Any man who would use a horseless carriage is a wimp. Real men ride horses or drive a coach-and-four. Horses was good enough for the last 30 centuries; they’re surely good enough for the next 30. If God had wanted men to drive infernal machines he’d have …
Those words were spoken in heartfelt earnest by many of our great-grandfathers. Who were just as wrong then as you are now.
Driving is an ignorant waste of a human’s time. It’s labor unworthy of an intelligent creature. It’s the kind of drudgery for which God invented first servants and later droids.
Now *that’s *the forward-leaning attitude, not the backwards-leaning one.
The future: It’s coming whether you (any you) is ready or not.
No driver can put “100% effort” (if by effort we mean attentional focus) into driving for more than a few minutes at a time. The amount of energy required to maintain that kind of focus is exhausting. If you’ve ever taken a performance driving class or participated in some other activity where high attentional focus is required for safety you’ll appreciate how even just a few minutes of focused effort is straining, and how in the lack of an immediate hazardous or impetus the mind natural wanders.
Assistance devices such as cruise control, traction control, collision avoidance, and so forth do serve to make the driver, passengers, and general public safer, and are objectively far more capable within the limits of their functions than a human driver. However, they also cultivate a viewpoint among many drivers that they can pay even less attention, and the more these systems are about to take over the driver’s workload the less attention most drivers will pay until a hazard situation develops, which is likely too late for response.
The first 90% takes 90% of the time. The first 90% of that last 10% takes 90% of the time. The first 90% of the remaining 10% of the original 10% takes 90% of the time. Etc.
What is the history of how long does it take from something is invented to how long it takes to become standard on vehicles? The electric starter, the automatic transmission, air conditioning in vehicles? More modern innovations? Why should self-driving cars be different?
The other aspects of this thread have been covered, but I find the voice recognition to be absolutely wonderful on my devices. If you haven’t already, you may need to train your device to your voice, so that it can improve the accuracy. Upon doing so, I’ve been able to speak over music, with wind noise, etc. and it detects things reliably enough. Without background noise, I speak entire paragraphs, with punctuation, only needing minor context-specific corrections.
But getting back to the topic, autonomous vehicles can’t come soon enough. I love driving, and do so for leisure among other reasons, but many people don’t share in my enthusiasm and would (or do) prefer to do other things with their time, when behind the wheel. The sooner we can replace them with something more reliable and consistent, the better.
I also don’t think it’s that far off, from a technical standpoint. However, from a social/economic/policy/legal one, there are a few barriers, I’m sure.
I think 5 years is incredibly optimistic, but not beyond the realm of possibility. I think IF the automotive industry did a technology/engineering sprint akin to the sorts of things we’ve done in the past like the space race, I think it’s absolutely achievable to make self-driving vehicles (SDV) reliable and inexpensive enough to be available to the general public. The problems don’t seem so much technical as much as they are in terms of infrastructure, availability, and socializing the technology. Really, I think it will take more like 10-15 years.
The biggest issue seems to be public perception. When the first person died in a wreck it became national news. People expect perfection, especially when we’re looking at technology that holds people’s lives in it’s hands, but I think that’s unrealistic. The real question should be total safety record over similar mileage in similar conditions which, in my limited exposure, seems to be generally in favor of SDVs. Also, how do they compare directly to humans in these fatal situations? Would a person in the same situation not crashed? Certainly, some situations may still be difficult if not downright unreliable for SDVs, particularly poor weather and road conditions, or situations where there may be surprises like neighborhoods (kids running into the street), highly wooded areas (animals), or construction zones.
In this sense, I don’t think we should let perfection be the enemy of the good. SDVs could reduce stress and total traffic particularly in cities and urban environments. Hell, speaking for myself with my commute in DC, even with my commute not getting any shorter, if my car could take some of that mental effort off of me, particularly the high congestion areas, it would be incredibly welcome. I’d be happy to still drive myself in situations that it’s not able to do effectively like when it’s snowy or icy.
I think this is also the sort of thing that has exponential growth. With just a few SDVs out there, we see zero or negligible benefit to the group as a whole. As we get more and more, there’s less unpredictability in the system and everyone, even those without SDVs will start to see benefits. If they perform better than an average driver, we’ll see fewer accidents, which means less traffic. Similarly, so many traffic jams are due to slower human reaction time, unnecessary breaking, poor merging, etc. As more of the driving is automated, we’ll see many of these problems reduce. Even better, this means safer roads, faster commuting, higher capacity for the same road infrastructure and probably less energy used on transportation too. I think once we get enough of a density that these become clear, it will push the general population over the edge on wanting one, and we’ll see exponential improvements all around, then we’ll have really market pressure to focus on these and less on traditional cars.
The biggest problem is infrastructure. I think some of the issues may very well require significant changes in infrastructure. For instance, I have a friend with a Tesla, and while his isn’t self-driving, he did get a loaner that had self-driving while his was getting routine work done. One thing he mentioned was that it didn’t know when lights were red or green if he was the first in line. Will we need to retrofit traffic lights and stop signs to give some kind of signal that SDVs can read? Maybe we’ll need some additional safety features aside from those just on the cars themselves and we’ll need information directly from the road itself, perhaps about the conditions or the like (Google Maps and the car itself can’t tell me if there’s ice or standing water on the road itself). For the foreseeable future, there will always be normal cars, will we need to provide some kind of updates to those to work with new standards akin to how digital to analog boxes were provided for old TVs when switching to digital broadcasting?
Beyond that, a more terrifying aspect of infrastructure is that it will benefit from sharing information, but now we have a system that might be subject to hacking or even terrorist threats, or just plain major failures. They need to be designed to be helpful but not be dependent on them, because we could see a situation where entire regions could come to a screeching halt if something happens to this infrastructure. This isn’t an insurmountable problem by any means, but it does require a lot of design and engineering.
Even if true, the guy driving unsafely will claim the self-driving car is at fault, and if the “driver” of that car wasn’t paying attention, might win. I witnessed an accident where the guy 110% at fault tried to blame the woman driving the car he hit.
The current design assumption is that truly self-driving cars will need machine vision. They’ll see and react to the exact same road signs and signals as humans do. The baby step devices available in cars now are using radars, lasers, and ultrasonics to detect simple environmental features like [distance to car] ahead. That won’t scale to all the things the car needs to sense to self-drive as opposed to just [auto not tailgate]. It’ll have to be machine vision in (at least) the normal visible spectrum.
As such, there’s no expectation the built environment needs to change at all for the early autonomous cars to fit in. When eventually they become the norm (as you predict) that’ll open the opportunity for infrastructure changes to better accommodate their new capabilities. Things like interleaving crossing traffic through intersections so stopping for traffic lights is much reduced.
It’s precisely because of the danger of hacking writ large that the cars will probably mostly just look out their window and drive. Rather than everybody communicating with potentially untrustworkthy actors or sharing a hackable map database they’ll just drive. Their inherently more predictable driving compared to humans will enable enough traffic gains to be worthwhile.
Fortunately there’ll be a record of exactly what happened in the computer driving the car and, presumably, in the cloud somewhere. Plus, if it’s not the only self-driving car in the area, all the others will have records too.
Self driving cars won’t be infallible, but they will be able to drive far more carefully than humans, and should have plenty of failsafes to prevent errors causing accidents. Accidents won’t become impossible, but hopefully can be rare enough that any one is a big deal, like plane crashes or whatever, and can only happen when there’s a massive coincidence of problems.
In two or three generations I expect people will look back on the injuries and deaths from driving with horror, the way we look back on surgery before anaesthetics or antibiotics.
I am all in favor of self-driving cars, because I live in LA and loathe driving. I look forward with joyful anticipation to the day when humans are no longer allowed to pilot motor vehicles, because every day I look out my window at morons speeding the wrong way down my one-way street in a residential neighborhood with three schools less than a quarter mile form me.
However, even if five years is a reasonable timeline for SELLING safe autonomous vehicles, it’s going to take 50 for the benefits to be apparent. The rollout will start with high-end cars, then in maybe 20 years they will become standard, but meanwhile there’s a squillion regular cars on the road driven by people who can’t afford the tech, or would rather own a car that can be piloted the wrong way down my street at 60 MPH.
All the lawn guys I see are driving ancient pickups. They’re not gonna run out to buy new cars. I saw a VW pickup the other day and I think it’s over 50 years since they were sold here.
I’m not sure about this. There really isn’t a way to prevent people from “gaming” the system. If traffic lights are eliminated, it would be too easy for people to override the autonomous features and go quickly through if the other cars are set to watch out for cars “breaking” the rules.
Things could quickly break down into a Wild West scenario with no rules. Or am I not understanding something key?
A properly programmed vehicle will not find itself in that situation. Navigation amongst various obstacles is an exercise in math, a thing that computers tend to be kind of good at. The self-driving car will identify the developing situation far in advance of when a human driver would and react to it long before it becomes a crisis.
Of course, proper programming is a story unto itself. If a driver ignores warnings to take control, the car ought not simply keep going until it crashes, it ought to stop, and force the driver to operate it manually (some starry-eyed technophiles speak of cars that have/will have no option for manual operation, which I consider not a very good idea). Good programming foresees as many edge-cases as possible and addresses them before they become a problem.
And sensors will eventually be able to handle weather. We think in terms of a visual world, such as we can actually see, but there is really no reason an automated vehicle has to suffer that limitation. A sensor will be developed eventually that will allow the vehicle to see the surface under the snow, to figure out where it should be going (unless the operator is intent on brodying across that snow-covered field). Sensor tech, of course, does not have to be limited to automated vehicles: some drivers would benefit greatly from a car that would warn them not to try that.
Is this true? I think we’re seeing situations where the autonomous cars are not doing well, so the question in why are what needs to be done about it. Manufacturers are obviously not going to be open about it now.
Of course, proper programming is a story unto itself. If a driver ignores warnings to take control, the car ought not simply keep going until it crashes, it ought to stop, and force the driver to operate it manually (some starry-eyed technophiles speak of cars that have/will have no option for manual operation, which I consider not a very good idea). Good programming foresees as many edge-cases as possible and addresses them before they become a problem.
[/quote]
I’m confused because the article you link to is about the driver of the car specifically stating that that the car did not provide warnings prior to the crash.
You can do that now at any four way stop-sign intersection. The other drivers are likely to pay attention and you can “sneak in line” if you so choose. The reason it doesn’t break down into a wild west scenario is that it increases your risk for very little gain, so it’s only done by the very few stupidest people. That still applies when most of the other actors are automatic.
When I look at how the development of ADVs are progressing I have to wonder if maybe the DOT (or the equivalent ) shouldn’t develop a framework like the one that is used for computer routers (OSI) where each vehicle is considered a “packet”, but leave the development “open source”. The lowest levels of infrastructure have to support the levels above, and the higher levels are restricted by the limits set at the lower levels, and they all have to work with each other.
Physical Layer- Roadways
Data Link Layer- Traffic Handling Algorithms, GPS and external sensors (traffic cams, etc)
Network Layer- Vehicle to roadway network negotiation
Transport Layer- Vehicle to traffic negotiation
Session Layer- Vehicle/Vehicle negotiation
Presentation Layer- ADV multisensor integration including driver senses
Application Layer- Man-Machine Interface
The idea is that each layer is closer to the driver the further down the chain, and my example could probably be better set up by actual programmers/mathematicians/engineers. Until we can have all of that sensor fusion I suspect that like most game changers it will be some time until it reaches the rural areas. I could quite easily see some HOV lanes become auto drive only in a major urban centre within 5 years quite easily.
