-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Things even FSD won't do
- Thread starter BridgeMojo
- Start date
CarlK
Active Member
I would agree if you said do not do instead won't do. Nothing is impossible long as it's not against the physics.
Then we will not have anything forever. An autonomous vehicle that has to rely on V2X to work will not be put on road unless all other vehicles and infrastructures are equipped for that. A simple chicken and egg problem.
Last edited:
Point conceded.
I too can see much of this happening.I think the most difficult hurdle for people to overcome is this: A self-driving car will never drive exactly the way they would, be it choice of route, following distance, whatever. It might drive a lot safer than you or I do, but it won't be the same. It will take time to develop confidence. Having misplaced confidence in automation too soon is dangerous, but lack of confidence when it's warranted could be dangerous too.
Uncle Paul
Well-Known Member
The most amazing form of FSD will happen when most vehicles are autonomous and can communicate with each other.
It will be remarkable when they not only know where each other are in physical space, but also what the cars around you are intending to do, and what their capabilities are.
Traffic will flow as one animal. Will travel in tight formation and make room for the ones that need to move to another lane to do so.
A car slowing down far ahead will signal to all the others that they also should prepare to slow down.
If a car wants to move over a couple lanes, other vehicles will expect that and give it room to manouver.
It will be remarkable when they not only know where each other are in physical space, but also what the cars around you are intending to do, and what their capabilities are.
Traffic will flow as one animal. Will travel in tight formation and make room for the ones that need to move to another lane to do so.
A car slowing down far ahead will signal to all the others that they also should prepare to slow down.
If a car wants to move over a couple lanes, other vehicles will expect that and give it room to manouver.
Mader Levap
Member
If you think it is realistic proposition, you are mistaken. Something like "everything moves onto v2v at same time overnight" will never happen. There will be decades-long period where autonomous vehicles and human-driven vehicles coexist on road. They will have to deal with each other. Period.
If you think it is realistic proposition, you are mistaken. Your autonomous car cannot assume "all vehicles/pedestrians/bicyclists" have transponders - THAT would lower safety drastically.
If you think it is realistic proposition, you are mistaken (Hmmm... I start to see pattern here). Do I even HAVE to explain idiocy (I can't describe that idea with less drastic word) of that one? Ever heard about something called "lawsuit"? Do you really think deliberately programming autonomous cars to hit people will end in anything else than "bankruptcy-inducing fines to carmaker, CEO in jail"? Even in "business-friendly" country like USA this kind of thing wouldn't go over well.
Did I say everything moves onto v2x at the same time overnight? No, I did not. I was referring to autonomous cars needed to have V2x from the start. For infrastructure it doesn't necessarily have to have V2X capability itself. But, it should be in an online database where information on it is kept up to date (lane closures, construction, road condition (ice, potholes, etc). The more about the environment that an autonomous car can know the better.
Did I say autonomous cars would assume everything had transponders? No, I did not. I advocate them as a way to add redundancy. There will still be lots of cases of things the vehicle didn't anticipate, and it will need to have ways of detecting them (animals, things blown onto the road, etc). With that being said some high speed roads could potentially require V2X communication of every vehicle traveling on it. This would likely only be autonomous only lanes on large freeways where the autonomous cars could travel close together like a train.
I'm simply of the belief that the best way to get to autonomous driving is through regulatory requirements. Where the requirements covered what the cars needed to have (v2x communications, sensors, level of computer redundancy, testing required, etc), and the roads themselves were approved for autonomous driving (a white list). Where that white list was constantly changing depending on road conditions, and construction.
The threshold for safety when it comes to autonomous cars has to be significantly better than human drivers or humans won't accept them. This is why we need a lot of redundancy.
As to solving the problem of humans bulling autonomous cars? My suggestion was clearly an exaggerated example, but how do you think autonomous cars should handle bullies? Ticketing the offender might be the only realistic way.
Last edited:
I wouldn't say V2X is necessary for an autonomous car to work, but by not having it we lose out on a huge benefit it offers autonomous cars.
To me it's only a question of how to implement it. There are already V2X vehicles on the road, and there are already things like intelligent traffic lights that will tell a properly equipped car what speed to go in order hit all greens.
I don't believe that regulatory requirements on V2X for vehicles/roads is anymore of a road block than what currently exist.
It's 2019, and we still don't have autonomous vehicles of any real kind.
Germany still hasn't approved an Audi A8 that has a very limited L3 style system.
So we're already hitting that "forever" as we try to implement it without V2X requirements.
@S4WRXTTCS and @Mader Levap I do find it interesting that in greentheone's development videos of autopilot in action, many of the vehicles are labeled with "No Rad sig." So the software is already listening for comm.
Oh! Maybe it does. Here I was thinking "no radiation signature," but that makes more sense. "Never mind."
CarlK
Active Member
What I meant is Tesla, or anyone else, can not design a system that relies on V2V to work unless every car on the road can do the same. A system like that will fail if just one car you meet at the intersection can not communicate with you. A workable system in the near future has to be able to do everything a human driver can do.
Yeah, of course. V2X is simply a way to try to fully utilize what an autonomous car can do as it can drive, and communicate anything out of the ordinary on the road (debris, potholes, stalled vehicles, etc). Essentially a seriously enhanced automated version of waze that benefited any vehicle equipped with V2X or a navigation system like Waze (assuming it gets connected to the V2X system).
V2X would also be used to communicate "agreements" with other V2X automated vehicles. This would significantly enhance the driving experience.
With V2X it hits the first five items on the OP's list (those are mostly about the autonomous system having up to date info on the road). Sure there are other ways to do those things, but V2X is a good candidate when it involves autonomous cars talking to each other or a human driven V2X vehicle.
Now you might say it's highly dependent on market saturation of V2X equipped vehicles, but we'd already have a lot if Tesla bothered to implement a V2X system.
CarlK
Active Member
Again "a lot" won't do. You need ALL cars on the road be able to communicate with you if your system relies on that to work. You can't miss even one car. It could help special situations like emergency vehicles or on limited access lanes for public transit for example but for general autonomous cars you just can't rely on that to work.
boonedocks
MS LR Blk/Blk 19”
Modern Lidar definitely works reliably in the level of rain or snow where vision is operable. Filtering the rain or snow out is done all the time.
Then again as things stand now even a slight sheet of wet sleet on the front bumper kills radar on a new Tesla and with it the whole Autopilot.
StealthP3D
Well-Known Member
Where "vision is operable" includes all conditions that doesn't bring all traffic to a complete standstill. Where current LIDAR systems are operable are a small sub-set of that. The problem with LIDAR is the light must make a round trip through the inclement conditions. You can't really "filter the rain or snow out", that's preposterous. At best, you can process the degraded data and try to work with what you have. But a vision system has the important advantage of only needing the light to travel one way (and it's all humans use to drive in inclement conditions).
True, Tesla made a cost/efficiency decision not to include a heated spot on the front bumper which makes sense when you consider that EAP is not rated for freezing conditions yet. It's trivial to upgrade that when the time comes but heating will not fix LIDAR's problem with fog, rain and snow.
StealthP3D
Well-Known Member
Call it whatever you want, it doesn't solve the inclement weather problem LIDAR has.
I'm surprised they even call that "snow". Notice that the visual image has no problem making out objects 1/2 mile away while the LIDAR image requires extensive "filtering" of false positives just to see 100 yards ahead! In normal snow you would have to filter the solid objects out with the false positives. So that's just not a solution for LIDAR in inclement weather. Vision has a big advantage as the video you posted makes clear.
Last edited:
It is not what I call it, it is what scientists commonly in the field call it — rain and snow definitely can be filtered from Lidar feed and this is being done every day. I don’t see why you would dispute a common technique.
Similar threads
