Autonomous Car Progress

[Jeff N]

Can't compare what you see on YT with what you experience. Afterall we all thought FSD was smoother after watching YT until we got it ourselves.

That's why we need independent people who can ride and compare on routes not selected by the company.

From personal experience, Cruise and Waymo drive far more comfortably than FSD 11.4.2.

FSD has improved a lot in recent months but it still has a long way to go in achieving low g-force passenger comfort. Stopping and accelerating in the Cruise or Waymo is very floaty — everything is typically soft and smooth and with no strong accelerations through turns. FSD is still robotic. FSD makes less obnoxious acceleration and braking much of the time now compared to earlier versions but it is still sudden and precise like a step motor. Cruise and Waymo average-out their braking and acceleration into smooth and predictable acceleration/deceleration curves like a professional chauffeur.

Cruise and Waymo still make mistakes or become momentarily confused but they do it elegantly. Tesla knows how to do this. Tesla’s regen braking to a full stop is very elegant and smooth but FSD doesn’t appear to use it and seems to control braking itself (poorly). If they could find a way to offload some of the normal braking details to the regular regen firmware it would not only do a much better job but it would also offload some of the cpu overhead from the FSD computer.

 

[Jeff N]

But clearly this is happening quite frequently (90 calls to 911 about stuck robotaxis in last 3/4 months. That is about 3 daily). That tells me robotaxis are not ready for driverless and need a safety driver at present. That is what SF is also saying - but the robotaxi companies have "bought off" state regulators who are imposing their will on SF. If SF had a say these robotaxis would not be allowed without safety drivers on the road at present.

True. Remote intervention appears to happen regularly — too regularly. Maybe an average of once per trip. But not in a way that makes Cruise unsafe, just annoying for others and cost inefficient for them. I assume that their strategy is to spend money to make remote interventions happen quickly with the idea that over time the need for interventions will scale down as the software improves. In my experience, the level and severity of interventions happening today is okay. Human drivers regularly get confused or do dumb things as well. But, they need to continue improving the system rapidly to make scaling up cost-effective.

 

[EVNow]

From personal experience, Cruise and Waymo drive far more comfortably than FSD 11.4.2.

Have you tried FSD from the back seat like you have tried Cruise / Waymo ? When you are in the driver's seat it will always feel very jerky since you can feel every small twist of the wheel. You can see in CYBRLFT videos, for eg., that a lot of his customers don't even know he is using FSD.

True. Remote intervention appears to happen regularly — too regularly. Maybe an average of once per trip. But not in a way that makes Cruise unsafe, just annoying for others and cost inefficient for them. I assume that their strategy is to spend money to make remote interventions happen quickly with the idea that over time the need for interventions will scale down as the software improves. In my experience, the level and severity of interventions happening today is okay. Human drivers regularly get confused or do dumb things as well. But, they need to continue improving the system rapidly to make scaling up cost-effective.

SF authorities don't think so. Cruise and Waymo need to fully and properly implement the MRC feature. Currently all they do is just stop in the middle of the road and call it "minimal risk".

 

[stopcrazypp]

SF authorities don't think so. Cruise and Waymo need to fully and properly implement the MRC feature. Currently all they do is just stop in the middle of the road and call it "minimal risk".

Welp, the recent mass shooting had Cruise blocking EMS once again. Even though Cruise rushed to do damage control saying it didn't block all lanes and that the vehicles could go around it (unlike previous incidents where they were bunched up or there was other traffic), I hardly think SFPD is convinced.
Stalled Self-Driving Car Appears To Block EMS From Reaching Mass Shooting Site

It looks like the police officer was trying to manually move the car, but was unable to. I believe SFMTA made that one of the suggestions, that if the companies were not going to have safety drivers, at least have a mechanism to allow emergency responders to manually move the vehicle off the road so emergency vehicles can get by. That call of course was ignored by the state and the companies.

Technically according to the SAE definitions, stopping in the middle of the road can still qualify as "minimal risk". Perhaps the standard needs to be made a little higher, especially in cases where obviously the car has plenty of space to pull over (as any human would have done if emergency vehicles were coming their way).

 

[EVNow]

Welp, the recent mass shooting had Cruise blocking EMS once again. Even though Cruise rushed to do damage control saying it didn't block all lanes and that the vehicles could go around it (unlike previous incidents where they were bunched up or there was other traffic), I hardly think SFPD is convinced.
Stalled Self-Driving Car Appears To Block EMS From Reaching Mass Shooting Site

This is beyond the pale. Unfortunately people like @diplomat33 and @Bladerskb will keep supporting these below-human capable robotaxis blocking the path of emergency responders saying "but Tesla". Horrible.

As I noted earlier - these robotaxis are "reducing" accidents by just stopping in the middle of the road when they are unsure. If someone else hits them they will claim it is not their fault.

Looks like these robotaxis may reduce fatalities in accidents but get people killed by other means.

It looks like the police officer was trying to manually move the car, but was unable to. I believe SFMTA made that one of the suggestions, that if the companies were not going to have safety drivers, at least have a mechanism to allow emergency responders to manually move the vehicle off the road so emergency vehicles can get by. That call of course was ignored by the state and the companies.

I don't think that is even enough - Robotaxis need to be able to respond and move to the side like human drivers. Otherwise they need safety drivers. There is zero reason why emergency responders need to move the cars ... they have an emergency to attend to, afterall.

 

[diplomat33]

This is beyond the pale. Unfortunately people like @diplomat33 and @Bladerskb will keep supporting these below-human capable robotaxis blocking the path of emergency responders saying "but Tesla". Horrible.

Stop your lies. I never defended Cruise in that incident. In fact, I've pointed out how bad Cruise is in the past. I agree that this Cruise stall that blocked police from a gunshot victim is beyond the pale. I do not defend it.

But don't lump Waymo and Cruise. It was a Cruise that blocked police from a gunshot victim, not Waymo. You cannot assume that Waymo is the same as Cruise. Waymo is far more reliable than Cruise. Waymo has far fewer stalls than Cruise and better safety. Don't act like they are the same. They are not.

I don't automatically defend Waymo. I acknowledge when Waymo needs to improve. But unlike you, I also acknowledge where Waymo is ahead in autonomous driving. The fact is that Waymo has demonstrated advanced autonomous driving that is very safe and reliable in many ODD, but not perfect.

And I've never said that FSD beta is horrible. I've pointed out where it is good and where it is not good. But the fact is that if FSD beta were driverless, it would be far worse than Waymo. It would have far more stalls and it would actually crash into other vehicles and cause injuries. Waymo is demonstrably far more reliable and far safer than FSD beta. Do I need to show you videos of all the times FSD beta is super unsafe and makes mistakes that Waymo never does?

 

[stopcrazypp]

I don't think that is even enough - Robotaxis need to be able to respond and move to the side like human drivers. Otherwise they need safety drivers. There is zero reason why emergency responders need to move the cars ... they have an emergency to attend to, afterall.

They shouldn't have to, but it's far better than the alternative of them just halting in the middle of the road and needing 15-20 minutes for someone from the company to come rescue it. This is especially true if they were bunched together. Just the ability of being able to move one of them would clear a path to pass. Then the company can deal with the rest of the mess.

 

[diplomat33]

As I noted earlier - these robotaxis are "reducing" accidents by just stopping in the middle of the road when they are unsure. If someone else hits them they will claim it is not their fault.

That is a very cynical take and completely wrong. They are reducing accidents by driving more cautiously than humans, by not speeding, not driving drunk, not losing control of the vehicle, not running a red light etc... That is where they are saving the most lives. In fact, the number of stalls per millions miles is so small that the effect on reducing accidents would be negligeable.

Looks like these robotaxis may reduce fatalities in accidents but get people killed by other means.

Look, the Cruise stall is beyond the pale. But we do need some common sense and perspective. How many people have died because an AV was blocking emergency vehicles? 5 maybe. How many lives were saved because an AV was not speeding or not driving drunk? Probably hundreds. I would venture AVs are saving way more lives than they are killing. I am not saying the Cruise stall was ok. Cruise needs to fix these stalls before they scale any further. Blocking emergency vehicles is unacceptable. But let's not pretend like AVs are killing lots of people because they block emergency vehicles all the time, they do not.

I don't think that is even enough - Robotaxis need to be able to respond and move to the side like human drivers. Otherwise they need safety drivers. There is zero reason why emergency responders need to move the cars ... they have an emergency to attend to, afterall.

Waymo and Cruise do move to the side like human drivers. And in most cases, they do. There are just some situations where they don't yet. Autonomous driving is a work in progress.

 

[EVNow]

They shouldn't have to, but it's far better than the alternative of them just halting in the middle of the road

Alternative is to have safety drivers until the massive companies can figure out how to do this properly. Citizens and emergency personnel shouldn't be asked to fill in for free.

 

[scottf200]

I installed 11.4.3 (2023.7.15) last night. On a drive with my wife today it attempted to pull out in front of a 45 MPH moving car that would have hit or had to swerve around us. I slammed on the brakes when I realized it was "going for it". It did stop at the stop sign ... "creep forward for visibility" (which was clear and long) before it started to go. Zero excuses for this behavior IMO.

A short while later it *slowly* "creep forward for visibility" in a couple spots that were wide open viewing for 30+ seconds as it approached the T intersection. It really is absurd.

I really wonder how much of this is its limited 90 (right) and 180 (left) degree visibility angles with modest cameras.
Comparison: Camera POV - Mobileyes SuperVision requirements on their partner Zeekr

 

[JHCCAZ]

I really wonder how much of this is its limited 90 (right) and 180 (left) degree visibility angles with modest cameras.

With the 2023.20.x software (which FSDb testers probably won't get for a while), it looks like they've enabled viewing for all of the cameras, including the formerly unviewable B-pillar cameras. This is documented to be true for Sentry mode, but I assume for dashcam recording/playback also.

When that happens, it will enable us to save dash lcam clips of these events, so that we can try to understand what the car thought it was seeing when it feels like it's about to commit to a turn right in front of oncoming traffic.

This happens to me sometimes also. I think at least some of these are due to a very clumsy "creep" adjustment, with enough jerk that it feels just like it's initiating the turn. It obviously knows how to creep slowly and gently, because we experience this all the time that stop signs as you mentioned. But when it lunges forward, you can't wait around to see if it's really going to throw you in front of the oncoming car - you just have to slam the brakes.

(BTW, your FSD experience message and my reply probably belong in the FSD 11.x thread. This thread is more for tracking and discussion of industry-wide autonomous car developments.)

 

[Doggydogworld]

But we do need some common sense and perspective. How many people have died because an AV was blocking emergency vehicles? 5 maybe. How many lives were saved because an AV was not speeding or not driving drunk? Probably hundreds.

The answer to both questions is zero.

 

[diplomat33]

@EVNow The Cruise did not block emergency vehicles according to the police and fire department. Also 9 were wounded but nobody died. So, the Cruise did not cause anyone to die because of blocking emergency vehicles.

But it turns out the initial perception created by the video may have been wrong, according to the San Francisco police and fire departments. They now say the robotaxi didn’t block emergency responders or otherwise get in anyone’s way after the shooting, which wounded nine people.

“The autonomous vehicle did not delay police, fire, or other emergency personnel with our arrival or departure from this scene. Furthermore, it did not interfere with our investigation into the shooting incident,” the police department said in an email.

The fire department said in a separate email that the vehicle “did not delay” fire personnel or paramedics. It added that the episode “could have been catastrophic but we are lucky that there was another lane we could use.”

That version of events is consistent with an earlier statement from Cruise, which is majority-owned by General Motors.

“Our car initially stopped as it was approaching an active emergency scene, then proceeded to perform a U-turn and pull over. Throughout this time, all vehicles, including emergency response vehicles, were able to proceed around our car,” Cruise said in a series of tweets early Saturday, hours after the shooting.

A spokesperson for Cruise said Wednesday that a further review of data from its vehicle again showed that it didn’t block any emergency personnel.

The viral video was only 13 seconds, and it didn’t show a wide sweep of the area.

A viral video of a 'reckless' robotaxi caused an uproar in San Francisco. Police say the internet got it wrong.

Online, people claimed the self-driving vehicle blocked emergency responders. Police say it isn't true.

www.nbcnews.com

 

[willow_hiller]

@EVNow The Cruise did not block emergency vehicles according to the police and fire department. Also 9 were wounded but nobody died. So, the Cruise did not cause anyone to die because of blocking emergency vehicles.

A viral video of a 'reckless' robotaxi caused an uproar in San Francisco. Police say the internet got it wrong.

Online, people claimed the self-driving vehicle blocked emergency responders. Police say it isn't true.

www.nbcnews.com

Ultimately, the questions I ask are:

1. Is Cruise in a fit state to be testing without a full-time safety driver?

2. Do they gain anything other than publicity for testing without a safety driver?

In my opinion, the answer to both is no. Removing the safety driver at this point is an unnecessary risk they're taking for the sake of good PR.

 

[diplomat33]

Ultimately, the questions I ask are:

1. Is Cruise in a fit state to be testing without a full-time safety driver?

2. Do they gain anything other than publicity for testing without a safety driver?

In my opinion, the answer to both is no. Removing the safety driver at this point is an unnecessary risk they're taking for the sake of good PR.

I also lean towards "no" for both. I definitely think Cruise jumped the gun with deploying public driverless at night in SF just to score a PR "first". And they launched driverless at night in Austin and Phoenix, also at night, in a puny geofence, also just for the PR "first" of claiming that they could scale to 2 cities in just 90 days.

 

[Bladerskb]

Ultimately, the questions I ask are:

1. Is Cruise in a fit state to be testing without a full-time safety driver?

2. Do they gain anything other than publicity for testing without a safety driver?

In my opinion, the answer to both is no. Removing the safety driver at this point is an unnecessary risk they're taking for the sake of good PR.

I voted Yes to both questions. If Tesla were to have done something similar and released something that was 1,000 times worse. You Tesla fans would be claiming it is the best thing since sliced bread and any delay would be killing people.

Oh wait they did and you guys responded exactly like that. They released one of the WORST and DANGEROUS software in FSD Beta when it clearly shouldn't be released and you guys praised and defended it. All you have to do is watch the old videos to see how horrific the software was at the times.

(You can make the case that version 10.69 or 11 should have been the version that is released)

I have watched hundreds of videos from Xpeng's CNGP and XNGP and Huawei's ADS (both door to door system).
The state these systems were released in were A planet better than the state FSD Beta was released in.
The same thing was the case with the release of NOA. It was horrific and you guys praised and defended it.
Other systems equivalent at release were no where near that state.

It demonstrates that Tesla is willing to release software that are not Ready.
What makes you think they won't continue this trend when they finally do robo-taxi?

 

[diplomat33]

New software update by Cruise:

The latest software release has rolled out to the driverless fleet, containing improvements in these areas:

• Shipped TSEL v15.2 with multiple improvements:
  • Improved driving behavior by 20% for higher speed roads, particularly during right turns and in intersections, yielding and asserting smoothly at speed.
  • Improved driving around double parked vehicles by 12.5% and turns onto arterial roads by 22%.
  • Increased reliability when pulling out of traffic and back into traffic to pick up and drop off riders.
  • Improved overall steering comfort by 5%.
• Shipped DLA v4, and KSE v14 which combined with LPC v7 further refines the detection and tracking of debris and animals by up to 50%, enabling the AV to more safely and smoothly drive around these objects.
• Shipped TREX v3 to improve lateral positioning by ~2X*, making AV driving more human-like.
• Further reduced trip interruptions through improved software stability, fault tolerance and additional fault recovery capabilities.

https://getcruise.com/news/blog/2023/may-04-software-release/

 

[willow_hiller]

I voted Yes to both questions. If Tesla were to have

This is called a "Whataboutism" and is a logical fallacy. Nothing that Tesla does changes the fact that Cruise could continue benefiting from having a safety driver.

 

[diplomat33]

I voted Yes to both questions. If Tesla were to have done something similar and released something that was 1,000 times worse. You Tesla fans would be claiming it is the best thing since sliced bread and any delay would be killing people.

Oh wait they did and you guys responded exactly like that. They released one of the WORST and DANGEROUS software in FSD Beta when it was clearly shouldn't be released and you guys praised and defended it. All you have to do is watch the old videos to see how horrific the software was at the times.

I have watched hundreds of videos from Xpeng's CNGP and XNGP and Huawei's ADS (both door to door system).
The state these systems were released in were A planet better than the state FSD Beta was released in.
The same thing was the case with the release of NOA. It was horrific and you guys praised and defended it.
Other systems equivalent at release were no where near that state.

It demonstrates that Tesla is willing to release software that are not Ready.
What makes you think they won't continue this trend when they finally do robo-taxi?

There is definitely a huge double standard with certain Tesla fans who don't want FSD Beta to be banned when it was incredibly unsafe but are now calling for Cruise and Waymo to be banned over a few "stalls" (and the Cruise stall was not blocking emergency vehicles as they claimed).

But your post does not answer @willow_hiller 's question. You say you would answer "yes" to both questions but you don't give your reasons. I'd be curious to hear your answer for why you voted "yes".

 

[Doggydogworld]

2. Do they gain anything other than publicity for testing without a safety driver?

At some point the "safety" driver reduces safety. A few years ago I went through every reported accident involving Waymo over a 3-6 month period. The only ones where Waymo was arguably at fault were caused by safety drivers. Sometimes the car was in manual mode the whole way. Sometime the safety driver intervened and caused a wreck. For example, one car veered to the left in slowish, multi-lane traffic. The safety driver intervened, steering the car back to the right and directly into the path of a lane-splitting motorcyclist who hit the right rear corner bumper.

Most memorable was the safety driver who accidentally disengaged the autonomous system while asleep on the highway. Fortunately he just ended up in the ditch and walked away.

The same thing happened when Waymo first moved into downtown SF. One driver felt the car was waiting too long at a green light, intervened and hit a skateboarder. All the "injury accidents" SF Transit called out in their protest letter to CPUC involved safety drivers.

Is Cruise now at the point where pulling safety drivers actually improves safety? I don't know, but it's possible. Safety drivers could certainly resolve stalls faster, which would be more convenient for others.

Maybe we should rename them convenience drivers. Or PR improvement drivers.