FSD discussion

[EVNow]

Source please, not just the graph. I want to dig into how that is reported, b/c something smells off there. That has not been my personal experience (in the last 3 months) with Cruise in SF.

Search for CA AV report. You get to see all the details and IIRC, description of each disengagement. If you actually find something fishy - don't forget to report to the CA authorities.

Disengagement Reports - California DMV

Autonomous vehicle manufacturers that are testing vehicles in the Autonomous Vehicle Tester (AVT) Program and AVT Driverless Program are required to submit annual reports to share how often their vehicles disengaged from autonomous mode during tests.

www.dmv.ca.gov

You may be able to get more information from Waymo & Cruise websites.

Waymo Safety Report - 2021 | Let's Talk Autonomous Driving

ltad.com

 

[Singuy]

I don't think they make similar mistakes. Their disengagement rates are in the range of 1 in 10k miles and Tesla's is 1 in 10 miles.

I don't even try to compare Robotaxis with FSD. They have a completely different business model (which doesn't seem to work at present). Tesla's FSD aim is to offer good ADAS and potentially L3 level at some point on consumer cars - so I compare that to MobilEye and other competitors.

Must be my imagination when I see waymo slaming on the brakes and cruise getting stuck.

I have some fsd routes if taken will have zero disengagement for tens of thousands of miles. Geo fencing, it just works. And if it doesn't then just geo fence some more. This is why waymo avoids traffic and difficult roads..but those numbers are great. Tesla however needs to do a unprotected left over 3 lanes of 60 mph cross traffic while needing to use the median with obstructed views. Either get that right or bust while waymo circles abandoned neighborhoods never needing to make an unprotected turn.

 

[bkp_duke]

Search for CA AV report. You get to see all the details and IIRC, description of each disengagement. If you actually find something fishy - don't forget to report to the CA authorities.

Disengagement Reports - California DMV

Autonomous vehicle manufacturers that are testing vehicles in the Autonomous Vehicle Tester (AVT) Program and AVT Driverless Program are required to submit annual reports to share how often their vehicles disengaged from autonomous mode during tests.

www.dmv.ca.gov

You may be able to get more information from Waymo & Cruise websites.

Waymo Safety Report - 2021 | Let's Talk Autonomous Driving

ltad.com

So, honest question, am I the only one that looked that raw data (did just now)? The number of reported incidents for Cruise is 9. NINE.

I've had that many myself in a Cruise vehicle in the past 18 months during my SF trips. Things that literally are listed for those 9 events, I have experienced.

I'm smelling a high degree of "self-reporting" bias in this dataset.

EDIT - and we have photos of videos of Cruise and Waymo being in accidents. Hitting the back of a bus, etc. None of those events appear to be listed in that dataset.

 

[EVNow]

So, honest question, am I the only one that looked that raw data (did just now)? The number of reported incidents for Cruise is 9. NINE.

I've had that many myself in a Cruise vehicle in the past 18 months during my SF trips. Things that literally are listed for those 9 events, I have experienced.

I'm smelling a high degree of "self-reporting" bias in this dataset.

Pickup the phone and report ! BTW, did you actually have more than 9 collisions (what they call "contacts") .... sounds absurdly unlikely. Or are these things what you thought weren't good driving ... ?

BTW, here is the Waymo research paper.

https://storage.googleapis.com/waymo-uploads/files/documents/safety/Safety%20Performance%20of%20Waymo%20RO%20at%201M%20miles.pdf

ps : I tend to believe Waymo data more than Cruise .... because I think Google has a lot of reputational risk compared to Cruise.

 

[EVNow]

I have some fsd routes if taken will have zero disengagement for tens of thousands of miles. Geo fencing, it just works. And if it doesn't then just geo fence some more. This is why waymo avoids traffic and difficult roads..but those numbers are great. Tesla however needs to do a unprotected left over 3 lanes of 60 mph cross traffic while needing to use the median with obstructed views. Either get that right or bust while waymo circles abandoned neighborhoods never needing to make an unprotected turn.

Thats why we shouldn't compare. BTW, if you do want to compare, you should make a lot of drives in SF in the area Waymo operates. But that will still be a small # and not really statistically valid.

Tesla's current business is consumer cars - which will encounter all kinds of cases. Waymo's business case is - they need to provide robotaxi services in large cities.

 

[bkp_duke]

Pickup the phone and report ! BTW, did you actually have more than 9 collisions (what they call "contacts") .... sounds absurdly unlikely. Or are these things what you thought weren't good driving ... ?

BTW, here is the Waymo research paper.

https://storage.googleapis.com/waymo-uploads/files/documents/safety/Safety%20Performance%20of%20Waymo%20RO%20at%201M%20miles.pdf

ps : I tend to believe Waymo data more than Cruise .... because I think Google has a lot of reputational risk compared to Cruise.

These are supposed to be user reports, they are supposed to be manuf reports.

I'm def smelling some cherry-picking here.

 

[Singuy]

Thats why we shouldn't compare. BTW, if you do want to compare, you should make a lot of drives in SF in the area Waymo operates. But that will still be a small # and not really statistically valid.

Tesla's current business is consumer cars - which will encounter all kinds of cases. Waymo's business case is - they need to provide robotaxi services in large cities.

There are plenty of head to head and we found that waymo just spend 5 to 10 extra minutes avoiding traffic areas or unprotected turns. That's why the performance doesn't impress me when 90% of the success is attributed to map routing.

 

[Singuy]

Waymo and Cruise has plenty of problems and they just skip many of them. The amount of crashes they are involved in are more than anyone expect, most of them due to cars hitting them vs them hitting other cars. This is due to their unnatural way of driving, going at or under the speed limit, slamming on brakes and do weird things. Dealing with irrational and impatient drivers on the road is not part of Waymo/Cruise's core competencies. It's about getting the passenger to the destination at low speed, routing for low traffic areas, and just straight out not be available if there's construction around the destination. Tesla wants to solve it all which will take time. They are not in a hurry to demonstrate L4 driving for angel investors.

 

[InTheMiddle]

Not a disagree, I just don't see Waymo's solution as either cost-effective or scalable long-term. It's more a proof-of-concept demonstration (which is at least better than what MobileEye has done).

Tesla's solution, while funded by FSDb early adopters, has the potential to properly scale up and through L2, L3, L4, and L5, while remaining cost-effective from a hardware standpoint.

As a stand-alone entity, I would concur that Waymo's solution would be very challenging to a) fund thru development / deployment to a point of self-sufficiency and b) to operate profitably at prices that are cost-competitive with human-driven taxis / rideshares.

However, consider Waymo purely as an extension of a large, deep-pocketed advertising company. Traditional taxis are painfully-ad-heavy at this point; first it was signage on the vehicles, then it was a top-hat added with additional rotating ads, and now even the presumed tranquility of the back seat is replaced with making you a near-captive audience while a seat-back screen right in front of you is running ads that you cannot turn off. Replace this anonymous-rider advertising with complete targeting of the individual rider (who booked the trip on an app) and the advertising revenue potential for Waymo is limited only by how much riders will put up with having ads forced upon them. (Sadly, repeatedly, I see humans willing to subject themselves to a great deal of ads to decrease or avoid even a minor cost.) Hence, while Waymo could go towards more niche markets (busses where per-install costs are amortized over more riders, non-emergency disability transport and other higher-value-per-ride scenarios, etc) to reach self-sustainability, I believe they will hit sustainability in Robotaxis by making the experience barely tolerable (but just barely) with endless ads to subsidize the ride cost.

(Note: MobileEye / Cruise have even less clear paths to potential sustainability unless one truly believes traditional vehicle OEMs a) will remain major players in the new vehicle market and b) will remain heavily supplier-dependent vs realizing they need to bring more capabilities back in-house.)

I do agree 100% that Tesla's solution has the potential to scale up all the way to Robotaxi-equivalence, and can remain cost-effective / affordable (to some). The real unknown is how that timeline plays out over the next <X> years. I think it's the right approach for a vehicle manufacturer to be taking, and am glad Tesla is pursuing vision and truly understanding the world around them as the solution vs LIDAR and focusing on object avoidance. I just think based on my experiences that <X> will be a longer timeframe than most here expect it to be (although shorter than some of the others here project in their pessimistic cases).

 

[Singuy]

Speaking of head to head, new Waymo vs FSDB today!

TLDR this very long head to head had Waymo falling behind FSDB by over 30 MINS. It took waymo so long that the Tesla could have made the trip back to the starting point and still beat the Waymo.

Tesla not only took the hwy, but also did not restrict itself to roads that are 35mph and made unprotected left turns. The Waymo went through multiple parking lots, apartment complexes, small roads, etc just to avoid traffic and unprotected left turns. In fact when arriving at the museum, the Tesla did an unprotected left to get into the parking lot while the Waymo went around the museum just to make a right turn into the parking lot.

 

[Knightshade]

Speaking of head to head, new Waymo vs FSDB today!

TLDR this very long head to head had Waymo falling behind FSDB by over 30 MINS. It took waymo so long that the Tesla could have made the trip back to the starting point and still beat the Waymo.

Tesla not only took the hwy, but also did not restrict itself to roads that are 35mph and made unprotected left turns. The Waymo went through multiple parking lots, apartment complexes, small roads, etc just to avoid traffic and unprotected left turns. In fact when arriving at the museum, the Tesla did an unprotected left to get into the parking lot while the Waymo went around the museum just to make a right turn into the parking lot.

If I can sleep in one, or be drunk in one, or be reading a book or playing a game in one.... but it takes longer to get there versus I have to actively be doing part of the driving task in the other but get there faster, I'm going to often prefer to go with the slower one.

Which is part of why these comparisons are so odd. They're comparing two entirely different things.

 

[uscbucsfan]

If I can sleep in one, or be drunk in one, or be reading a book or playing a game in one.... but it takes longer to get there versus I have to actively be doing part of the driving task in the other but get there faster, I'm going to often prefer to go with the slower one.

Which is part of why these comparisons are so odd. They're comparing two entirely different things.

This is fair, but the comparison exist to show the capabilities of the actual drive.

 

[Knightshade]

This is fair, but the comparison exist to show the capabilities of the actual drive.

Yeah but since one of those is "Take a lot more potentially risky driving situations on because there's a backup human paying attention" it doesn't seem especially useful comparatively.

When Tesla has an offering that removes the human from the system...or I suppose when Waymo has one that puts them back in... you'd get a comparison that might tell you something interesting.

Right now it just confirms the obvious-- super conservative driverless vehicles take longer to get places than human-driven ADAS vehicles do.

 

[Singuy]

If I can sleep in one, or be drunk in one, or be reading a book or playing a game in one.... but it takes longer to get there versus I have to actively be doing part of the driving task in the other but get there faster, I'm going to often prefer to go with the slower one.

Which is part of why these comparisons are so odd. They're comparing two entirely different things.

You are paying 45 dollars to sit in a metal box 30 min longer than an uber that most likely cost 50 dollars. This system is DOA until it actually drives like a sane person Tesla style.

Actually uber may cost the same or cheaper because you took up 1hr of time in a waymo and only 30 mins of time in an uber. Time is money in the taxi world.

 

[Webeevdrivers]

No talk about FSD (or the lack thereof). At some point they need to rip the bandage off and indicate it’s not going as planned. Long term it will be better if they fess up now.

 

[JRP3]

If I can sleep in one, or be drunk in one,

My criteria for even considering FSD.

 

[Singuy]

No talk about FSD (or the lack thereof). At some point they need to rip the bandage off and indicate it’s not going as planned. Long term it will be better if they fess up now.

Yup, Waymo and Cruise will one day need to fess up how it's not going as planned. Just need to lose a few more billion I guess before that happens.

 

[Doggydogworld]

.... Waymo falling behind FSDB by over 30 MINS.

Way back when I was in 2nd grade 28 was less than 30.

Tesla not only took the hwy, but also did not restrict itself to roads that are 35mph and made unprotected left turns.

Waymo takes 45 mph roads and makes unprotected lefts. In fact, it did so multiple times on this very video. Why lie?

In fact when arriving at the museum, the Tesla did an unprotected left to get into the parking lot while the Waymo went around the museum just to make a right turn into the parking lot.

Another lie. Waymo took the most direct route from the N. McClintock / W. Chandler intersection, heading east on W. Chandler then turning right on S. Chandler Village Dr. and approaching the museum from the north. Tesla was on the 202 highway, which runs south of the museum, so after exiting onto the frontage road it turned left onto S. Chandler Village Dr. and approached the museum from the south.

Waymo obviously needs to add highways to offer full metro area service. It's not a technical issue. They navigated 10 different 100 mile city/highway loops with zero interventions in 2009. Highways are easier than surface streets. But the liability is much, much higher. Tesla has a simple solution to the liability problem -- blame the driver. Of course that only works for consumer toys, not actual Robotaxis.

 

[heltok]

It's a fun debate to argue who is where in the race. No real definitions of what race and hard to judge where in any such race anyone is. Imo pretty pointless debate. Imo better just to state your beliefs around concrete future events and not argue or post some statements about the past and debate if they are true or not.

I have worked a bit with Mobileye as a customer. They are pretty fun, pretty phd:y with lots of math and they tend to project a lot of confidence. I am not super impressed by their progress over the years, as Tesla has run into the struggles of real world AI and hitting local maximums, so have Uber and Waymo. Mobileye has a large customer base for their products, imo Tesla has a huge advantage in that when they release Cybertruck or $25k car they can just slap HW4, AEB, Autopilot and FSD for a pretty small incremental sum. Ford are not as fortunate with F-150 Lightning who have to pay Mobileye/nVidia pretty large amounts and same holds for Toyota, VW etc. How much extra would you be willing to pay for Autopilot? I for sure value this >$2k/car.

Tesla choose the harder path, no Lidar, not using HD map for road markings etc. Waymo might be able to scale up maps, but updating maps will be tricky and you run into the problem of what to do when the map and the camera disagrees. Mobileye have more cars in their fleet, but are they as good as Alphabet with scaling maps? How long after a road has changed to you need to update the map? How can you be sure that a road has changed just because a camera disagrees with the map? Finding the edge cases here will be long march of nines. More complexity... It will be a tricky project, one that Tesla can ignore but instead have to solve the harder problem of vision. But once they have solved vision it will give them advantages in Optimus etc.

Anyway I think there is room in the market for several players and wish them all luck. I think that advancements in AI will lessen the time to play catch-up and in the end it will be the sum of all the parts that matter and I think Tesla has many of the parts with vertical integration from charging network, cars, AI, compute etc...

 

[insaneoctane]

So, I am wondering about Elon's comments on FSD during the shareholder meeting. He said he's currently going days without safety-related interventions. That is certainly not my experience. Is that most likely because he's got a different build, or his interpretation of safety-related intervention vs mine? I think FSD is a great product showing great potential, but his representation today doesn't match my personal observations...