If you don't suffer from Red/Green color blindness/weakness
like ~7-10% of males do "the grass is greener over the septic tank"
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
- Status
like ~7-10% of males do "the grass is greener over the septic tank"
level 5 is attempting to replace the driver, as best as I understand the definition.
Expecting the system to not require intervention if a human driver requires intervention would seem to be beyond the scope.
If I know a flash flood is coming and therefore flag you down to avoid crossing the bridge over the creek, even though you have no way of detecting it, do you feel it's within a Level 5's ability to be able to do so?
If a human has no way of knowing that a tsunami is coming, and decides to drive down to the waterfront, and a policeman has to intervene, do you expect a Level 5 system to not require the same intervention?
I think you are describing a "Level 6: Omniscient", system...
Reciprocity
Active Member
Weather should not impact FSD anymore then it would a person. Meaning if its to dangerous to be on the roads as a person, then its probably weather conditions that are extreme and no car should be on the road. The car will not need to see the road so snow covering the road is not an issue. The car should not be on the roads during a hurricane or tornado and those are situations where people should not be on the road either. The snow would have to be so deep as to block out all landmarks, which could happen, but people couldn't drive in those situations either. The bar should not be set above driving in conditions that no human can drive. The bar should be set to conditions where people can drive safely and no higher. There are some cases where there could still be issues like after a very large snow storm and snow is plowed and covers many landmarks on the side of the road, but this should only impact areas where there are not many landmarks. Traffic lights and most signs will not be covered by plowed snow. Maybe some rural areas where landmarks are much more spaced out could be an issue, but that really depends on how good the maps are and if the car can see the road on a plowed road, which has an new advantage of having snow walls as boundaries that the car should be able to see. The car could also position itself based on other cars around it, though in most rural areas that would be impacted would not have much traffic.
These are certainly complex problems, but any Level 5 system will have to be as good or better then people at negotiating the problems. Remember that there are people who have never driven in certain conditions and have never driven on certain roads. This system will have driven on every road multiple times and some roads millions of times. This system will have driven in every form of weather that is local those same roads multiple times. It is this advantage that must compensate for what will inevitable be an imperfect system. It only needs to be 10x better then the average humans not perfect.
Reciprocity
Active Member
If you don't suffer from Red/Green color blindness/weakness
Hah.. no I am not color blind at all. I always pass those tests where you look at the color dots and can see the image. We just had our system service so I know its good to go and I wont drive on the lawn until I find out exactly where it is. Thanks for the heads up!
Has anyone started implementing a "self-driving" computer system which can respond to a policeman flagging it down? Or which can take verbal instructions from a human? Not yet, as far as I can tell. I don't know of a single self-driving system which responds to someone at the side of the road frantically gesticulating by pulling over and asking "What's wrong?"
In that case, *when the policeman gives instructions, a human driver must implement them*. The human must be ready to intervene at any time, if only with the brake pedal.
Optimists have just gotten way ahead of themselves on fantasies of self-driving. I get it, it would be nice to have self-driving cars, but get realistic, guys. The developers haven't even really started identifying the problems; it's going to take a while.
Reciprocity
Active Member
Level 5 does not require any human intervention and does not even require controls for a human. It does not however have to be perfect. If the system gets confused, the car needs to be able to pull over and turn on the hazards just like a person. This is for Level 4 and 5 BTW. Level 4 is very similar to level 5 with the main difference being that drivers need to be able to take over when the car warns them of an issue. This hand off is typically at least 20 seconds in advance. The other main difference between 4 and 5 is that a person needs to be in the car because they need to be able to takeover and no one has to be in the car for level 5. Tesla network requires level 5, but some form of network could be created without. For example, people could pickup cars at a lot and the car could drive them to their location. The car would then park and wait for another ride. At some point, the owner would need to get the car back home. Not ideal, but certainly an option. I would love to pickup a car at the airport and have it drive me around time on my business trip and then back to the airport. These passengers would need to be able to take over so they would need to have a valid drivers license and insurance as well some idea that they might need to take over.
I am curious what your thoughts are for level 4-5 are if it was geofenced to only urban/suburban areas, and interstates? If they could get that much it would be a huge percentage of the addressable market. I do agree that the car being able to figure out where to park seems like a thorny issue, that is very different software from the rest of the driving.
Already a techincally solved problem, 20 years ago, though most of them still have someone to "hit the brake pedal". I think Vancouver Skylink doesn't.
And the *reason* is that it's normal to totally control the railroad track environment in a way in which we do *not* control the road environment. A highly controlled road environment will be a relatively easy problem -- which is why it's going to be easy to have self-driving on expressways ("no pedestrians / no horses / no bicycles", as the sign says around here).
"Freeway entrance to freeway exit" could be done before the end of the year, it's the squirrelly "surface streets" which are hard.
I don't think it matters very much right now on level 4 versus level 5. A level 4 car that can handle 99.999% of highways, 99% of suburbs, and 95% of cities is good enough that for lots and lots of people and that's 100% coverage sufficient to be indistinguishable from level 5 almost all the time. The remaining percentage of time would be equivalent to humans that also fail in various circumstances. For others, it's 0% level 5 at the same time because of where they live and where they need the car to drive. The problem for automakers and ADAS component manufacturers that wait for level 5 to ship, or rely on essentially level 5 to ship is that the time factor of experience at level 4 may be quite significant. And data collection at level 4 is a huge differentiator. The test fleets of 3 (Apple) or 200 (GM Bolt) or whatever under < 1,000, even with a much more extensive LIDAR, radar and camera sensor suite isn't going to cut it. They simply won't encounter enough scenarios in enough various locations as real drivers driving around their own cars in all sorts of wacky places.
I think as we have hundreds of thousands to millions of level 4 cars out there and the areas that give ADAS systems trouble would be under pressure to change. As it stands, traffic engineers alter our road scapes all the time, and wherever ADAS systems are geo-fenced away from or where safety is compromised, likely the complaints from drivers/riders will force change. And the most problematic of these locations likely are also the most problematic for human drivers.
I think as we have hundreds of thousands to millions of level 4 cars out there and the areas that give ADAS systems trouble would be under pressure to change. As it stands, traffic engineers alter our road scapes all the time, and wherever ADAS systems are geo-fenced away from or where safety is compromised, likely the complaints from drivers/riders will force change. And the most problematic of these locations likely are also the most problematic for human drivers.
Last edited:
Reciprocity
Active Member
This is one of the reasons that Tesla is so focused on vision systems over Lidar. Lidar can only tell you there is an object at certain distance, it can not definitively tell you what that object is. With Vision systems, you could have the car react to police hand gestured instructions. All you need is a million pictures of policemen giving hand singles. Again, this will come from the current data gathering that Tesla is doing. Not only that, but how the driver reacted. My point is that if it is possible, Tesla is using a methodology that can answer any situation because its the same way you and I do it. We see it, we interpret what is going on based on our experiences and we react. I would suggest that the system will do a much better job then a fresh teen driver sooner then you think.
Reciprocity
Active Member
Reciprocity
Active Member
Quite frankly there is not a trillion dollars on the line for autonomous trains. Really no comparison to the lengths companies will go to, to get that money.
racer26
Active Member
We've also seen in previous cases where there were expiring incentives that in essence what happens is the expiring incentive pulls forward demand, and you get a temporary sales boost much larger than the organic growth rate, followed by a glut afterward where the demand was pulled forward from (because now the product got substantially more expensive overnight), and then a return to the general trend formed by the organic growth rate.
Really big incentives will always artificially inflate demand. Big incentives expiring even more so.
HK's incentive was truly massive - their congestion problems are so large that they solve them by taxing cars HUGELY. EV's used to be exempt from that tax, and that made a Model S cost its owner approximately the same as a Camry in dollars out of their pocket. Of course that's going to make the Model S an abnormally popular car. Now that the exemption has expired, Model S has returned to costing its owner 2-2.5x what a Camry does (as it is in the rest of the world), and that has changed the popularity curve dramatically.
I am cracking myself up thinking about car jackers dressed as policemen flagging down self driving cars, and directing them to drive into waiting semi trucks.
We will be nowhere close to self-driving in 99% of suburbs or cities for many years. Highways are easy. Suburbs and cities have totally whacked-out road designs, all the time, and they'll have to go through them one by one. The most heavily populated places with the most Teslas will probably have local self-driving first (so, the quirks of the Bay Area will probably be worked out long before the quirks of rural NY).
Do NOT expect the roadways to change to make self-driving easier. Why? Because the way the roadways are now -- the unpredictable, hairy environment -- is largely down to cheapness. If people ask to make the roadways better for self-driving cars, the answer will be "are you gonna pay for it?" Maybe Mountain View and Palo Alto will. Flint won't.
Do NOT expect the roadways to change to make self-driving easier. Why? Because the way the roadways are now -- the unpredictable, hairy environment -- is largely down to cheapness. If people ask to make the roadways better for self-driving cars, the answer will be "are you gonna pay for it?" Maybe Mountain View and Palo Alto will. Flint won't.
Reciprocity
Active Member
That could happen with human drivers, but that's exactly what I thought when I wrote it. Car would have issues with driving into the back of a semi autonomously as well. But you also won't be able to Jack one with the finger in your pocket trick either.
erthquake
Active Member
This is why we have to "wait and see" on the value of the Tesla Network and other shared mobility platforms. I don't think Adam Jonas is factoring this complexity into his autonomous ride-sharing valuations/timelines.
In the near term, the most important marketing value of (nearly) full autonomous driving is the safety of passengers as well as other cars. If Tesla can demonstratively (and quantitatively) show that Autopilot saves more lives, regardless of how "perfect" it is, it's going to dramatically increase their technological moat.
Thanks for the well reasoned response. From an investor perspective I think that if they could get it to work in the 1% of the landmass of the world (The richest 1% where most teslas are sold) , plus the interstates, that would sell more cars than they can possibly produce for decades to come. from an investment perspective, who cares if it doesn't work in Flint? Would you want to send your new Tesla out by itself in Flint to pick up strangers? Of coarse if you look at it from a social justice perspective, it would be nice if it worked everywhere.
The one area that I could imagine roads being improved for self driving that would not cost very much is in striping. I would imagine that there are small changes to the way they stripe roads, that wouldn't cost any more than what they already do, that could make things much easier for self driving cars. Unfortunately, by the time they figured it out and implemented it, the software will probably leapfrog the need for it.
Reciprocity
Active Member
If your concern is roads, then you shouldn't be concerned. High def 3D map tiles and drivable paths solves the very efficiently and effectively. I'm more concerned with things like taxi cab drivers and people texting on their cell phone while crossing an intersection. How the car will react when someone crosses head on into it's lane because they are texting. Railroad crossings is for sure an issue, but easily enough solved. Isn't there a system to track trains?
Level 5 is no easy task but the rewards could never be bigger and getting larger every day with more distracted drivers. EV + autonomous is an end game for ICE that might take much longer without autonomous. But stakes.
Edit: We are all not just dreaming up FSD, Tesla is telling us that it will deliver it. Elon has been prone to exaggeration but I think with something like this that he would be more careful as it would and will damage their reputation and credibility more that just about anything besides missing on the model 3 production significantly.
Last edited:
- Status
Similar threads
- Sticky
