Och

iTrader: (3)

Respectfully, I also don't think you understand technology in depth. Once again, to negotiate all the possible scenarios on the road it is necessary to have full human intellect and senses. It is delusional to think that current technology is anywhere near approaching human intellect, and frankly I don't think it ever will.

The traditional cars are made to accommodate existing infrastructure, and human drivers can negotiate all sorts of scenarios. On the contrary, for the automated cars to be deployed en masse, they would require new infrastructure that accommodates them, and they would need to herd pedestrians, bicyclists, and human drivers. And that doesn't even address weather conditions, animals, mechanical breakdowns and damaged roads.

SW17LS

If you don't think in the future we will have AI that has "human intellect and senses" you're fooling yourself. A lot sooner that 15,000 years too lol.

Thats just a ridiculous statement. 15,000 years ago people in China had just started making pottery. There were still sabre toothed cats.

I have 100% confidence that there will be "100% self driving cars" at some point in the future of humanity. I can't even fathom being such a luddite that I would make a statement that anything will NEVER be achieved.

Och

iTrader: (3)

Well, if computer AI ever reaches the level of human intellect, or even outsmarts us, the first thing these self driving cars are going to do is run all the humans over so they don't have to drive us around anymore.

mmarshall

Running without humans is one thing, but are they going to figure out how to produce their own fuel? .......even if it is electric power.

Och

iTrader: (3)

If they so smart, let them figure it out.

Dave600hL

If you read my earlier posts you will see I am agreeing with you (The physics obviously states that a car no matter how good a brake system can stop it, if they are not detected in time nothing will help stop that car). And my take is if it was possible in that situation for a human to do better (At this stage of ML tech). As for my ML comment, I was merely stating that at some point the sensors are going to detect the object. But, in this case it is hard to say without the data, my guess is that the sensors detected something but it was not deemed an issue by program until it was just too late. If it was deemed an issue it could have taken evasive actions like swerving out of the way and would be able to do this safely b/c the program will have data on what other cars are around it or not and do this decision making in milliseconds. A human would have to have either memory of his/her surrounds at that particular moment or look in the mirrors and by that time it will be too late.

But, that is precisely the whole premise behind ML. If given enough data the program will be able to tell BETTER than a human that someone may possibly dart out in front of the car. As this tech gets better, there will be sensors all around the car detecting objects all around it to get as much data as is can.
For example, if you give the application a scenario where if it were to see an object on the sidewalk accelerate towards the road, the app can calculate how long it would potentially take to get to the road and in front of the car if it were to continue on its current path. It can then decide if to apply brakes or change lanes or swerve out of the way and b/c it has all the data from everywhere around the vehicle it can make the best decision for that situation all in a matter milliseconds. It is only a matter of time until an app will be able to make a better judgment than humans based on the circumstance at hand.

Machine learning.

MattyG

The NYT article shows a Google Earth image of the median past the crosswalk where the woman attempted to get across. Lots of bushes to cover her until she decided to dash across. Was she aware of what she was doing, ie. crossing into 40 mph traffic lanes and did she detect the oncoming car?

https://www.nytimes.com/interactive/2018/03/20/us/self-driving-uber-pedestrian-killed.html?hpw&rref=automobiles&action=click&pgtype=Homepage&module=well-region&region=bottom-well&WT.nav=bottom-well

SW17LS

I have no doubt that they can and will if people have the will to design machines that can grow to that capacity.

Och

iTrader: (3)

If an object is accelerating towards the road it's a relatively easy scenario, but even then the car needs to identify the object. Is its a human, or an animal, or perhaps a trash bin that was picked up by the wind? Would be helpful to know, especially if there is a tractor trailer behind you that will most likely rear end you if you slam on the brakes. But again it's a simple scenario, there are a lot more scenarios that computers simply can not apply judgement to - say a deer standing still on the side of the road, children playing ball on the sidewalk, a drunk person walking and wobbling on the sidewalk, erratic behaving person on the sidewalk, mechanical failures, potholes, snow falling down from a truck in front, oil leaking onto the road from the car in front, and so on. The autonomous car could be on fire and keep driving until catastrophic failure, or it could drive alongside another car that is on fire. This whole "AI" is just nothing like human intellect.

MattyG

^^^ In order for AI to work, it would need to work on a mass level then. You couldn't just have it on some vehicles and not on others. And you would need the infrastructure to help that AI work properly. In this case, you would have to sacrifice aesthetics and put up barriers to make sure that pedestrians cannot cross away from a marked lit crosswalk.

Or you would need specific autonomous lanes that would only allow such cars. This seems to be an attempt by Uber/Volvo to release a vehicle out into the wild and see if it was able to test well enough to handle the tasks. Now the question becomes, did the engineers ever look at a scenerio where a pedestrian shows up unexpectedly in the exact wrong place and at the precisely wrong time?

Och

iTrader: (3)

If you look at the location on google maps, its looks like a perfectly smooth road that is completely deserted. No cars, no pedestrians. This half assed wannabe autonomous car couldn't handle a single pedestrian! How would it fare somewhere in Manhattan with hordes of distracted pedestrians, cars, bikes, etc?

Och

iTrader: (3)

Ok, so like I said a few posts earlier, in order for autonomous cars to work we need to herd humans and build infrastructure to accommodate those self driving cars instead of vice versa?

Understand, a car is NOT autonomous until it can negotiate with all the scenarios on the road. If its going to require specific lanes and limit movement of pedestrians, then its not autonomous but rather a limited capability automated vehicle, and frankly its not worth the bother.

And I sure hope people take offense if governments decide to build barriers for pedestrians not to interfere with cars. We are not in **** germany and streets are for people.

Dave600hL

Then you don't understand machine learning. I am programming in it now and we have only just scratched the surface with the level of the tech right now. It will be able to do all those things you say and decide better, all it needs is the data. I am not sure if you understand how it all works, but the application needs data and tons of it. The more data it collects the more accurate it will be able to make decisions.
ie, Snow falling from a truck, it does not need to know it is snow, but if you show the application millions of scenarios of objects falling from traffic in front it will be able to decide what is the best course of action based on all those other million times an object fell from a vehicle in front. It will be able to see potential outcomes and decide which is the best coarse of action for the situation the car is in at that point. And it will have a detailed analysis of where exactly other cars positions are in regards to itself or other objects and in milliseconds make the decision on which way to go.

But, you have to understand that for ML to work it needs bucket loads of data and at this point even with all the data available today, it is only a drop in the bucket as to what it will need to function at the levels I am describing.

Och

iTrader: (3)

You do understand that at the end ML translates to algorithms and there are limitations, right? AI can never interact with the world in all the abstract ways that humans can.

Och

iTrader: (3)

A self driving F1 car can certainly beat a human driver, provided that its setting a lap record on an empty track. Not so much when there are other cars around.

Take a look at racing video games. They have very advanced graphics, and very advanced physics. It is AI that they can't nail. In every video game AI cars suck, especially around human controlled cars. And video games have relatively static environments and limited physics, objects and movement. Real world is a lot more complex.