Waymo’s awesome remote fleet response system is the only way they can have so many cars constantly operating fully autonomously without any safety driver on board. It’s clear and evident in almost everything that they do that they obviously understand the importance of reputation and safety record and how crucial it is for public trust in adoption of this new technology. The engineers they got from Anki are wicked smart too. (RIP to one of the coolest little robotics manufacturers) It’s been awesome to watch them grow over the years. I am hoping they will soon allow them to travel on highways as well. I can’t wait to see what the next generation Waymo’s will be like as well, since they’ve been using the jag ipace platform for about 4 years now. Especially with the recent big leaps everyone has been making in their NPUs and CPUs, Honestly did a double take seeing it was you covering this. But really glad that you are.
Wow, this is actually huge. Even though it requires a human 'driver', I suspect 1 could cover maybe 100 vehicles or so. And that'll increase as the software gets better. Robotaxis really are nearly here. And after seeing this, I think Waymo and Tesla will be leading the way. Exciting times
The Waymo visualisations are massively detailed and accurate. I wonder if they are combining the Video Detail with the Computer analysis to augment the details. Such as the exact type of vehicle / shape variation, Pedestrian’s Shopping Bags, UPS Truck with detailing of UPS on the side…. Great job Waymo !!!
Yup, but totally not important. Tesla replaces objects with ugly default 3d models while Waymo spends process power to let their 3d rendering look the same as in real world. No effect on self driving capabilities though, just visualisation.
@@kafiluz4317 Yeah, but if it's the interface for the operator, I'm sure an operator would prefer to see some actual humans doing human things, not just dummies "walking" around.
@@kafiluz4317 This is a video about how Waymo remote operators work. I'd like FSD to not lean left or right in the lane (or just accept my corrections without disengaging), but that's a topic for another video.
This is very interesting. Impressive stuff from Waymo here. Likely though, they've just paved the way for Tesla to implement their own flavour of this remote intervention protocol.
It's great that Waymo published this and their approach is very pragmatic and smart. Unfortunately, they didn't publish a very important aspect: What is the remote operator intervention per X number of miles? That number essentially tells you how scalable their approach is. If they need 1 operator per 10 cars, that is never going to work on even a large scale. 1 operator per 1000 starts to get practical and at 1 operator per 10000, we are talking amazing.
There are about 5 million Uber drivers who drive more than occasionally. At a rate of one fleet response agent per 1000 Waymos, it would take 5000 agents to cover 5 million Waymos. Currently there is a total of a few hundred Waymo vehicles.
Great job explaining how Waymo is operating. I've used Waymo a number of times in SF, but haven't noticed an intervention. I'd love to try FSD, and I'm thinking of leasing a 3, but I need to sell my old '88 Porsche 911 first.
Even with some intervention demanded by the car. It could be safety critical (not like today, the car would be in control but hard breaking and asking for help), "just in case" (slowing down or staying in its place) or to improve learning in a specific situation that they lack data of (just a message asking to take over to get some data).
I like the dropping in a car to tell Waymo where to go to resume the journey. This video is interesting and informative. Are the remote operators human? I ask because it navigates so differently than FSD.
I wonder....Will this suggestion from the operation be count as an intervene? because it's look like intervene to me. However, I think we also have to find a better metrics to measure the readiness of self-driving car. Should count how much it's impact/inconvenience other people on the road as well, not just full blown accident.
Waymo vehicles are operating on the freeways for road-testing until regulatory approvals. You can find some of the videos of freeway driving. The Waymo technology is capable, just not approved by each State regulators.
@@Rayjon10857 Re: The Waymo technology is capable, just not approved by each State regulators. Waymo started transporting passengers in 2018 but by 2024 they've "expanded" into just a couple of locations. They dont even cover the whole of Phoenix and they've been there for 5 years. They've had plenty of time to get approvals. All this points to a flawed, non-scalable strategy, not a capable technology. However, I think they're changing their strategy and you can see it in some of the Phoenix videos by JJRicks. About 8 months ago their path planner began twitching like Tesla's did in the early days. Its different in SF where its the old software as you can see the path planner very stable and very much on its rails. I think if Waymo ends up winning the race to FSD, it'll be because they skip "driving AI" entirely and jump to something much more akin to AGI.
What do you mean, the car is still driving by itself while their interaction is occurring based on what it thinks is safe. It is not like it's stopping in the middle of the highway.
Except Waymos are crashing everywhere, driving the wrong way, stopping in the middle of the road, and making dangerous maneuvers. There's videos everywhere out here in AZ of this. Especially in constructiom zones / dense areas. These are dangerous and need to all be recalled and the company shut down
The other measure to note is how many road miles are in the ring fenced areas for the Waymo driver to navigate in. Tesla may need 1,000x interventions today but it probably has 10,000x the road miles to deal with.
Gasp! So you're actually saying that Tesla will need a remote fleet response team after all? I'm shocked I say, shocked, at this sudden reversal of opinion. You'd better watch out for all those Tesla fanbois out there, they may turn on you..... Tesla are obviously dragging their feet on implementing this. I look forward to hearing them admit that their cars aren't actually capable of human-level autonomous reasoning and communication. And there is still the issue of sensor/system redundancy: fine, no lidar, but surely advanced radar is coming back? The longer this goes on the more cars will be out there with aging/missing hardware that they promised owners would have level 4 FSD "sometime this year". I smell an angry mob/class action lawsuit.
Re: And there is still the issue of sensor/system redundancy Why? We drive fine with vision only. Why do you think a different sensor is needed when demonstrably its not.
@@medhurstt Yeah, vision only systems do just fine in heavy rain, fog, poor lighting….. Every manufacturer is doing some sort of sensor fusion to handle these conditions, except Tesla. I don’t think NHTSA is just going to just roll over and give Tesla alone a pass on this, especially for the early level 4 systems, but I guess that is the gamble they’re willing to take. ua-cam.com/video/33tGvrKKi1k/v-deo.html
@@medhurstt And the other 49 vehicles? Three people died….. These incidents are pretty common: ua-cam.com/video/2nHCRPWRU2I/v-deo.html ua-cam.com/video/-QCmuqS0Kac/v-deo.html ua-cam.com/video/2pia7PkXTvc/v-deo.html Also, I’m not aware of any proposed commercial truck autonomous driving system that is using vision only. Even the Tesla semi is radar-equipped for autopilot, although no FSD yet.
@@marksinclair701 But that's an example of cars and trucks going way too fast for the conditions. Its an example of people driving too close to stop safely and going too fast for the conditions. Any self driving car would/should have been going slower at a safer distance and likely have stopped safely but that doesn't stop the moron behind slamming into them. Consider a Waymo in the middle of that. Do you think it would have done better than the rest? No of course not. Any FSD solution, to be safe, has to drive such that its not going too fast to stop before the limit of its sensors (whether they're vision, or whatever) given the conditions (eg ice and snow). That means if its pea soup fog, the speed has to be very slow and there's no reason in principle why vision is any different from other solutions.
The intersting question is if Tesla can do the same , if it’s technically doable ? That would mean Tesla could start a robotaxi fleet even when it still have some interventions , and this should affect the stock price and move it quicker upwards as this scenario would be possible much earlier than a robotaxi that is totally independent .
The Tesla phone app can already “intervene” by sending the car a new route while it is driving. Supposedly Tesla will have a robotaxi app and similar communication of changing instructions shouldn’t be too difficult.
Tesla FSD is not at a point where safety critical interventions are low enough to allow level 4. I highly doubt tesla will go from level 2 to level 4 for FSD. They will probably have FSD be level 3 for a few months before adding in a level 4 option to the robotaxi. Hard to say though. If 12.4 is a 10x improvement and then 12.5 is a 10x improvement on that we'd be at 100x better than right now which would be better than human. I'm hyped to see level 4 sometime this week if Elon's time scale is good.
@@mrperson1324 Tesla strongly relies on driver - AI assumes the driver is watching. It isn't a bad thing as long as driver is not stupid and doesn't try to break safety systems. I think it's enough to lower the threshold where AI decides to perform more risky actions, without asking a driver. It would be good enough even right now.
Tesla could use either the waymo system like this , or connect to LLM , like Grok , where the passengers themselves can help FSD and tell it what to do if it’s “lost” , or even passengers driving the cars themselves , and then the cars drives themselves only to the next customer (empty). This would be easier to get through regulations. There is now , IMHO , so many possible usecases already with existing technology , so even if FSD from today didn’t get any better , it could still be used somehow and contribute to revenue , thus also reducing stock risk / downside substantially .
@user-po8or5jo5q Yeah but I'm not sure it's safe enough to do that. Waymo vehicles are already safe even without the intervention. They can manage a safe state (maybe not without annoying someone or blocking traffic) and understand when a situation they may need help is present. They will not hit someone but FSD has hit cars. They are on another level of safety.
At a high level, you can think of the levels of automation as: Level 0: No automation at all Level 1: Very light automation (cruise control, etc.) Level 2: Some automation but requires human attention at all times Level 3: Can self-drive but require intervention in severe conditions Level 4: Highly autonomous Level 5: Completely autonomous Its claimed Waymo operates at L4 but I think its pretty clear its really L3 where "severe conditions" does include weather but also unusual road conditions. Cones were mentioned. I think this kinda means Waymo have gone for a multi level implementation of autonomy. Most of the driving is performed by the car and some of the driving is directed by humans remotely....but that structure will surely support an intermediate level of autonomy with direction from a central AI and fallback to human direction. So Car AI then Centralised AI then human. Its the only way they could hope to expand! My2c
This kind of operation is clearly level 4 to me. Level 3 would be much more limited and more like L2 with the big difference that the driver doesn't have to pay attention and monitor the car.
@EinzigfreierName It comes down to the frequency of intervention, and there's only really one channel that regularly covers waymo drives, and he's had multiple interventions documented. He puts the car into situations where intervention is more likely and it happens.
@@EinzigfreierName The point is that Waymo appears to be L4 only because its small scale. If there were a million Waymos out there then it'd be far more obvious that they're not L4 as the reported incidents would be much higher and more obvious. There's no specific definition for L3 vs L4, its subjective.
@@medhurstt I agree there is no strict border between L3 and L4. But the number of miles driven between interventions and the situations which need intervention don't depend on the number of cars on the road. To me these vehicles are autonomous enough to call them level 4. Just look at Mercedes Drive Pilot in comparsion for a level 3 system. Totally different league.
Tesla shouldn't copy this. I think calling for Tesla to imitate this brute-force crutch is giving up just when true self-driving is almost achieved. Tesla must just drive themselves, period. No compromises. IN the game of faith you get off when you are asked to believe beyond your capacity.. I sense y'all think FSD is as good as it gets. I disagree. Teslas will drive themselves with 0 human guidance.
Waymo’s human is intervening just like when a human intervenes FSD. Try to imagine Tesla’s FSD stopped and stuck. Then the human intervenes by taking over and driving around the truck and back onto the road. That’s essentially what the Waymo human agent did when he/she gave instructions to the car to drive around the truck, onto the sidewalk and back onto the road.
@@erwile Waymo car was waiting for something, such as the truck to move. The human told the Waymo car to go around the truck. The car went on the sidewalk, which was bad driving. The human dropped a car icon on the road, which told the car to drive there. Play word games if you want, such as calling it "strategy", but the human told the car how to drive. Similar interventions happen with FSD.
@@jlp0 if a human jump in front of the car after the remote controller sent the path , the Waymo car would have been responsible for the urgence manoeuver. With fsd it's solely the human driver
@@erwile False. There are numerous videos showing the FSD stopping for cats crossing the road, child mannequins pulled in front of the car, and stopping for pedestrians or cyclists who aren't even on the road but started to move towards the road to cross. Also, based on statistics, FSD is safer than human drivers and human drivers are safer than Waymo and Cruise.
@@jlp0 hey man, I'm pro-tesla and I watch too much hours of fsd video, relax. I know human + fsd is safer than only a human driving. But if a human take over fsd, it's the driver responsibility to brake or avoid anything (even though the car can act if needed, you shouldn't rely on it). Another example, if you take over Tesla fsd, if you let the steering wheel with no fsd, you will probably go off-road. Waymo remote controller are not driving (they're not taking over if you think about it, not in the sense of fsd) so they rely on the car to avoid collision if needed, or follow the route. They give a rough path and the car has to brake / avoid anything on the road. Tesla doesn't do that, which is fine, because it's not a level 4 system right now, just a level 2 ++++++
I would say, just not hitting another car is not enough. The car should also not put himself into a position where it is likely being hit by a other car, like going wrong way, ignoring Stop sign or red light. So the law should state, do not hit anything and do not do something so that is becomes likely that somebody else is hitting you.
Waymo's system best fits the definition of Level 3 where it can ask the human driving (who in this case is remote) to intervene in the driving decision making.
I don't think so, the human is not really driving, it's just giving hint on where to drive. they're Level 4. The car is fully responsible for driving and can't ask someone to "really drive". Level 3 would mean someone would have to take over immediately.
The car is driving though. Heck, they could simply not respond to the intervention and the car would keep driving. It can respond on its own although maybe not super effectively.
For me, the takeaway is that a system like Waymo will never be able to scale efficiently because it requires too many guard rails that don’t scale. We already knew it needs accurate, up-to-date, geofenced, HD city maps, and it basically rides on invisible “rails”, but this revelation about emote safety drivers makes my assessment even worse. Every time Waymo tries to scale (more cars, expanded geofence area, more cities), the long tail of corner cases will grow, and so will the need for remote interventions.
The more interventions, the easier is to train AI model based on them. The more data they collect and the better the model, the less interventions. Honestly I prefer Tesla's approach, but it doesn't mean Waymo will fail. Possibly they might win the race, it's only just a beginning.
It's true that this approach doesn't scale too well and that's why robotaxis will not take in teh next few years. Because there is not better approach so far.
@@EinzigfreierName You dont think the Tesla approach of millions of cars providing training data across most of the drivable world isn't a better approach? We've already seen the benefit of a full AI stack over hardcoded rules. But its going to require seeing the edge cases and Waymo just doesn't do that and for all intents and purposes, never will.
@@medhurstt Both approaches have advantages and disadvantages. If regulations change or you want to behave the vehicle differently, this is much easier accomplished with hard coded rules instead of training the AI for it. But my main argument is - if Teslas approach is superior why are there no Tesla robotaxis on the road yet? I'm not even fully convinced about the "vision only" approach will work for that. I wouldn't be surprised to see LIDAR and radar on the first real Tesla robotaxi. And for sure they will be remotely supervised and geofenced just as Waymos are.
@@EinzigfreierName Re: if Teslas approach is superior why are there no Tesla robotaxis on the road yet? Because they're solving a different problem to Waymo. Waymo solved a very small subset of driving and used shortcuts by putting the car on rails using the lidar for very precise positioning in their HD map whereas Tesla are solving driving completely for all scenarios. Its orders of magnitude more difficult. Lidar is unnecessary as v12 proves but that was always going to be the case. Take the worlds best driver and give them perfect attention. That's where vision only can get us. Do you think that's not good enough? And if not, why not?
Tesla Robo Taxi will need some kind remote response system to control which Robo Taxis will be selected to pick up a passenger at a certain location as you can't have a bunch of different Robo Taxies going to the same location at once for that one customer. Secondly you will still need to have some geo fencing to keep your Taxi from going into inner city neighborhoods that more dangerous than suburban ones where this type people likely damage or steal your car, keep it from going out of State and to keep them out of government, airport, military, etc. restricted area. Removing the steering wheel will pedals will be a big mistake because if something happens and needs a human to take over the driving and to move it to a safer place for it to be two. So, you will also need a recues team too! So, at the end of the day there is a lot of work to put together a Robo Taxi Network and it going to be very expensive one too! You just don't send your car out the road and make money. Just my opinion.
the wheel could be replaced by a joystick plugged into the usb-c or bluetooth (the cybertruck does not use a mechanical steering). Automating which car goes where is quite easy (uber is doing it with human driver...), mapping dangerous area too (you can disable/discourage areas where the system has issues, just use human labor to label regions (it could also drive slower or not park here), until an update can address that) A rescue team will be needed but that's not a big deal. Nothing is really expensive in these systems, and as the cars improves it would be less and less expensive per profit.
Of course can Waymo cars drive themselves. But there are cases that need to be resolved by a human. Tesla has not even shown driverless rides yet. Why do you believe they can do it without any human intervention? They can't.
@@EinzigfreierName Tesla can because I use it everyday and I am just forced to touch the. steering wheel when I don’t need to. ua-cam.com/video/MA12MNFxwoA/v-deo.htmlsi=q2kopcwAYSd1O69c
Im sorry, but they mislead orginally and tried to cover up their actions. This fits in line with that mentality and unless they show live dual feeds from randomly selected vehicles and simultaneously showing the support centers ...for days at random times...unedited...it will remain lies and half truths until then. If they want trust back they need to beyond transparent. Until then this is propaganda.
Waymo’s awesome remote fleet response system is the only way they can have so many cars constantly operating fully autonomously without any safety driver on board. It’s clear and evident in almost everything that they do that they obviously understand the importance of reputation and safety record and how crucial it is for public trust in adoption of this new technology. The engineers they got from Anki are wicked smart too. (RIP to one of the coolest little robotics manufacturers)
It’s been awesome to watch them grow over the years. I am hoping they will soon allow them to travel on highways as well. I can’t wait to see what the next generation Waymo’s will be like as well, since they’ve been using the jag ipace platform for about 4 years now. Especially with the recent big leaps everyone has been making in their NPUs and CPUs,
Honestly did a double take seeing it was you covering this. But really glad that you are.
And when they have millions of cars worldwide?
Thanks for helping explain the Waymo remote driver concept. Now I can understand how it actually is working...
Wow, this is actually huge.
Even though it requires a human 'driver', I suspect 1 could cover maybe 100 vehicles or so. And that'll increase as the software gets better.
Robotaxis really are nearly here. And after seeing this, I think Waymo and Tesla will be leading the way. Exciting times
The Waymo visualisations are massively detailed and accurate. I wonder if they are combining the Video Detail with the Computer analysis to augment the details. Such as the exact type of vehicle / shape variation, Pedestrian’s Shopping Bags, UPS Truck with detailing of UPS on the side….
Great job Waymo !!!
Yup, but totally not important. Tesla replaces objects with ugly default 3d models while Waymo spends process power to let their 3d rendering look the same as in real world. No effect on self driving capabilities though, just visualisation.
@@kafiluz4317 Yeah, but if it's the interface for the operator, I'm sure an operator would prefer to see some actual humans doing human things, not just dummies "walking" around.
@@vaddimka It's about priorities. First I would like to see all objects that FSD can see before they make "augmented reality" modelling like Waymo.
@@kafiluz4317 This is a video about how Waymo remote operators work. I'd like FSD to not lean left or right in the lane (or just accept my corrections without disengaging), but that's a topic for another video.
@@vaddimka you're right. It's just some kind of a Tesla channel as well ;-)
This is very interesting. Impressive stuff from Waymo here. Likely though, they've just paved the way for Tesla to implement their own flavour of this remote intervention protocol.
Yes, what matters is measuring and minimising time critical interventions that cannot be handled remotely.
Thank You folks for All that you are doing for our Planet Earth
Peace.. Shalom.. Salam.. Namaste
🙏🏻 😊 ✌ ☮ ❤ 🕊
It's great that Waymo published this and their approach is very pragmatic and smart. Unfortunately, they didn't publish a very important aspect: What is the remote operator intervention per X number of miles? That number essentially tells you how scalable their approach is. If they need 1 operator per 10 cars, that is never going to work on even a large scale. 1 operator per 1000 starts to get practical and at 1 operator per 10000, we are talking amazing.
There are about 5 million Uber drivers who drive more than occasionally. At a rate of one fleet response agent per 1000 Waymos, it would take 5000 agents to cover 5 million Waymos. Currently there is a total of a few hundred Waymo vehicles.
That's actually really cool
Great job explaining how Waymo is operating.
I've used Waymo a number of times in SF, but haven't noticed an intervention.
I'd love to try FSD, and I'm thinking of leasing a 3, but I need to sell my old '88 Porsche 911 first.
Really helpful that you shared this
"Okay AI, now just keep improving."
AI: Goes looking for situations it can't handle. Becomes thrill-chasing adrenaline junkie.
Interesting
Thx Omar !
As FSD gets better and better, I wouldn’t mind this type of question and answer “intervention” over manually taking over to correct behavior
Even with some intervention demanded by the car. It could be safety critical (not like today, the car would be in control but hard breaking and asking for help), "just in case" (slowing down or staying in its place) or to improve learning in a specific situation that they lack data of (just a message asking to take over to get some data).
That’s smart
Fair reporting
I like the dropping in a car to tell Waymo where to go to resume the journey. This video is interesting and informative. Are the remote operators human? I ask because it navigates so differently than FSD.
Grata conhecimento e vida nos liberta
awesome
Good information ❤
Love ur narrated content
I wonder....Will this suggestion from the operation be count as an intervene? because it's look like intervene to me. However, I think we also have to find a better metrics to measure the readiness of self-driving car. Should count how much it's impact/inconvenience other people on the road as well, not just full blown accident.
My, how the tables have turned. Now the automated device has to reach out to the human tech support 😂
I’ve always wondered if the Fsd team could ask the system why it did something.
Great, but does it learn from it’s interventions? Or does it repeat its mistakes?
how many miles per disengagement
What it sounds like to me is that its not full self driving.
Nothing is at this point.
12.4 not in yet?
The explains why not high speed and highway, there is just no time to wait for this interactions.
Waymo vehicles are operating on the freeways for road-testing until regulatory approvals.
You can find some of the videos of freeway driving.
The Waymo technology is capable, just not approved by each State regulators.
@@Rayjon10857 Re: The Waymo technology is capable, just not approved by each State regulators.
Waymo started transporting passengers in 2018 but by 2024 they've "expanded" into just a couple of locations. They dont even cover the whole of Phoenix and they've been there for 5 years. They've had plenty of time to get approvals. All this points to a flawed, non-scalable strategy, not a capable technology.
However, I think they're changing their strategy and you can see it in some of the Phoenix videos by JJRicks. About 8 months ago their path planner began twitching like Tesla's did in the early days. Its different in SF where its the old software as you can see the path planner very stable and very much on its rails.
I think if Waymo ends up winning the race to FSD, it'll be because they skip "driving AI" entirely and jump to something much more akin to AGI.
What do you mean, the car is still driving by itself while their interaction is occurring based on what it thinks is safe. It is not like it's stopping in the middle of the highway.
Except Waymos are crashing everywhere, driving the wrong way, stopping in the middle of the road, and making dangerous maneuvers. There's videos everywhere out here in AZ of this. Especially in constructiom zones / dense areas. These are dangerous and need to all be recalled and the company shut down
They must have been sleeping when the Arizona one crashed into that pole.
How many miles between each assistance ?
The other measure to note is how many road miles are in the ring fenced areas for the Waymo driver to navigate in. Tesla may need 1,000x interventions today but it probably has 10,000x the road miles to deal with.
Gasp! So you're actually saying that Tesla will need a remote fleet response team after all? I'm shocked I say, shocked, at this sudden reversal of opinion. You'd better watch out for all those Tesla fanbois out there, they may turn on you.....
Tesla are obviously dragging their feet on implementing this. I look forward to hearing them admit that their cars aren't actually capable of human-level autonomous reasoning and communication. And there is still the issue of sensor/system redundancy: fine, no lidar, but surely advanced radar is coming back? The longer this goes on the more cars will be out there with aging/missing hardware that they promised owners would have level 4 FSD "sometime this year". I smell an angry mob/class action lawsuit.
Re: And there is still the issue of sensor/system redundancy
Why? We drive fine with vision only. Why do you think a different sensor is needed when demonstrably its not.
@@medhurstt Yeah, vision only systems do just fine in heavy rain, fog, poor lighting…..
Every manufacturer is doing some sort of sensor fusion to handle these conditions, except Tesla. I don’t think NHTSA is just going to just roll over and give Tesla alone a pass on this, especially for the early level 4 systems, but I guess that is the gamble they’re willing to take.
ua-cam.com/video/33tGvrKKi1k/v-deo.html
@marksinclair701 How it the example of a moron in a truck going way too fast for the conditions, an example of vision only inadequacy?
@@medhurstt And the other 49 vehicles? Three people died…..
These incidents are pretty common:
ua-cam.com/video/2nHCRPWRU2I/v-deo.html
ua-cam.com/video/-QCmuqS0Kac/v-deo.html
ua-cam.com/video/2pia7PkXTvc/v-deo.html
Also, I’m not aware of any proposed commercial truck autonomous driving system that is using vision only. Even the Tesla semi is radar-equipped for autopilot, although no FSD yet.
@@marksinclair701 But that's an example of cars and trucks going way too fast for the conditions. Its an example of people driving too close to stop safely and going too fast for the conditions. Any self driving car would/should have been going slower at a safer distance and likely have stopped safely but that doesn't stop the moron behind slamming into them.
Consider a Waymo in the middle of that. Do you think it would have done better than the rest? No of course not.
Any FSD solution, to be safe, has to drive such that its not going too fast to stop before the limit of its sensors (whether they're vision, or whatever) given the conditions (eg ice and snow). That means if its pea soup fog, the speed has to be very slow and there's no reason in principle why vision is any different from other solutions.
This is as I'd expect. Remote teleoperation is essentially a corner case solve by humans. Problem is FSD requires autonomous corner case solving.
"Am I on fire?"
"Yes."
"Why?"
"Because you're in San Francisco. "
Be cool if Tesla used Grok (or similar) to also potentially ask passengers for assistance maybe before sending to a remote operator.
The intersting question is if Tesla can do the same , if it’s technically doable ? That would mean Tesla could start a robotaxi fleet even when it still have some interventions , and this should affect the stock price and move it quicker upwards as this scenario would be possible much earlier than a robotaxi that is totally independent .
The Tesla phone app can already “intervene” by sending the car a new route while it is driving. Supposedly Tesla will have a robotaxi app and similar communication of changing instructions shouldn’t be too difficult.
Tesla FSD is not at a point where safety critical interventions are low enough to allow level 4. I highly doubt tesla will go from level 2 to level 4 for FSD. They will probably have FSD be level 3 for a few months before adding in a level 4 option to the robotaxi. Hard to say though. If 12.4 is a 10x improvement and then 12.5 is a 10x improvement on that we'd be at 100x better than right now which would be better than human. I'm hyped to see level 4 sometime this week if Elon's time scale is good.
@@mrperson1324 Tesla strongly relies on driver - AI assumes the driver is watching. It isn't a bad thing as long as driver is not stupid and doesn't try to break safety systems.
I think it's enough to lower the threshold where AI decides to perform more risky actions, without asking a driver. It would be good enough even right now.
Tesla could use either the waymo system like this , or connect to LLM , like Grok , where the passengers themselves can help FSD and tell it what to do if it’s “lost” , or even passengers driving the cars themselves , and then the cars drives themselves only to the next customer (empty).
This would be easier to get through regulations.
There is now , IMHO , so many possible usecases already with existing technology , so even if FSD from today didn’t get any better , it could still be used somehow and contribute to revenue , thus also reducing stock risk / downside substantially .
@user-po8or5jo5q Yeah but I'm not sure it's safe enough to do that. Waymo vehicles are already safe even without the intervention. They can manage a safe state (maybe not without annoying someone or blocking traffic) and understand when a situation they may need help is present. They will not hit someone but FSD has hit cars. They are on another level of safety.
At a high level, you can think of the levels of automation as:
Level 0: No automation at all
Level 1: Very light automation (cruise control, etc.)
Level 2: Some automation but requires human attention at all times
Level 3: Can self-drive but require intervention in severe conditions
Level 4: Highly autonomous
Level 5: Completely autonomous
Its claimed Waymo operates at L4 but I think its pretty clear its really L3 where "severe conditions" does include weather but also unusual road conditions. Cones were mentioned.
I think this kinda means Waymo have gone for a multi level implementation of autonomy. Most of the driving is performed by the car and some of the driving is directed by humans remotely....but that structure will surely support an intermediate level of autonomy with direction from a central AI and fallback to human direction. So Car AI then Centralised AI then human. Its the only way they could hope to expand!
My2c
This kind of operation is clearly level 4 to me. Level 3 would be much more limited and more like L2 with the big difference that the driver doesn't have to pay attention and monitor the car.
@EinzigfreierName It comes down to the frequency of intervention, and there's only really one channel that regularly covers waymo drives, and he's had multiple interventions documented. He puts the car into situations where intervention is more likely and it happens.
@@medhurstt Of course it happens. Otherwise it would be level 5 already.
@@EinzigfreierName The point is that Waymo appears to be L4 only because its small scale. If there were a million Waymos out there then it'd be far more obvious that they're not L4 as the reported incidents would be much higher and more obvious.
There's no specific definition for L3 vs L4, its subjective.
@@medhurstt I agree there is no strict border between L3 and L4. But the number of miles driven between interventions and the situations which need intervention don't depend on the number of cars on the road. To me these vehicles are autonomous enough to call them level 4. Just look at Mercedes Drive Pilot in comparsion for a level 3 system. Totally different league.
They pay fiverr i fluencere 5$ an hour
Tesla shouldn't copy this. I think calling for Tesla to imitate this brute-force crutch is giving up just when true self-driving is almost achieved.
Tesla must just drive themselves, period. No compromises.
IN the game of faith you get off when you are asked to believe beyond your capacity.. I sense y'all think FSD is as good as it gets. I disagree. Teslas will drive themselves with 0 human guidance.
Waymo’s human is intervening just like when a human intervenes FSD. Try to imagine Tesla’s FSD stopped and stuck. Then the human intervenes by taking over and driving around the truck and back onto the road. That’s essentially what the Waymo human agent did when he/she gave instructions to the car to drive around the truck, onto the sidewalk and back onto the road.
no, driving is not like telling where to go, the remote people are not responsible for how the car drive, they only give a "strategy" .
@@erwile Waymo car was waiting for something, such as the truck to move. The human told the Waymo car to go around the truck. The car went on the sidewalk, which was bad driving. The human dropped a car icon on the road, which told the car to drive there. Play word games if you want, such as calling it "strategy", but the human told the car how to drive. Similar interventions happen with FSD.
@@jlp0 if a human jump in front of the car after the remote controller sent the path , the Waymo car would have been responsible for the urgence manoeuver. With fsd it's solely the human driver
@@erwile False. There are numerous videos showing the FSD stopping for cats crossing the road, child mannequins pulled in front of the car, and stopping for pedestrians or cyclists who aren't even on the road but started to move towards the road to cross. Also, based on statistics, FSD is safer than human drivers and human drivers are safer than Waymo and Cruise.
@@jlp0 hey man, I'm pro-tesla and I watch too much hours of fsd video, relax. I know human + fsd is safer than only a human driving.
But if a human take over fsd, it's the driver responsibility to brake or avoid anything (even though the car can act if needed, you shouldn't rely on it). Another example, if you take over Tesla fsd, if you let the steering wheel with no fsd, you will probably go off-road. Waymo remote controller are not driving (they're not taking over if you think about it, not in the sense of fsd) so they rely on the car to avoid collision if needed, or follow the route. They give a rough path and the car has to brake / avoid anything on the road. Tesla doesn't do that, which is fine, because it's not a level 4 system right now, just a level 2 ++++++
I'm pretty sure they're doing this before the release of Fsd 12.4. They know something that we don't.
Tesla should be already copying all that
I would say, just not hitting another car is not enough. The car should also not put himself into a position where it is likely being hit by a other car, like going wrong way, ignoring Stop sign or red light. So the law should state, do not hit anything and do not do something so that is becomes likely that somebody else is hitting you.
Waymo's system best fits the definition of Level 3 where it can ask the human driving (who in this case is remote) to intervene in the driving decision making.
I don't think so, the human is not really driving, it's just giving hint on where to drive. they're Level 4. The car is fully responsible for driving and can't ask someone to "really drive". Level 3 would mean someone would have to take over immediately.
The car is driving though. Heck, they could simply not respond to the intervention and the car would keep driving. It can respond on its own although maybe not super effectively.
I want tesla to release a feature where we can remote control our tesla using a PS5 controller to do robotaxi remotely 😂
For me, the takeaway is that a system like Waymo will never be able to scale efficiently because it requires too many guard rails that don’t scale. We already knew it needs accurate, up-to-date, geofenced, HD city maps, and it basically rides on invisible “rails”, but this revelation about emote safety drivers makes my assessment even worse. Every time Waymo tries to scale (more cars, expanded geofence area, more cities), the long tail of corner cases will grow, and so will the need for remote interventions.
The more interventions, the easier is to train AI model based on them. The more data they collect and the better the model, the less interventions.
Honestly I prefer Tesla's approach, but it doesn't mean Waymo will fail. Possibly they might win the race, it's only just a beginning.
It's true that this approach doesn't scale too well and that's why robotaxis will not take in teh next few years. Because there is not better approach so far.
@@EinzigfreierName You dont think the Tesla approach of millions of cars providing training data across most of the drivable world isn't a better approach? We've already seen the benefit of a full AI stack over hardcoded rules. But its going to require seeing the edge cases and Waymo just doesn't do that and for all intents and purposes, never will.
@@medhurstt Both approaches have advantages and disadvantages. If regulations change or you want to behave the vehicle differently, this is much easier accomplished with hard coded rules instead of training the AI for it. But my main argument is - if Teslas approach is superior why are there no Tesla robotaxis on the road yet? I'm not even fully convinced about the "vision only" approach will work for that. I wouldn't be surprised to see LIDAR and radar on the first real Tesla robotaxi. And for sure they will be remotely supervised and geofenced just as Waymos are.
@@EinzigfreierName Re: if Teslas approach is superior why are there no Tesla robotaxis on the road yet?
Because they're solving a different problem to Waymo. Waymo solved a very small subset of driving and used shortcuts by putting the car on rails using the lidar for very precise positioning in their HD map whereas Tesla are solving driving completely for all scenarios. Its orders of magnitude more difficult. Lidar is unnecessary as v12 proves but that was always going to be the case. Take the worlds best driver and give them perfect attention. That's where vision only can get us. Do you think that's not good enough? And if not, why not?
Tesla Robo Taxi will need some kind remote response system to control which Robo Taxis will be selected to pick up a passenger at a certain location as you can't have a bunch of different Robo Taxies going to the same location at once for that one customer. Secondly you will still need to have some geo fencing to keep your Taxi from going into inner city neighborhoods that more dangerous than suburban ones where this type people likely damage or steal your car, keep it from going out of State and to keep them out of government, airport, military, etc. restricted area. Removing the steering wheel will pedals will be a big mistake because if something happens and needs a human to take over the driving and to move it to a safer place for it to be two. So, you will also need a recues team too! So, at the end of the day there is a lot of work to put together a Robo Taxi Network and it going to be very expensive one too! You just don't send your car out the road and make money. Just my opinion.
the wheel could be replaced by a joystick plugged into the usb-c or bluetooth (the cybertruck does not use a mechanical steering). Automating which car goes where is quite easy (uber is doing it with human driver...), mapping dangerous area too (you can disable/discourage areas where the system has issues, just use human labor to label regions (it could also drive slower or not park here), until an update can address that) A rescue team will be needed but that's not a big deal. Nothing is really expensive in these systems, and as the cars improves it would be less and less expensive per profit.
lol waymo sucks balls
Needing remote control means car cannot drive itself like Tesla FSD AI
Of course can Waymo cars drive themselves. But there are cases that need to be resolved by a human. Tesla has not even shown driverless rides yet. Why do you believe they can do it without any human intervention? They can't.
@@EinzigfreierName Tesla can because I use it everyday and I am just forced to touch the. steering wheel when I don’t need to. ua-cam.com/video/MA12MNFxwoA/v-deo.htmlsi=q2kopcwAYSd1O69c
Im sorry, but they mislead orginally and tried to cover up their actions. This fits in line with that mentality and unless they show live dual feeds from randomly selected vehicles and simultaneously showing the support centers ...for days at random times...unedited...it will remain lies and half truths until then. If they want trust back they need to beyond transparent. Until then this is propaganda.