicles accelerate and decelerate at the same time, you must remove _everything_ that might introduce unexpected variance.

Which basically removes people.

A simple fact is that faster moving traffic is necessarily less dense; the gaps between vehicles must be larger to account for small variations that matter more and more at speed.

I'd say it's more about reducing unnecessary stoppages where possible. The wave is triggered by someone braking ahead for whatever reason, we're looking to prevent the unnecessary wave, not necessarily stop the initial braking event as it may have been necessary.

Let's improve driver's ed and impose more licensing restrictions. Seems like pretty wide variations in standards between States.

Let's ban distracting billboards in more States, as in Maine, Vermont, Alaska, and Hawaii.

Let's expand annual vehicle inspections to more States.

More surveillance tech, automation, and regulations aren't the answer.

> you must remove _everything_ that might introduce unexpected variance

I'm not sure that follows. Cars that communicate can accelerate and brake together even in unexpected situations.

You need the increased space, which means you have the traffic waves.

You need the space because of variations in cars; some have better brakes than others, some may be heavier so need more time to slow down, others may be on wetter patch of road, etc.

And one car may not even get the signal, so only slows down when it observes the vehicle ahead of it doing, an observation that needs time.

Or a car starts accelerating as the one in front just stalls.

It may all be better than human reaction times, but for robustness, which is really very necessary, you're going to get the same dynamics.

And this all assumes only good actors; somewhat optimistic in my view.

Waves are caused by starting at different times, and connected vehicles can coordinate their start time. Yes, you need more space with speed, but a fleet can handle that with different acceleration curves instead of varied starts.

This isn't entirely true, connected vehicles would still have some delay due to radio propagation time, but it's ns per vehicle instead of hundreds of milliseconds. Additionally, you can entirely compensate for it in ways you really can't with humans.

Also a connected vehicle could potentially tell much if the car ahead is actively accelerating. Reducing it's need to brake if it can tell there's sufficient distance to negate braking (a bit more risky).

> Cars that communicate can accelerate and brake together even in unexpected situations.

A horde of cars where 100% of them consistently operate in a failure-free state and have comms that can't be hampered by the environment - that group could maybe do this.

Fair. If coms fail though, you're no worse than today. Vehicles don't need to travel any closer, just start together.

There won't be any. If you want to do that kind of stuff you'll take a wheelchair to the nearest "activity centre" where you'll be able to move your appendages around to simulate some kind of neolithic locomotion. It will be considered quite a niche pastime, though, as you can just take pills to remain happy and in shape.

I've been advocating for railside cameras so that the driver can see things further down the track. The accidents happen at crossings of which there are relatively few.

If you put a cam and a computer with a crosswalk it can rigorously figure out (and transmit) someone is crossing the road. Very much more so than a vehicle approaching from around the corner.

It will just be the new victim blaming: the cyclist got run over because they didn't have an expensive responder on their back!

It would be awesome on any grade-separated highway

all grade separated highways eventually exit onto not-grade separated roads, and often tailbacks are the result of delays happening off the highway system.

> often tailbacks are the result of delays happening off the highway system.

Only if there is something seriously wrong with the road system. A highway ought to have higher throughput than any surface road. What you describe is not normal or nominal

City centers are often places with intense demand aimed only at one or two exits off of a highway. Other cases are airports, sufficiently large shopping centers, large events, etc.

Is it? "Large blocks of vehicles moving in tandem" is a technology we have today: trains. Are trains that bad for pedestrians and bikes?

You mean the trains that go by once every 15+ minutes, and are confined to a track with no way for the operator to do anything but brake or speed up, compared to every few seconds with the ability for the driver to take control at any moment?

But I guess this would work/be status quo for non-autos if we kept the signals so peds and bikes knew they could still cross and probably not get run over by someone who decided to switch back to manual control.

> You mean the trains that go by once every 15+ minutes, and are confined to a track with no way for the operator to do anything but brake or speed up, compared to every few seconds with the ability for the driver to take control at any moment?

Yup, those trains! They'd have a shorter and more irregular schedule, but autonomous convoys would behave pretty similarly to that. Outside of emergencies like drivers assuming control to swerve into pedestrians, I guess.

Plus, like you say, no need to remove any of the infrastructure of really safety assumptions of today, just augment.

how many pedestrians and bikes are directly crossing busy train tracks at grade? traffic lights signal cycle every minute or so.

the issue is that unlike trains, roads are so numerous that they are hard to avoid, and it is financially unrealistic to bridge or tunnel for non-motorized users across every road, particularly if you want that crossing to be accessible.

> how many pedestrians and bikes are directly crossing busy train tracks at grade?

Near me? A few. Not sure how busy, but I get caught on my commute about once a week.

While I totally agree with your points, I don't see how this is any worse than today. Connected convoys don't need to go at every green light (like that one robot planet from Futurama) they can wait for pedestrians. The biggest difference would be clearing intersections quicker.

You say it would not make things any worse and I disagree.

If anything, it might codify how hostile parts of the US already are to pedestrians, with beg buttons that may or may not work properly. More efficient cycles with less gaps may give people even less time to dart across during a cycle where they may not have the green but there is no cross traffic.

If traffic clears the intersection faster, you can have more efficient cycles without changing the cycle time. The effect would be larger, not smaller, gaps.

That would assume that transportation departments in the US would step away from trying to boost car throughput at all costs. That’s their current MO; car travel is like a gas that expands to fit all available space.

The intersection of paths taken by pedestrians and trains is much more limited than that for pedestrians and cars.

And notably, trains don't stop for pedestrians.

Or another way:

Roads are a shared resource. Train tracks are not.

No, because they're confined to completely separate tracks, which they don't share with pedestrians or bikes at all. In fact, when the two meet, we call it a "train crossing" and have special "cross guards" to keep pedestrians, bikes, and other vehicles well away from the train. Otherwise, we generally try to keep the tracks away from other modes of transit, by either putting them in tunnels ("subways") or elevating the tracks, or just not putting them near roads to begin with and giving them special right-of-way.

None of this resembles the way we treat cars at all.

I guess different countries have different rules, but my country have classes of roads which are not shared with pedestrians or bikes. There are generally not that many traffic lights, but there are exceptions.

The biggest use of simultaneously acceleration in those situations would be around road maintenance and other situation where road speed or number of lanes are reduced, with heavy congestion as a result.

As both a driver and a pedestrian I would love that engineering solution.

PLEASE build human scaled walkways _away_ from roadways. In my climate zone, please also provide roofing over them to shade from solar and downpour events.

Crosswalks. The "car crossings" for pedestrians and bikes are called crosswalks. They're dedicated sections of road, typically selected for high visibility, specially demarcated and often even specifically signaled. High-speed traffic is generally kept away from other modes of transit by means of elevated highway sections (in high-density areas).

My dude, this is literally exactly what I had in mind.

Good. Go to any society on earth with excellent transit (Singapore, Korea, Japan, etc). They all HATE using it and love cars so much it makes American car enthusiasts look mass transit lobbyists.

Cars are freedom, cars are status, and the people who don’t want them are peons in relation to those who do. This is a fact of living in societies with the best transit in the world. Using it is simply admitting that you’re poor.

sounds great in the US where cars are first class citizens in cities, but it feels like a loss for pedestrians in more mixed cities.

Though I suppose, mixed cities will ultimately push cars out, which will separate the two better and allow the car world to do whatever automated works it wants without harming anyone

Not all roads are in cities, though. I'd be fine with "manual" driving in a city if I could turn on smart-cruise-control on the motorway and let it do its thing at 40/60/70 knowing that I can relax just a little bit more. Doubly so if in an EV world, the cars can talk to each other and a central network to say "I'm going to need to charge in X km, so I can use charger A, B, or C" and they communicate to minimise wait times across the board.

I can almost squint at this and see 'self driving cars over a long enough period of time in reality are just .... trains with cars connected by wifi instead of physical beams'.

Except able to go to different places, with the much lower cost of road infra vs. rail infra.

The system will not be used for efficiency in the wider sense, merely in the car centric sense of increased throughput.

As another commenter has pointed out such a system makes life for other road users: cyclists, pedestrians, horses, most uncomfortable (to put it it exceedingly mildly).

Sounds like reinventing trains, but honestly why not with all the existing road infrastructure

The 'tech bros reinventing trains' refrain fails to take into account that the same people would love 1/100th of the coverage of the global road network for rail.

Unless someone, for instance from remote, crack some cars to send false signals, let's say a police mandating stop when an armored bank van pass by, signaling to also open doors meanwhile another armored semi-autonomous car from a very active activist suddenly accelerate crashing onto an elementary school group on a trip stating was the activist driving, and the smart-blood test state he/she is on drugs and alcohol while he/she was effectively not and wasn't driving at all or controls was not operational being by wires... etc etc etc...

You can't design the world as anyone is a good actor. Most are indeed good actors, but most and all are different quantities.

It turns the whole world into a computer science problem. A distributed database of state with some malicious data. With various asynchronous processes that have different versions of the data. All needing to make decisions with incomplete data.

The point is that the holistic approach can't work because there no "single complex system" but "many individual actors" under our control: we can't program the world, we can only program some specific actor between semi-independent others. Yes, we have States, rules in them, agreements, standards etc but we can't still program the whole not considering the program a valid model of the reality.

That's why the approach should be very careful in what to trust or not. Not only: there are various situation where it's not possible to avoid a crash but there are few possible crash options, the human might choose badly (for him/her, for someone else) but it's a personal choice. A machine choice it's the responsibility of the machine vendor if any. So even if today there are essentially a lack of norms on that topic being "so new essentially not existent", tomorrow we will need to state clearly it a car crash while self-driving ALL consequences, positive and negative must be on the car's OEM. As a result such ADAS will be less and less "acting" trying to protect their OEM more than anyone else, defeating the initial purpose.

The sole solution for this responsibility is that such systems are not made by a company but by an open community, so they are a product of humanity not of some vendor, and we are all partially responsible and partially in control of them. This of course can only exists in a society where universities are OPEN and FREE for all, funded by the public not some private interests, so all can participate depending on individual skills and will, not on wealth. FLOSS must be mandatory to avoid making it a business at all.

Something theoretically possible of course, but very unlikely in current societies...

I wouldn't want all cars to accelerate simultaneously since I don't think they can all brake simultaneously

Not arguing for this, just thinking it though from a purely technical approach.

If the cars could come to a consensus about the maximum common braking ability between them, they could also coordinate all of them stopping at that rate.

The results of a brake failure would probably propagate through a larger amount of traffic if everyone was packed more densely.

>From a standstill, all vehicles waiting could accelerate simultaneously rather than create pressure waves due to human reaction times.

How would this work? If even a single car has a malfunction it would cause a massive pile up? The amount of work people will do to avoid using trains is insane

I assume the cars automatic braking and collision detection will just override whatever would cause an accident. Or a person's foot.

I was at a stoplight the other day and I was about 30 cars back. I could still see when the light turned green. I counted ~30 seconds before the car in front of me moved at all. I did not make it through the light.

If the cars could talk to each other, they could all start moving together (slowly) and then accelerate and spread out, resulting in much higher throughput and preventing traffic jams.

It seems this comment is paraphrasing the following video by CGP Grey: https://youtu.be/iHzzSao6ypE

Reply on news.ycombinator.com