I don't know how they do that. I understand IF...THEN...ELSE but I don't understand how the learning process works. I see the results every time I get in my car and it works. When they switched from programmed instructions (v11 and earlier) to the end-to-end neural net training (v12 and later) the change was dramatic.
It's similar to ensuring that a human has been adequately trained.
Brakes didn't get much better for a long time. Four wheel drum brakes were awful.
In this case, safety means improvement. I doubt Waymo or anyone else is predicting zero accidents. But they are stating they will steadily improve; and Waymo at least has the evidence.
One might argue that despite Internet commerce being a thing for decades, that problem hasn't been completely solved:
arstechnica.com
If Waymo bot#736 screws up and learns something in the process, all Waymo bots learn the same lesson. If I screw up and learn something in the process, do you also learn the same lesson?
One might also argue that the problem has been mitigated to a certain degree.
True. But reasonable people may disagree about whether the level of security sufficient to enable e-commerce is equivalent to the level of security sufficient to enable autonomous driving.
Oh no? What does it mean?
I don't know that it's going to happen that fast. I do "know" that we are not going to figure out UBI in time, and we are going to be in for a world of hurt for a while. Most likely, Great Depression 2.00, We Added A Zero (tm).
Bot Lizards!
I'm 50, so I might be able to do that. Someone who's 25 now will never make it to 65.
I would imagine there will be a phased approach. Over the road highway driving is fairly predictable and also easy to automate.
The truck should be able to detect worsening traction and automatically slow down. I would guess that because of that, the autonomous truck is more likely to be the hazard that gets plowed into by human drivers than the one who's plowing into a pileup.
That's a stupid human trick right there. Human over-the-road drivers are paid by the mile, so there is a lot of motivation to enter conditions that are terrible and get through them as quickly as you think you can. An automated vehicle won't have the emotions or get-there-itis that a human driver will... And it can drive for 20 hours at 30 mph just as happily as it can drive for 10 hours at 60 mph.
We already have Waze, smartphones (iPhone anyway) with crash detection. Put those two together and you can get data very quickly.
Just like Tesla's FSD system already identifies worsening conditions, limits speed, and warns the supervising driver.
I'm literally looking out my office window at the side of a building 1,000' away (per Google Maps) and can make out bushes, people, and section lines on overhead doors. I can also notice movement in the foreground for objects much closer to me. So I've got better vision than all of those camera capability in a wider field of view. Again, I'm not saying it can't be done, I'm saying the number of camera layers, and ability to spot/detect objects (which is must be programmed to look for) is a lot more than what is currently being used. The offset that favors the automation is that it doesn't get its attention pulled away to something in the semi truck messing with radios, food, seat comfort, or a lot lizard that got picked up at the last truck stop, lol.
My car also reacts to threats it shouldn't be reacting to like a pile of leaves/shadows on the ground when backing up. Should be fun seeing what panic response the automated semi trucks do when it comes across something it misidentifies in the software.
The company doing this has been running these trucks for some time with safety drivers with no incidents. How much longer should they do so?
Peeking and any other technique could be developed. But you are impatient, so you want to pass the semi in front of you. That computer in the truck, will just plod along and only look to take action when the semi in front of it is going too slow, that there is a rule break. I doubt the programmers/trainers (depending on which solution) would bother over taking a semi until the speed differential is closer to ten MPH between desired and actual. The risks are really just not worth it.
Now this I can get behind (ha!). Having done a number of coast-to-coast drives at the behest of Uncle Sam, it was always mildly infuriating to have a truck swing over into the left lane and then spend the next 20 minutes/30 miles inching slowly ahead before finally getting back over. In the meantime, two miles of traffic has backed up behind that rolling road-block.
I tend to slow down a bit when I’m the one being overtaken in this scenario. Makes life a lot more comfortable for everybody.
I think that is mostly due to fleet trucks and the governors put in them. Back around 2006 I actually did the math on the distance and time from when they left to when they got back for another load on some owner operators when I contract managed a number of pork grow out sites. I didn’t have anything to do with shipping, only the load out crews, but those dudes ran very hopped up Pete’s and commonly averaged over 80 mph from north iowa to St. Joe Missouri (calculated using zero unloading time, only distance there and back, they had to be rolling ALOT faster than that to make the numbers they did).
You're a gentleman and a scholar. I'd buy you a beverage of your choice if I could. It generally wasn't terrible on the flats, but on some of the mountain passes it was like watching paint dry, presumably because the truck getting passed understandably didn't want to give away any momentum going up the hill.
Yeah, I suppose if I have to choose between a 18-wheeler coming up behind me at 100mph and a truck slowly overtaking in front of me, I'll take the latter since I enjoy living.
I drove semis on the farm as a kid…no scales on the farm, so the first couple of loads were occasionally a little heavy until we figured out how wet the corn was. Don’t think I was ever much over 90,000 lbs, though.
Been in enough tight spots that I know how quickly they can develop even when nobody’s trying to be an *******.
I suppose that would be up to the company and their risk-tolerance for massive lawsuits when the first accident occurs, lol. I'm sure it's been hashed out by their executives on what they think their exposure is.
This is also how many humans will loose jobs. Not only will machines be better/faster/cheaper, but they will be more insurable. At some point insurance companies will either charge more for humans and make them even more expensive, or simply refuse to write policies on human drivers because of the higher risk compared to machines.
That already happens. Tesla's insurance charges less when the percentage of miles driven on autopilot is high. Looks like that is available in Texas and Arizona so far and offers a discount when FSD is used more than 50% of the time. It's not a huge difference. 10% on the applicable part (collision?) of the insurance premium.
Hadn't read that. Very cool! Even 10% has got to be pretty close to paying for the cost of FSD?
One way or the other, the cost of paying those claims will come from the operator of the vehicle. Either the insurance company charges you or the vehicle manufacturer charges you. If the vehicle is safer, the overall cost will be lower.
Careful. That flavor of thinking is how we almost lost GA a while back...
No, we “almost lost GA” because people were successfully suing manufacturers who weren’t responsible for the damages.
I agree. But if we continue to let the manufacturers walk away from their part, there will be a limited incentive to improve quality. They'll push back, like boiler manufacturers did 200 years ago, but in the end...sadly probably after they kill a bunch of people...they'll become safer.
I know. But the difference is that everyone wants self-drive and the software industry isn't going anywhere. They've been operating without any real incentive to improve quality control forever, and working with that industry I don't see any effort at all toward self-improvement.
We almost lost GA because people sought deep pockets and payouts had to be large enough to make an impact, rather than a settlement for the million dollar cap of an individual's policy. Time and time again the courts found that the manufacturers absolutely were responsible, otherwise there wouldn't have been a threat to the entire industry.
I agree that free markets don't push for quality. That's why DOT has to create standards and mandate that manufacturers comply with them. Market forces alone are insufficient for many kinds of change.
I'm good with testing to set a baseline and eliminate risk of criminal liability, but not civil. If there are enough accidents with self-drive that they can't either self-insure or afford the insurance for the products, then they shouldn't be making them. I say that because the actual liability will need to be paid - medical bills, pain/suffering, etc. We shouldn't expect people sharing the road with self-drive to assume the risk for the other vehicles, nor should we have the government bail out the manufacturers.
IF liability can be proved, a suit will be successful.
And you'll find that is not correct. Perhaps read the supporting documents for the General Aviation Revitalization Act to see how it wasn't 100% fault of the OEMs. Plus there are plenty of tort examples of what actually happened and why. If what you state was true, then GARA would have never passed and GA would have disappeared.
In my experience, its not so much proving the liability, but who has the ultimate "standard of care" over that liability and most importantly which attorney is a better salesman in court.
If the industry cut the number of accidents in half, it would be a huge victory for society. And still plenty of accidents to get sued out of existence.
