patrick.net

I think the abundance of augmented driving technology that we have now, that we didn't have 10 years ago or more when the Self Driving car craze took off. Has now gotten to the point it over shadows Self Driving technology and is deemed more practical by the consumer than self driving cars.

I would like to see more technology put in cars, like bendable led screens, that can wrap around side posts and other car features where the worse blind spots are, to make those obstructions invisible, and the objects in those blind spots can be seen more clearly. I think in a decade or so, all of these cars with augmented safety features will communicate each other. I can see technology for getting on highways, and exiting from the far left lane, where cars will communicate with each other either speeding up and getting out of the way, or slowing down and giving way. Cars will speed up or slow down to provide appropriate distance between cars when traveling, or to prevent idiots like to drive up next to a car and pace them blocking the highway and seeing the pace. Technology that will either speed one of the cars up and slow the other down so they are staggered, and cars can negotiate around them.

Aren't self driving cars a form of mechanical masturbation that God hates?

Not about cars, it has always been about trucks. They mention self driving cars in hopes of drumming up investment capital from the general public. If you can at least handle the long haul portion of shipping on expressways, you'd cut labor costs probably by 80%. Self driving is the easy part.

If you've been in a Tesla or even high end car with lane assist, it's done already for 98% of what a trucker would drive milage wise outside of fueling up and delivery off expressway. Would likely be safer too since most accidents are caused by fatigue and not actual driving. Problem is it would wipe out millions of jobs overnight. It's more a problem of fuel or batter capacity to get the truck to go 1-2k miles without stopping.

I'm curious how AIs will respond to situations like a highway patrolman directing traffic to drive on the grassy shoulder around bodies and car debris in lanes. Also, areas where the lanes have been restriped differently from how the slabs were originally poured, causing two or more competing sets of visual cues as to which is actually "the lane." Night with wet pavement can be pretty tough.

Another 50 years before technology can ever handle this. It's all but disappeared, which means multiple companies encountered the same 'roadblocks' to development.

Perception, as I said before, is the issue, not making the vehicle react to it. Until a robot can figure out how to open a door without programmer guidance...

I think Obama's admin pushed unattainable glory that current technology wasn't ready for. Like someone from the Industrial revolution, making campaign promises about Broadband internet and more communication satellites. When they didn't even have electricity, electronics, tube technology, into the transistor, to the semiconductor, to the processor and computers and all of the evolution of the appropriate technology to support the next.

There's been a whole generation trying to claim the glory of Scientists 100 years from now. Talk about trying to steal valor.

And it's not just in this area of science. In the meantime though our personal computing power or control over the computing power the consumer has. Is diminishing every year.

Intel crashed and burned, burning the candle at both ends, trying to keep up with Moore's Law, bullshit the software creators were selling the hardware makers to make their latest models obsolete as soon as the first batch is sold out. Only for what? To end up with a computer end user who needs nothing more than a computer with the power comparable to 2003 in most instances. Because most users are using apps hosted in the cloud where all the processing is done, and the client today just needs to display the content. My point being they keep talking about Quantum Processors, and they don't even make software today, where an i9 Processor is required. An iCore 3 will work just fine for desktop app let alone cloud computing.

Automan Empire says

I'm curious how AIs will respond to situations like a highway patrolman directing traffic to drive on the grassy shoulder around bodies and car debris in lanes. Also, areas where the lanes have been restriped differently from how the slabs were originally poured, causing two or more competing sets of visual cues as to which is actually "the lane." Night with wet pavement can be pretty tough.

I don't think that a self driving car could handle Pittsburgh with a little snow.

Its somewhat interesting to see people's takes on AI and self driving cars from outside the space. While circling the target, the comments don't quite hit what is actually happening. And that is a mix of AI and traditional code on several different levels.

To be brief, AI is used to detect objects within vision systems. Integrating over several frames of video will likely give you a strong position and velocity for those detected objects. However, if the vision system fails to detect an object, there is still potentially lidar or sonar that can detect "something" that you probably dont want to crash into.

Developers make tools that show which sensors are detecting which objects.. perhaps by color coding them, and then also work on algorithms to create "sensor fusion" where two sensors "seeing" the same thing are able to reconcile it to actually being the same thing.

A third layer of AI is then added to this which is an object motion predictor. In other words, if a car is going straight in the lane next to us, but the road is about to turn, that car is likely going to turn in the same way that we are, and so we wont collide. What if the person's tire blows and they swerve into us? In the same way we can't predict that, neither can the AI. But what the AI CAN do that we can't is respond to the variation in the expected plan much faster than a human.

These plan generators need lots of work to react to road conditions, speed limits, lines on the road, unexpected object movement, etc. The key thing is, pretty much the safest thing you can do in AI the majority of the time is to simply hit the brakes. At some speed, say under 15 miles per hour, this is an almost foolproof plan to avoid all potential injury.

Its as we try cranking up the speed where there become situations where there simply isn't enough time or braking distance to avoid a collision due to some unexpected event (someone running a red light).

The place where regulators can help is requiring for instance that vision systems are supplemented by lidar systems as a backup. Lidar systems are already preventing human crashes. They could also start making driving tests for automaker AIs. We've actually already seen a reduction in traffic accidents with some of the AI safety systems. I'd imagine the first step in AI on the road might be the driving system simply letting the driver drive, but preventing accidents when possible (which is already rolling out).

I'm also guessing that at some point, the AI driver safety record will start being touted to lower insurance rates. And soon after that, people will think it is ridiculous to drive a car yourself with accident rates so much higher than when an AI is there. Then they will get used to the convenience of being able to drink when they go out for lunch or dinner, get out of the car and let it park itself, go fetch things from the store for you, pick up kids from school, or friends from the airport, or let you sleep as you go on car vacations to far off destinations. The car interiors will actually change to become "mobile living spaces" with specializations for dining, sleeping, playing games, working etc.

I would actually be surprised if garages don't change to become more of a "car dock" that gets you and your family and stuff into and out of the car more comfortably.

I know without a doubt AI that has been available and used in industrial applications are quite capable being hobbled together and prove proof of concept.
But the problem relies on sensors, hardware and software, that can fail eventually on the road, I see lots of glitches with the radar in my Mazda 3, that if my car was self driving I could get in a bad wreck. In industrial applications, an optical more motion sensor gets obscured or fails, the worse that happens is they lose a batch of product. Where in the real world they can cause millions of dollars of damage and untold human carnage.

AI is awesome in many industrial applications, and can be proven to work in less controlled environments. That's when humans, need to access and think of all of the ways those failures can happen when considering if they should. I don't think they wanted to do that with self driving tech.
Like those first generation of hoverboards, everybody wanted one. Then it started reporting how they were bursting into flames and causing house fires.
So then they came out with another generation of them that didn't catch fire. That too didn't last long, they had several at our huge office. They encouraged people to use them to create a hip cool fun environment. The HR dude was the biggest proponent of them. Then one day he fell and busted his ass and realized what I did the first time I saw them. They are a liability. They banned them and the non electric scooters.

I think had helicopters been invented today instead of in the early 60's when they really took off. The push would be for them to replace cars for commuting.

I agree with the sentiment about long haul trucking being the most realistic use case in the near future.

I think they will start using truck drivers to drive the truck from the factory to the freeway, get out and get driven back to the factory by a coworker. Rinse, repeat. Same for receiving - wait for the truck near the freeway and drive it through the city to the destination. The truck can drive the long haul itself.

Humans will still be used for city driving, with all of it's difficult unexpected scenarios to deal with, for a long time I think. Part of that though is because even if self driving achieves statistical superiority, fear from citizens, regulators, and politicians will still impede adoption, or make compliance costly. But long haul trucking seems within reach for AI - they mostly drive straight or hit the brakes. For scenarios where the truck needs to navigate something difficult on the hwy, I think they can just let someone drive remotely. Highways have excellent wireless coverage now on most routes, and our overlords are quickly building satellite internet to cover every square inch of the earth, ensuring always-on connectivity. Spacex, Blue Origin, and OneWeb all have plans to launch many tens of thousands of LEO satellites, providing good speeds w/ low ping. So, the truck can hit the brakes while a standby driver takes over remotely.

Zak, you seem to be in the industry. What do you think will happen with trucks?

I'll repeat what I said earlier, Amazon bailed out Zoox investors (if they're lucky, they got their money back) by buying them for over one billion dollars. That said, if anybody has an impetus to get self driving vehicles working, it's Amazon.

It would be much easier to build a self flying plane than a self driving car. No obstacles or people pulling in front of you. Autopilots are pretty advanced anyway, they can even do a full instrument approach. Taxiing could be controlled by buried wires in the taxiways and runways.

It's all so Cool And Hip, A.I., etc.

Legal liability can likely get in the way of it.

HeadSet says

It would be much easier to build a self flying plane than a self driving car

See Can Cargo-Carrying Drones Jump Over Air Freight’s Logistical Logjams?
Also, vertical takeoff and landing.

B.A.C.A.H. says

Legal liability can likely get in the way of it.

Since some random pedestrian was killed by a self-driving Uber test car in Phoenix back in 2018, the public should expect at minimum, a million dollars in survivor benefits for its victims, considering that in the current world, vehicular homicide involves jail time.

HeadSet says

It would be much easier to build a self flying plane than a self driving car. No obstacles or people pulling in front of you. Autopilots are pretty advanced anyway, they can even do a full instrument approach. Taxiing could be controlled by buried wires in the taxiways and runways.

And it'll be under the FAA, where safety and documentation are paramount, unlike the streets of Phoenix (or any other city) where the mayor threw in some kickbacks at Uber to test their vehicles on an uninformed public.

Booger says

I don't think that a self driving car could handle Pittsburgh with a little snow.

And much of the rest of the country outside of SoCal (and the Southwest) where there's snow, sleet, heavy rain, fog, and a host of other situations where the navigation could be rendered inoperable and you're looking at a major accident.

they still dont have an answer who is legally responsible for when the car crashes and kills someone.

TrumpingTits says

I think a better hypothetical moral test would be you're driving on a narrow street, and an out off control wide truck is barreling down the street head on to you.
On the right is 1 person or on the left sidewalk is 5 people, which side would you take?

I'm not going to be on the tracks, at a switch and wouldn't know which is which at that point. But if I were close enough, I would just yell, "TRAIN!!!"

TrumpingTits says

NancyPelosiHaircut says

they still dont have an answer who is legally responsible for when the car crashes and kills someone.

And they won't. Because politicians will have to be accountable for that. The insurance companies will push this issue, tho.

Right now, they're charging that Uber backup driver for negligent homicide in AZ.

https://www.wjsu.org/post/backup-driver-autonomous-uber-suv-charged-negligent-homicide-arizona

I'm sorry but this is wrong. When a vehicle is 'self-driven', a driver is not in some 'driving mode', He's in some distracted state, listening to music, reading, or whatever. Sure, he's suppose to take control when the computer tells him but in that case, why have the computer at all and just drive the car?

The company, Uber, is liable and not the test driver. They should pay the victim's family a million or two.

We all know what's coming: two tiered insurance. Guess which policy will be cheaper (bonus points, guess who the first pat.netter to complain will be?)

Hircus says

But long haul trucking seems within reach

Agreed, that's already happening. Also the $ saved when they caravan and draft each other makes it worth it.

Rin says

m sorry but this is wrong. When a vehicle is 'self-driven', a driver is not in some 'driving mode', He's in some distracted state, listening to music, reading, or whatever.

This might evolve in case law once end users are being ferried about by mature self driving technology. She was in a testing and development platform of what's still just an advanced prototype. This is grist for the "You had ONE fucking job!" mill. I do agree with the finding that the pedestrian was partially at fault for crossing in the dark with no lights while not in a crosswalk and disregarding approaching traffic. "Pedestrians always have the right of way" is abused by self-serving pricks as badly as "The customer is always right." People in the ghetto-ass neighborhood where my shop is located are especially bad about stepping randomly off the curb and mean mugging the people who had to screech to a panic stop to avoid hitting them. Recently I pulled out for a right turn when a homeless punk speeding the wrong way on the sidewalk ran his bike into my fender. He cussed me out, and complained, "You didn't even look!" I said of course not, you're riding against the flow NOBODY is looking for a speeding conveyance approaching in that direction, and to ponder having "He had the right of way" on his gravestone. That's when he punched the mirror, cussed me out more, and sped away in the wrong direction in traffic lanes now causing another car to honk and swerve 60 feet further. So yeah I'm unsympathetic to belligerent jaywalkers finally meeting their chrome plated fate.

Wait another 70 years.

"Well Done, Accord 452"

www.youtube.com/embed/F2iRDYnzwtk

Hircus says

But long haul trucking seems within reach for AI - they mostly drive straight or hit the brakes. For scenarios where the truck needs to navigate something difficult on the hwy, I think they can just let someone drive remotely.

Are all the trains broken?

Automan Empire says

and to ponder having "He had the right of way" on his gravestone

lol

Booger says

Hircus says

But long haul trucking seems within reach for AI - they mostly drive straight or hit the brakes. For scenarios where the truck needs to navigate something difficult on the hwy, I think they can just let someone drive remotely.

Are all the trains broken?

Why do they use long haul trucking now?

NoCoupForYou says

Wait another 70 years.

"Predigested food cooked by infrared. "

This was a time when magazines actually had Lime Jello with Green Olive jello mold recipes.

So what technical problem is the self-driving vehicle suppose to fix? It appears that the primary motivation is to remove ppl from the transportation industry and save some money.

Here's the thing, oil refineries are in a contained space. The same goes for manufacturing of rubber tires, steel pipes, etc. This development, between the 1870s and the 1970s, spurred the modern world and today, it's done in safe, contained environments where the average pedestrian isn't being lambasted by chlorine gas or sulfuric acid.

In contrast, aside from the firing of individuals, self-driving cars will not have nearly the same effect on society as the aforementioned. It's at best, a consolidation of money from national trucking firms like Schneider or local livery services like Boston Taxi, into one or two monopolies like Uber or Waymo.

HeadSet says

It would be much easier to build a self flying plane than a self driving car. No obstacles or people pulling in front of you. Autopilots are pretty advanced anyway, they can even do a full instrument approach. Taxiing could be controlled by buried wires in the taxiways and runways.

You know aviation better than I from past comments. I agree that drones or planes are more realistic than self driving cars. You can already plot a path with a drone from launch to landing. Weight is the biggest issue. If they can lessen the weight of batteries in half, we'd have drones flying people around tomorrow. I'm actually a little surprised that Musk isn't taking this on. It would benefit his EV's and also open up pandoras box for air transportation.

WookieMan says

I'm actually a little surprised that Musk isn't taking this on.

People will be falling from the sky onto other people. Let small package transport in narrowly defined lanes build out the tech first.

Tenpoundbass says

If the AI has NEVER seen a large white sign on the left hand side of the road

On his 570 AM Saturday morning program Wheels with Ed Wallace in the DFW area, Ed ridiculed A1 and the example he used was what happens where there's a blonde, blue-eyed seven year old girl with pigtails in the path of the A1mobile to make it sufficiently heart rending. That was two years ago and today that would be considered some kind of phobic thing.

Hircus says

Are all the trains broken?

Why do they use long haul trucking now?

Because they are stupid, I think.

call what? are you saying that there were people that actually thought self driving cars would happen before lets say 2050?

Something to contemplate. In three years time when Tesla is selling a sedan for $25k, Elon Musk will have all the data he needs to determine if autopiloted vehicles are safer than human drivers. At that point , a Tesla auto insurance subsidiary can offer a substantially discounted insurance plan to autopilot Tesla drivers. Trout will be the first to complain....

https://driveghost.com/

CBOEtrader says

https://driveghost.com/

That's a bit scary too.

I've always wanted cars to have inner rims that would allow them to ride on train tracks as well.

CBOEtrader says

https://driveghost.com/

"Equipped with 8 cameras capturing 360˚ HD clarity, Ghost sees it all, from merging vehicles to shadowed pedestrians."

Still no lidar?

Back after a break... Self driving Trucking is already here and operating on our highways.

TuSimple in San Diego has routes running from Pheonix to I think El Centro. Exactly what we thought... long stretches of straight, pretty empty road with almost no weather.

I think at this point they have moved off of safety drivers and just have remote control drivers. Not positive on that.

As far as cars, there are a couple competing business models out there. Uber and Google as we know, but then also Cruise(GM), Zoox, Argo (Ford), Baidu(China) and a few others. The GM, Ford, and Tesla plays are clear.. branded navigation for their car line. Uber and Google are taking the taxi approach, but on car manufacturers platforms, so not integrated. Zoox is starting with integration on Toyota Highlanders, but they are actually developing a "self driving platform" that is like a car, but actually isn't. It's more of a cube on an electric chassis, with no ability for a human to drive. This is also targeted at the taxi market, but clearly differently.

An interesting wrinkle in the Uber/Taxi market is I've heard nobody wants to share ride spaces in the COVID market.. Makes sense. Also of note are some other players like CommaAI. They wanted to develop a small video only plugin that consumers could buy third party to drive their car. They actually put something out on the market and its almost terrifying. They were shut down by regulators, but then put out an open source product that they claim is just for experimentation, and not for on-road use. I BET there are people letting that thing drive around for them though. I think Comma may now be out of business.

Finally, we have the drone space. It's interesting in that in 3d space, we can use communication to maintain separation, and there aren't really obstacles to need avoiding in the air in the same way they are on the ground. Some of the smaller drone AI makers targeting the military might be able to compete with Boeing since Boeing has never really been known for their software excellence. Obviously we have higher safety challneges though in that you can't just "hit the brakes" in the air with no energy maintenance cost. Seeing a place to land and doing it safely is a rather important thing. ALTHOUGH, it would be HIGHLY interesting to see dirigible Air busses flown autonomously. They can actually get decent speeds over medium distances (10-20 miles), if we could figure out how to load and unload them. Might be an interesting alternative to bulldozing in train tracks for public transport in the suburb commute spaces.

https://jalopnik.com/autonomous-cars-are-getting-in-the-way-of-emergency-res-1850393777

Autonomous Cars Are Getting in the Way of Emergency Responders in San Francisco