-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
I'm not really sure where to put this as it could go in a few different threads. Please feel free to move if needed.
This was a slide from Autonomy Day, 2019. I want to focus on the miles....16mph x 16 hours / day or ~112 hours per week (let's ignore the logic that the demand for robotaxis is not constant for 24 hours/day, the vehicle will need time to charge, etc.).
Let's go to the latest Earnings Call (from yesterday):
Did Elon just *HALVE* the projected "utility hours" of a Tesla robotaxi or am I misunderstanding?
Additionally...
Level 4 does still allow for human override as an option, but does someone want to explain how a robotaxi network would work where a person has to sit in the driver seat to intervene? Who's assuming the risk in that scenario? Would Tesla even roll out robotaxis where a rider has to constantly monitor the vehicle?
This was a slide from Autonomy Day, 2019. I want to focus on the miles....16mph x 16 hours / day or ~112 hours per week (let's ignore the logic that the demand for robotaxis is not constant for 24 hours/day, the vehicle will need time to charge, etc.).
Let's go to the latest Earnings Call (from yesterday):
Did Elon just *HALVE* the projected "utility hours" of a Tesla robotaxi or am I misunderstanding?
Additionally...
If you assume that with no driver, an intervention or disengagement results in a crash....at ~10,000 miles per disengagement every single car in the fleet is crashing about once/year at "normal" 10k miles/year drive time. If Tesla is anticipating a robotaxi will drive more like 90k-100k miles...at 90k miles per disengagement that's still one crash per vehicle in the fleet per year. That's not remotely insurable. And Elon is touting "a very small number of interventions per mile" as good progress and evidence that level 4 FSD is close (this year). I mean, come on. Just do some math. On average, there is one vehicle crash in the U.S. for every ~530k miles (6m vehicle crashes per year, 3.2 trillion vehicle miles traveled).
Level 4 does still allow for human override as an option, but does someone want to explain how a robotaxi network would work where a person has to sit in the driver seat to intervene? Who's assuming the risk in that scenario? Would Tesla even roll out robotaxis where a rider has to constantly monitor the vehicle?
Last edited:
Daniel in SD
(beta)
Disengagement numbers are only loosely related to the collision rate. For example in Waymo's autonomous testing they found that 99.9% of their disengagements would not have resulted in a collision had they not occurred (https://storage.googleapis.com/sdc-...Waymo-Public-Road-Safety-Performance-Data.pdf). You have to go back and simulate what would have happened.
The Disengagement Myth
In a few weeks the California DMV will release disengagements data from Cruise and other companies who test AVs on public roads. This data…
medium.com
Or maybe FSD Beta would have swerved back had you not intervened?Of course there is zero chance that Tesla achieves human performance this year (1 severe collision (>12mph) per 2 million miles or so according to their safety report).
Obviously the manufacturer of the system will be responsible if the collision is caused by a fault in the system. A robotaxi can't required a driver!
Knightshade
Well-Known Member
Over-reading might be more accurate than misunderstanding.
he just tossed out a number off the cuff to point out one RT would in terms of driving hours replace many human driven cars. Actual # of hours will vary modified by a bunch of factors.
Assuming owner doesn't need it for personal use, and demand were always there, that'd be more than either the 12 Elon cited or the 16 on the slide you mention- nearer 20 most likely if there's a supercharger in the area.
But some owners WILL want it for personal use some of the time...and some owners will live someplace demand isn't always 24/7, or a supercharger is less proximate at the moment- so they'd be running less than 20- how much less will vary.
After a certain point the difference isn't all that important for most discussion.... but obviously it could run for however many hours the owner wishes, modified by the demand in their specific area, minus time spent charging.
Level 4 allows for a driver to disengage, but Level 4 cannot require a driver to be in the driver's seat and alert. Level 4 must be capable of bringing the car to a safe stop in a safe place any time it cannot handle a situation by itself. A robotaxi must be able to operate entirely without a driver. It can be geofenced and therefore need not be Level 5, but within its area it cannot require a driver. Otherwise it's not a robotaxi: It's just a ride-share with driver-assist features.
By definition, a car that requires a driver to make the decision to disengage is Level 2 or lower.
Elon is just being Elon, promising Level 4 "real soon now" when in fact the technology for wide-area, non-geofenced Level 4 is years away.
Not every driver-initiated disengagement would have resulted in a crash. When I'm using EAP I disengage very pro-actively any time I'm not entirely confident of the car's ability to stay safe. Some of those times the car would probably do just fine. So the driver-initiated disengagement rate is not what the crash rate would be. But the numbers are still way far away from ready for L4.
That said, I do not believe Tesla will initiate Level 4 before it is safer than human. Or to put it another way: Elon's proclamations are total bull, but Tesla is, I believe, fully committed to safety.
Updates from the original source since:
- April 2021
Gating factor is achieving & proving higher safety with pure vision than with vision+radar. We are almost there. FSD Beta V9.0 will blow your mind.
Source - April 2021
True. Anyone paying attention to the rate of improvement will realize that Tesla Autopilot/FSD is already superhuman for highway driving & swiftly getting there for city streets.
Source - May 2021
We had to focus on removing radar & confirming safety. That release goes out next week to US production. Then a week or two to polish pure vision FSD & v9 beta will release. Difference between v8 & v9 is gigantic.
Source - May 2021
I think we’re maybe a month or two away from wide beta. But these things are hard to predict accurately. The work we had to do for pure vision driving was needed for FSD, so much more progress has been made than it would seem.
Source - May 2021
Pure vision Autopilot is now rolling out in North America. There will be an update of this production release in 2 weeks, then FSD beta V9.0 (also pure vision) a week later. FSD subscription will be enabled around the same time.
Source - January 2022
I will be shocked if we don't achieve FSD safer than a human this year
Source
Indeed, Elon talks nonsense when it comes to timelines .. this is hardly news. Oh, and politicians lie as well. And that frozen food pie? Doesnt look anything like the picture on the box.
The point is, what Elon says about timelines is garbage .. but at the end of the day, he does seem to inspire people to deliver. Tesla DID put EVs on the map, and in a huge way. SpaceX DOES regularly land and re-use rockets. And Tesla IS in the process of putting ADAS cars in the hands of anyone who can afford them. Notice anything here? First, they are all things people "in the know" scoffed at and said were BS, and they are all HARD problems, in the sense no-one has ever done anything like it before, or even knew it could be done.
Personally I think Elon is a pretty awful guy, but he does push people to get "impossible" things done, even if takes way longer than his eternal optimism predicts.
It’s in the Washington Post article, it happened in April 2019:
“In April 2019, at a showcase dubbed “Autonomy Investor Day,” Musk made perhaps his boldest prediction as Tesla’s chief executive. “By the middle of next year, we’ll have over a million Tesla cars on the road with full self-driving hardware,” Musk told a roomful of investors. The software updates automatically over the air, and Full Self-Driving would be so reliable, he said, the driver “could go to sleep.”
Investors were sold. The following year, Tesla’s stock price soared, making it the most valuable automaker and helping Musk become the world’s richest person. Full Self-Driving followed Autopilot, which was launched in 2014 and went on to allow cars to navigate highways, from steering and changing lanes to adjusting speed. Full Self-Driving aimed to bring those capabilities to city and residential streets, a far more difficult task.”
When I bought my 2018 Model 3, Tesla said it had all the necessary hardware for full self-driving, which would allow it to operate as a robotaxi (to distinguish from merely "can mostly drive itself while you monitor your car with your hands on the wheel) once the software was available. Turned out it didn't. I think there were two significant hardware updates that would need to be applied to run current "FSD" software on my car, and that is still nowhere near robotaxi-capable.
I'll eat my hat if the Teslas being sold today, with today's hardware, are ever capable of operating as robotaxis.
Of course, I don't think Tesla is making that promise for new Teslas being sold today. Now "FSD" just means that you can drive it in the city and you won't have to intervene very often. Though I would hate to be anywhere near when someone tries to operate "FSD" on South Kihei Road, which is the only road that runs the length of Kihei, other than the highway.
Tesla still has no idea what hardware will be needed to run the as-yet unfinished software.
I'll eat my hat if the Teslas being sold today, with today's hardware, are ever capable of operating as robotaxis.
Of course, I don't think Tesla is making that promise for new Teslas being sold today. Now "FSD" just means that you can drive it in the city and you won't have to intervene very often. Though I would hate to be anywhere near when someone tries to operate "FSD" on South Kihei Road, which is the only road that runs the length of Kihei, other than the highway.
Tesla still has no idea what hardware will be needed to run the as-yet unfinished software.
Similar threads
