[nishantvyas]

Correct.

It’s amazing, how much forgiving we can be for human errors ( accidents every year) but absolutely not for machine/autonomous vehicles, even when, statistically speaking, machines may make better decision much faster(or at-least no worse than human judgement)... I guess feeling/perception of being in control is more important to us...

Other interesting observation I find in every autonomous vehicle discussion is, how we only focus on edge cases... when in reality every tool that we use (including the car we been driving) today are built for general use case and operate under mostly a control environment.

Rather if we think autonomous car as additional pair of eyes and hands when we need it most might serve us well in short run before the technology gets mature over next decade or two.

I’ll be really happy and relaxed if my car can mostly (70-80%) drive it self to daily commute or next trip to LA; expecting it to be my chauffeur is bit too much, personally.

[beefield]

This is why I wonder why platooning technology is so much less hyped. Give me a platooning hardware kit to my current car, enough users after which I can join on my longer trips and literally more than 95% of my self driving car needs are covered. I really do not care if I need to drive a 10 minute stint on the city every now and then. And If I do, I can take a taxi. But getting my hands off the wheel and eyes off the road on the highway is what would have real utility to me.

[dmoy]

Platooning kind of messes up highway traffic because people need to either cross a ton of lanes to get in and out of the platoon (if the platoon is on the left), or non-platoons can't get on/off (if the platoon is on the right). If everything was forced to platoon on highways it would work. But that's like 30-50 yrs later, after the tech is introduced, barring some really radical legislation with huge popular and state support.

[beefield]

Not sure if it would be techically possible to have gaps in the platoon after every five cars, but at least it would be trivial to set max size of one platoon to something reasonable.

[dmoy]

Sure, yea, I don't mean like it becomes literally impossible to merge in. Just that for any reasonable size platoon, it disrupts more traffic than it saves. Given existing roads and the existence of not-platooned cars on the same roads, it doesn't really work.

Maybe it works for long haul trucking though.

[pnutjam]

You honestly trust some random driver at the front to make your decisions?

[beefield]

Well, that sounds pretty much like flying commercial flights or traveling by bus. So I guess based on my travel history, answer must be yes.

[pnutjam]

When I fly, it's a trained pilot. Even for driving is stricter licensing for cdl.

[masonic]

  amazing, how much forgiving we can be for human errors ... but absolutely not for machine

A given human fail produces one event. A flaw in autonomous driving software can mean thousands of failure events.

Also, when a human driver's negligence results in injury or severe damage, criminal charges result. That's a deterrent. With autonomous driving, you can't prosecute an algorithm.

[nishantvyas]

Yes. point about the charges would be a thing, something that should be debated...

would "use at your own risk" vindicate the company behind autonomous vehicle? or owner is responsible for his vehicle's actions? i guess never in the history, we had so much advance automations in direct hands of consumer...

As for the failure, I have reasons to disagree... if autonomous cars are working under "unsupervised learnings", my assumptions is, it most likely will makes different decisions for same scenarios based on data on hand.... so thousand's of failure events... though it may look similar may or may not end in same results... similar to how we would react when faced to some unknown situation on road... your scenario might more likely to play out for bad batch of hardware devices/sensors/lidar/camera etc in autonomous system...

[ghaff]

>or owner is responsible for his vehicle's actions?

If it's sold as fully autonomous, i.e. significantly beyond Tesla's system today, I don't see how the manufacturer could not have the liability. How comfortable would you be to use a car that could expose you to severe criminal liability because some company made a mistake with their software?

[masonic]

liability is assumed. I speak to criminal prosecution like an impaired human driver would face in addition to financial liability. The automated vehicle would face no criminal exposure.

The company responsible would also have a clear incentive to alter/destroy any damning evidence gathered in telemetry.

[ghaff]

>The company responsible would also have a clear incentive to alter/destroy any damning evidence gathered in telemetry.

Not saying it doesn't happen. But now you've gone from a product liability case which rarely has individual criminal consequences to actions that clearly do.

If/when we get to this point, it will be "interesting" though. Outside of maybe the medical area, there aren't many examples of consumer-facing products that, when used as directed, kill people because sometimes "stuff happens." And people generally understand that's just the way it is.

It's not out of the realm of possibility to imagine government-approved autonomous driving systems that insulate everyone involved from liability so long as they're used and maintained as directed. See e.g. National Vaccine Injury Compensation Program. I'm not sure it's likely but it might become a possibility if manufacturers find they're too exposed.

[TeMPOraL]

> I’ll be really happy and relaxed if my car can mostly (70-80%) drive it self to daily commute or next trip to LA;

There's a caveat here that this 70-80% must be contiguous and the car must be superhuman-level reliable in that segment. Otherwise, the "additional pair of eyes and hands" significantly increase the danger. If your car suddenly decides that it can't handle something and asks you to take over in the last second, you won't be able to handle it either.

[ghaff]

My assumption is that you get to this for some subset of highways in some subset of weather conditions with some special rules in place (maybe mandatory maintenance schedules?).

Which is actually a big win as long highways drives are boring and probably have a decent chunk of more serious accidents.

It doesn't give you the robo-taxi use cases that are what a lot of urbanites care about the most. But it would be a nice safety and comfort add-on for how a lot of people spend many hours of their weeks.

[skywhopper]

The problem is that you encounter edge cases on every drive, and you need to be ready to respond. The car may be able to handle 80% of the trip, but one of those edge cases will sneak up on you and the car. How long would it take you to regain situational awareness and safely maneuver the car in the event of some unexpected situation after you've been cruising in self-driving mode for an hour? Ten seconds? Five? Can you do it in one? What if the car doesn't realize it can't handle the situation at all?

Like any risk, you also need to consider the impact of getting it wrong. If an audio assistant gives you the wrong answer to the population of your hometown, no big deal. But if your car thinks everything is okay and drives you into a stationary fire truck on the shoulder of a freeway when you are travelling at 70mph, the downside of that edge case is infinitely worse.

Sure, humans can make these mistakes, too. But the fact is that your notional world where computers are able to make smarter decisions than humans about how to drive doesn't actually exist. No one has figured out how to make it work. And they won't anytime soon. They've solved all the easy parts. But it turns out there's a lot more involved in driving than all the billions of dollars poured into the problem so far can figure out.

[nishantvyas]

https://www.youtube.com/watch?v=9SexsvIO4vE internet is full of these examples.

My point is, computer with,

- more data, (historic on how to act on certain situation, live data for event i.e. sensor data, lidar/radar data, images) vs human driver who would not have access or the ability to process these.

- faster and parallel processing vs human driver

- single focus/goal (of driving from x to y safely and making appropriate decisions to achieve it) vs human driver (with "physical limitations", "emotions", "hormones" and other things that makes up "life") is more likely to be distracted...

computer with all of above advantages compared to human driver may able to make better informed decision much faster than human driver can do (and when it doesn't it's hard to know/prove if human driver would consistently make better decision every time for same situation)

having said above, I agree that tech is in its infancy and it's gonna take a decade or two to be matured and even after that human intervention just in time in some cases would be needed but for the most controlled/learned environment (which is 70-80% of total driving on day to day basis) these systems would be immensely helpful.