Elon: "Feature complete for full self driving this year"

[linux-works]

mething there (stationary vehicle under an overhead sign).

I think this will also be the case with HD maps. You can use them as guidelines. If the road has 3 lanes, but covered with snow, people still drive in these invisible lanes because they've driven there before (human HD map / familiar). The AV could not know this without that map. But you cannot rely on them for safety, or override drivable areas or emergency braking ,because the HD map says its safe. Perception needs to be the main rule.

if everyone had a communications channel to talk to each other, this would be an extra dimension of information. if you can authenticate (big if), then you can rely on shared experience and even very very current information (like, from the 4 cars that are in front of you).

it may be that sorting out all the information dimensions (sensors, static info, dynamic info, learned/shared info) becomes the new hard problem.

 

[ChrML]

if everyone had a communications channel to talk to each other, this would be an extra dimension of information. if you can authenticate (big if), then you can rely on shared experience and even very very current information (like, from the 4 cars that are in front of you).

it may be that sorting out all the information dimensions (sensors, static info, dynamic info, learned/shared info) becomes the new hard problem.

Yeah! It won't be possible until at least a few companies have developed self-driving, and someone decides a standard for information sharing should be built.

But it will definitely happen. And will increase safety and drivable speed when you can see around corners etc... It'll be a 5th dimension of info.

I don't think sorting the info will be a big problem. It'll be a observational landscape and known decisions (like turn indicators) in a standardized format. Appended to internal environment with a slightly lower trust value than own observations.

 

[linux-works]

v2x is such a standard, at least in an early phase. I don't know of any others.

 

[emmz0r]

if everyone had a communications channel to talk to each other, this would be an extra dimension of information. if you can authenticate (big if), then you can rely on shared experience and even very very current information (like, from the 4 cars that are in front of you).

it may be that sorting out all the information dimensions (sensors, static info, dynamic info, learned/shared info) becomes the new hard problem.

won't happen. Also a security risk, and has minor advantages. If humans can manage to drive around, computers eventually will. Only advantage I can think of is negotiating who drives in unregulated intersections where humans traditionally would use gestures and eye contact. But won't happen since literally 100% of the cars would need to have the same system and same communications protocol and working software.

 

[arghx7]

Vehicle to vehicle communication would take decades. They would have to force every car off the road that doesn't have it, or wait for the natural turn over of the fleet. There are probably millions of cars on the road that don't have antilock brakes/stability control, and that has been required since something like 2012 model year in the US. Only now are carburetors effectively off the road in most places and the last carburated car was I think 1990 model year.

 

[linux-works]

I can tell you that there are vendors with early silicon that supports v2x and there are vendor test suites and software packages you can buy to integrate with.

all it takes is enough cars using it, and enough people seeing the benefit, for it to really take off.

I don't think its that many years away. in fact, I have a feeling that its 2 years away on some production cars, give or take a year. but way way less than 5 years, pretty sure...

no idea about tesla, though. they don't use industry 'standard' software from the regular vendors (iirc).

 

[CrayZ1]

Cadillac has been shipping cars with V2V hardware since 2017, and all cars with V2V & V2X since 2018. Once V2V gets about 25% saturation AND mfr cooperation, it will start to work. It will essentially expand your sensor range and safety of driver's aids. Classic example would be somebody about to blow through a red light. Your car approaching the intersection might not see it, but it could receive data that report the speed and direction of the scofflaw. Or a motorcycle splitting lanes in heavy traffic is another.

In a fantasy world with 100% saturation of V2V, you don't need lane markers, speed limits, red lights, stop signs, etc. Those exist to keep cars from hitting each other.

What I find disturbing, is that a 12 mph accident with a minor injury involving a fully autonomous car (a motorcycle sideswiped an AV, MC at fault, AV broke no traffic laws), made international news. Hundreds of articles about it. But did you read anything in the MsM about the V2V rollout? It seems there are forces at play in black helicopters that are trying to stop Autonomous Technology. Oh, that MC minor crash, the AV MFR settled with the defendant even though not at fault.

 

[linux-works]

won't happen. Also a security risk, and has minor advantages.

minor? why so?

here's how I see it. some car ahead sees a road hazzard. they swerve to avoid hitting it. the car behind also see it and does that. maybe only 2 of the 5 cars ahead of you have v2x installed; that's still 2 bits of info, IN ADVANCE (before YOU hit the object) that you car can use to make decisions on.

for queuing, it can be helpful. you can slow everyone down gradually, when there is an accident or obstruction.

and yes, authentication is needed. we've had security and authentication models that work; we also have a lot of amateurs who make products and have no clue about security. hopefully, the iot guys who know nothing about security (and could not care less) are NOT the same ones who end up building the next gen v2x systems

here's a direct analogy: the cars up ahead slow down and they put on their brake lights. that is an early form of v2x (in a sense, if you think of it that way). its an indication from one 'car' to another about what it plans to do. same with turn signals. same with horn. lots of ways for one car to 'tell' the others. so why should a packet radio be any different?

 

[Daniel in SD]

I'm very skeptical that V2X is worth the cost in the short term. Long term I could see it improving traffic flow. Maybe we could replace carpool lanes with autonomous lanes in the near term?
When I watch my favorite self driving demo video I wonder where V2X would help. Would every person have a brain implant that would transmit their intentions?

What I find disturbing, is that a 12 mph accident with a minor injury involving a fully autonomous car (a motorcycle sideswiped an AV, MC at fault, AV broke no traffic laws), made international news. Hundreds of articles about it. But did you read anything in the MsM about the V2V rollout? It seems there are forces at play in black helicopters that are trying to stop Autonomous Technology. Oh, that MC minor crash, the AV MFR settled with the defendant even though not at fault.

The concern trolling on this subject is absurd. Everyone in power wants self driving cars. $80+ billion dollars have been invested, big business is for it, the government is for it, there is literally no organized opposition that I'm aware of.
U.S. to outline strong support for self-driving cars at CES
The first pedestrian killed by a car back in 1896 was also well reported. Unusual events often make the news!
The only information we have about who was actually at fault in the Cruise accident is from the police report. It seems quite possible to me that if a human were driving the accident would not have occurred.

 

[CrayZ1]

I'm very skeptical that V2X is worth the cost in the short term. Long term I could see it improving traffic flow. Maybe we could replace carpool lanes with autonomous lanes in the near term?
When I watch my favorite self driving demo video I wonder where V2X would help. Would every person have a brain implant that would transmit their intentions?

The concern trolling on this subject is absurd. Everyone in power wants self driving cars. $80+ billion dollars have been invested, big business is for it, the government is for it, there is literally no organized opposition that I'm aware of.
U.S. to outline strong support for self-driving cars at CES
The first pedestrian killed by a car back in 1896 was also well reported. Unusual events often make the news!
The only information we have about who was actually at fault in the Cruise accident is from the police report. It seems quite possible to me that if a human were driving the accident would not have occurred.

I read the accident report back then. The Bolt wanted to change lanes, but never left it's lane. A small motorcycle who was splitting lanes assumed the Bolt would complete the change, but it could not. The bike was going faster than traffic, and swerved into the lane, hitting the side of the Bolt and falling down at 17 mph. Getting cut off while making a lane change is very, very common in congestion. People try to stop you. A human does the thing when somebody tries to block them from a lane change, they get back to the center of the lane to try again.

I split lanes for 150 miles a day for about 6 years straight and never hit a car. Luck? That and experience. I only split the 1-2 lane (the bike was in 2-3 split), I do not enter a lane with a car in it. I do not assume cars will drive like I want them to.

Minor crashes happen all the time, and they are seldom $75,000 in damages.

 

[DanCar]

I don’t think tesla has that big of a head start on full self driving as we think. I seen kia sonata has summon now.... and quite a few cars offer steering assist with adaptive crews control.

Partial list of level 2 cars: Lane centering - Wikipedia

 

[Daniel in SD]

I read the accident report back then.

Did the police look at the video from the Cruise vehicle?

$75,000 in damages.

Cruise has taken billions of dollars in funding. If their vehicles get into fewer accidents than a human driver I can't see how liability could be a problem even if they are wrongly held liable for accidents where they are not at fault.

 

[emmz0r]

minor? why so?

here's how I see it. some car ahead sees a road hazzard. they swerve to avoid hitting it. the car behind also see it and does that. maybe only 2 of the 5 cars ahead of you have v2x installed; that's still 2 bits of info, IN ADVANCE (before YOU hit the object) that you car can use to make decisions on.

Devils advocate here.

You can judge that by visually looking at the situation as well. Also that is mission critical information that has to have 0 latency, and if the server is down, you're screwed. If you have direct car-to-car communication it means that it has to be super secure, and make sure 110% that you can't spoof the signal of your own car sending incorrect information. Causing cars to behave erratic and disrupt the whole thing.
In car-to-car communication you are actually sitting on the "server" in your own car, which can be tampered with.
I view the downside of that way bigger than the potential upside.

For traffic jams etc you can use the server, that is sort of in place at Tesla where it can choose a detour if there's too much traffic. Yes I know it's very "low level" / "basic" version of realtime traffic queue arrangement.

 

[im.thatoneguy]

That's not what L5 is. For one, buying tickets is not a DDT so it is not required for any autonomous car to do. Second, I don't believe L5 includes off road driving.

I'm not saying the buying of the ticket is required for L5, I'm saying the trip will require tickets. And if the ticket requires tickets then there will have to be adaptation on the part of the ferry terminal.

The transaction is not only a ticket, at least for my local ferries, there is also a verbal exchange on where to go and what to do when you buy a ticket. If you just drove through an automatic ticket booth it would also need an adaptation to both query where you want to go from the vehicle/driver (probably through some sort of web API) and then it would need to respond by "proceed to loading zone 5".

If you're implementing an entire stack to communicate with AVs due to the ticketing requirements, then it stands to reason they won't stop at ticketing but will also automate the role of the human traffic managers and have an "AV loading time" in which they simply load all of the AVs through a wireless signal to the waiting autonomous cars. What's most likely is that every manufacturer will look at the cost of debugging ferry boarding and say "Don't care, the humans can drive for 5 seconds." and not bother implementing it at all until the entire process is entirely integrated with the ferry system.

 

[diplomat33]

I'm not saying the buying of the ticket is required for L5, I'm saying the trip will require tickets. And if the ticket requires tickets then there will have to be adaptation on the part of the ferry terminal.

Except that trip is not required for L5 either. That's ludicrous. L5 does not require that a car actually self-drive across continents including embarking and disembarking from a boat or plane autonomously. The SAE defines autonomous driving as performing the dynamic driving tasks to do a trip on roads. And L5 can be restricted to a single continent. For example, L5 can be restricted to just the continental US.

 

[im.thatoneguy]

Second, I don't believe L5 includes off road driving.

The definition is:

As specified in J3016, level 5 is distinguished from level 4 by the fact that it is not operationally limited to a specific operational design domain and can rather operate on-road anywhere that a typically skilled human driver can reasonably operate a conventional vehicle.

But their "on-road" definition is

“On-road” refers to publicly accessible roadways (including parking areas and private campuses that permit public access) that collectively serve users of vehicles of all classes and driving automation levels (including no driving automation), as well as motorcyclists, pedal cyclists, and pedestrians

The aforementioned 4x4 road that I drove a rental Kia through with planks for bridges and puddles large enough to consume a small rental vehicle was a "Highway" and mapped. So that qualifies as a publicly accessible roadway that a skilled human drive (me) could reasonably operate a conventional vehicle on.

 

[diplomat33]

The definition is:

"As specified in J3016, level 5 is distinguished from level 4 by the fact that it is not operationally limited to a specific operational design domain and can rather operate on-road anywhere that a typically skilled human driver can reasonably operate a conventional vehicle."

But their "on-road" definition is:

“On-road” refers to publicly accessible roadways (including parking areas and private campuses that permit public access) that collectively serve users of vehicles of all classes and driving automation levels (including no driving automation), as well as motorcyclists, pedal cyclists, and pedestrians"

The aforementioned 4x4 road that I drove a rental Kia through with planks for bridges and puddles large enough to consume a small rental vehicle was a "Highway" and mapped. So that qualifies as a publicly accessible roadway that a skilled human drive (me) could reasonably operate a conventional vehicle on.

Maybe your case would fall under L5 but that might be iffy IMO. "Publicly accessible roadway" means some type of road. The definition you quote even mentions parking areas and private campuses. That is what the SAE is talking about. I don't think the SAE is talking about some dirt path barely wide enough for a car and some bridge with wooden planks.

But when I say "off roading" I am talking about driving on terrain where there is no road or discernable lane at all, like riding across an empty corn field. That would most definitely not be L5.

 

[im.thatoneguy]

L5 can be restricted to just the continental US.

Not necessarily. It can only be restricted to the Continental US for the purposes of licenses, not operational limitations. They address this scenario specifically in section 8.8

For example, an ADS-equipped vehicle that is capable of operating a vehicle on all roads throughout the US, but, for legal or business reasons, cannot operate the vehicle across the border in Canada or Mexico can still be considered level 5, even if geo-fenced to operate only within the US. The rationale for this exception is that the geo-fenced limitation (i.e., US, only) is not due to limitations on the technological capability of the ADS, but rather is due to legal or business constraints, such as legal restrictions in Canada and Mexico/Central America that prohibit level 5 deployment, or the inability to make a business case for expansion to those markets.

They offer an "Out" for Level 5 which is "there is no business case to ..." which essentially makes a further mockery of Level 5. You can call any limitation on Level 5 a restriction due to "the inability to make a business case".... in which case you can take nearly any level 4 system and say "There isn't a business case to expand to back roads" "There isn't a business case to expand to ferry terminals" "There isn't a business case to expand beyond cities without a million people..." "There isn't a business case to expand to heavy snow." etc.

They should have stopped at there are legal restrictions and called it good. But then they knew someone would say "But what about driving on the island of Java and recognizing its road signs? Isn't that a geo fence?" And they just said "Ok fine if you don't want to sell a vehicle which can drive in Java then you can call it Level 5 even though it's geofenced." So you would have to implement Javanese capability while not ever actually selling in Java due to business constraints. How do you prove your system can operate in Java if you don't test in Java due to business constraints? You can't. So people will either lie and say there are no geo restrictions technologically or they will just accept L4. Because L4 is fine.

 

[hill]

minor? why so?

here's how I see it. some car ahead sees a road hazzard. they swerve to avoid hitting it. ......snip....

uh huh
on ice, & so now you're plowing through the guardrail, into the river, hoping you can break the glass to swim 10 feet up.
Thank you Full Self Driving

.

 

[diplomat33]

Not necessarily. It can only be restricted to the Continental US for the purposes of licenses, not operational limitations. They address this scenario specifically in section 8.8

I am well aware. What I am saying is that is preposterous to say a car is not L5 if it can't autonomously buy tickets and embark and disembark on a plane or boat to cross continents. Buying a ticket is a not a DDT so it is not included as part of autonomous driving.

The SAE is very clear that L5 means performing the entire DDT on public accessible roadways. So getting on planes or boats is not part of L5.