We'll know Tesla is serious about robotaxis when it starts hiring remote operators.
See full article...
See full article...
On self-driving, Waymo is playing chess while Tesla plays checkers
- Thread starter JournalBot
- Start date
The Emperor of the Reply Guys must see to the welfare of his realm
Upvote
24
(25
/
-1)
Counter-Point: beginner drivers
Maybe AVs need to have a sign similar to those used by driver's schools for other drivers to see.
Upvote
13
(13
/
0)
Why do you love two huge corporations?
That's a serious question. How can you LOVE a corporation?
You said that with a a straight face? You don't think it's odd to LOVE a collection of people whose collective behavior doesn't give a fuck about you?
Loving any company is incredibly immature and not healthy at all. Please see a therapist. I'm not joking and not being sarcastic. It is literally bad for your health.
Upvote
42
(62
/
-20)
enlightened.doggo
Ars Scholae Palatinae
The only data advantage Tesla has is more pictures on the camera roll. Waymo vehicles have entire sensor arrays collecting far more detailed information about the world. State of the art ML algorithms for image recognition have error rates in the low percents. Even a 0.1% error rate could be deadly in a FSD context. The Tesla strategy will likely never produce safe FSD and instead remain a–to be fair, relatively advanced–driver assist.
Upvote
28
(33
/
-5)
Longmile149
Ars Scholae Palatinae
Upvote
29
(33
/
-4)
Excellent article. I'm on record as being deeply impressed by what Tesla have achieved but also being skeptical about how much further they can take it - this article's points about flat tyres and similar are highly relevant.
I get annoyed by both the Tesla-stans and the "but LIDAR!" comments (of which this thread contains two already).
It's abundantly clear that Waymo are playing in a different league to basically everyone else. Tesla are nowhere near. I thought Cruise were close for a while, but that turned out to be sadly wrong.
For LIDAR, there's nothing inherent about it that allows the problem to be solved - it can only ever be a factor, as with everything else. A large part of the reason that Tesla didn't go with it was cost - they simply would never have been able to fit it to all their cars like they could and can with cameras. What's interesting is that the cost of LIDAR has now decreased - and amortised across the life of a taxi, would it really be a factor? Especially compared to the support staff this article highlights.
Just one more factor in Waymo's rollout strategy's favour - how would either Waymo's or Tesla's approach work for issues in semi-remote freeways? That's a different level of problem to being in the city centre a few miles from a support base.
I get annoyed by both the Tesla-stans and the "but LIDAR!" comments (of which this thread contains two already).
It's abundantly clear that Waymo are playing in a different league to basically everyone else. Tesla are nowhere near. I thought Cruise were close for a while, but that turned out to be sadly wrong.
For LIDAR, there's nothing inherent about it that allows the problem to be solved - it can only ever be a factor, as with everything else. A large part of the reason that Tesla didn't go with it was cost - they simply would never have been able to fit it to all their cars like they could and can with cameras. What's interesting is that the cost of LIDAR has now decreased - and amortised across the life of a taxi, would it really be a factor? Especially compared to the support staff this article highlights.
Just one more factor in Waymo's rollout strategy's favour - how would either Waymo's or Tesla's approach work for issues in semi-remote freeways? That's a different level of problem to being in the city centre a few miles from a support base.
Upvote
17
(25
/
-8)
This timing of this announcement was only made to garner support for the upcoming (June 13) votes on Musk's pay package, whether Tesla will re-incorporate in Texas, and his brother Kimbal Musk's and Rupert Murdoch's son James board of directors positions.
While I wish Tesla were capable of unveiling an actual robotaxi on August 8, this will be yet another vaporware whose sole purpose was to serve to pump up the stock and allow Tesla to be used as a piggy bank.
Upvote
43
(47
/
-4)
Which is funny considering the CyberStuck manages to be $100k for a piece of junk. They could have easily fit LiDAR in the budget had they not, like, completely failed to engineer an actual truck. F150 Lightning is a superior vehicle in literally every way at half cost. What's that 50k even going to??
Ford could probably throw a couple LIRAR pucks on there and still come out cheaper.
Upvote
18
(24
/
-6)
Because the power to license drivers is devolved to the states, not the federal government. So each state has the power to do things differently.
Upvote
44
(44
/
0)
That was my thought when I read "City officials can erect a geofence to keep Waymo vehicles away from emergency scenes."
Absolutely the wrong mindset and approach. If there's a fire or some other emergency, the response crew has more immediate and important things to do than pull up some app, open the map, and draw up a restricted area. Not to mention situations can quickly evolve and that area could change with it. The cars need to able to handle the situations as they currently are.
Upvote
-4
(13
/
-17)
Upvote
15
(18
/
-3)
I doubt the geofence gets set by the firefighter on the scene. Presumably it's done by an SFFD dispatcher in an office somewhere. Even if you assume self-driving vehicles are working perfectly, it's still helpful to have a way to route them away from areas that get congested due to a big fire or car crash.
Upvote
37
(38
/
-1)
If by "wait" you mean "testing on public roads and killing people," sure.
Whatever you gotta tell yourself to avoid that awkward feeling of cognitive dissonance. Wouldn't want to utilize the incredibly meta cognitive abilities of your massive primate brain!
Upvote
8
(15
/
-7)
Easily solved, add cowctachers to the front of Firetrucks and Ambulances; both trucks are already made of base super duty structure. The driverless cars will be out of the way one way or another.
Then let the liability fall on the companies and charge the execs with whatever obstructing emergency vehicles is.
Side note - it's asinine that these mega companies get away with forcing emergency drivers to reroute - but HOLY SHIT if one of us low class citizens do, it's potential jailtime for us under a class 2 misdemeanor. Maybe not a one off incident (that would be a fine and points on my license); but being a repeat offender = potential jailtime+plus likely enough points on my license for a suspension.
Same thing should apply to the companies - execs held legally accoutnable and potential jailtime and they should acrue points against their license if they live ina State that uses a point system (like California or Florida). If enough points are acrued - not only does whate3ver exec lose their license but the company gets kicked out of the State for the duration. The fucking bullshit they get away with.
So glad there are laws for richy folks and corporations and different laws for us common folk.
Upvote
-8
(7
/
-15)
Or you know, no one will make auto taxis anytime soon. Even if one happens, it'll only be in very gated areas with perfectly mapped roads. The general purpose go anywhere autotaxi is still sci fi
Upvote
4
(8
/
-4)
Because machine intelligence operating in the physical world is interesting, I read this stuff, but most of my brain is still screaming in protest. We have way more than enough tech and knowledge to make a sensible, highly-automated transport network, operating on infrastructure that is smaller and cheaper, with astronomically improved safety and pollution.
The only real obstacles are a capitalist economy with its heels dug in on cars, and a public that is intolerant of long-term projects. Those aren't nothing, but all these $$$ developing pseudo-AGI in order to keep almost every mistake we made on transit alive, expensive, and generally detrimental to the human condition aren't nothing either. This society is broken and foolish.
The only real obstacles are a capitalist economy with its heels dug in on cars, and a public that is intolerant of long-term projects. Those aren't nothing, but all these $$$ developing pseudo-AGI in order to keep almost every mistake we made on transit alive, expensive, and generally detrimental to the human condition aren't nothing either. This society is broken and foolish.
Upvote
19
(23
/
-4)
Eh... taxis services could actually scale pretty hard. Lots of people only own cars because their is no other choice.
I currently take a 4 door gas powered car with a 300+ mile range, trunk, and 5 seats to and from work every day. I do not need a gas powered vehicle that big to get to work, but I do need a gas powered car that big to do other activities. If I could reliably have a single seat electric commuter car with a 50 mile range pull up when I want to go to work, and then I could reliably order a larger long range car when I need it, I wouldn't own a car, and the world would be better for it. I'd use a tiny fraction of what I use in my daily commute if I was in a single seater electric car, and I wouldn't bother to own a larger car for the rare times when I need them.
I know lots of people love their cars and live in places where they always want one one hand, but that REALLY isn't everyone.
Really, single seat electric commuter cars seem like a massive business around any major city; especially those with traffic problems that could dramatically benefit from suddenly having all of their commuter cars shrink in weight and size by over half and stop emitting CO2 and other pollutants. You get more road capacity, less pollution, and fewer resources being used to make large multipurpose vehicles that sit idle all day in a parking lot.
Taxis services; especially small single seat electric commuter cars, could easily scale to be one of the biggest new markets that is currently left untapped, and it would be a massive environmental victory.
Last edited:
Upvote
13
(18
/
-5)
@peterford ...
(1) have you not been paying attention to the real world reviews by Cybertruck owners since they started being distributed or have you been living under a rock for the past several years ? The Cybertrucks are garbage and just shy of bricks on wheels that rust.
(2) posting "Micro$oft" instead of the actual company name enough times will get you warnings and suspension from Ars. That's against TOS; heads up.
(3) as for someone being a subscriber and having an account that goes back at Ars since near the beginning of the site does not mean that reader/commentor should be held to some higher mythical imaginary standard you perceive as relevant to jack shit. By even posting that comment YOU need to grow TF up and move on with your lfie. @Emon is a made up screen name on the interwebs that you don't know and should not have a personal grudge against and not give a second thought to after you finish reading an article. Post your comment relating to the article, maybe have a polite exchange over the topic at hand and move TF on. It's not worth the pissing contest.
Upvote
12
(25
/
-13)
It's been downvoted because the OP fundamentally missed the main point of the article: Waymo is succeeding by solving the easy problems first and incrementally moving on to the harder problems. There's no law that says Waymo has to simultaneously have cars in Phoenix AND Duluth. If Waymo can make it work in Phoenix, but it doesn't work in Duluth, that's OK. They may even have a viable business by solving the easy cases.
So maybe try thinking for yourself instead of just blindly rooting for Team Musk and getting all pissy when legitimate criticisms are discussed.
Upvote
52
(57
/
-5)
Upvote
11
(12
/
-1)
There's a reason they do this in San Francisco, and it's not just because the corporate offices are there: FSD becomes almost impossible when there's snow on the ground (or, in many places, when snow is on EVERYTHING, including signs). YOLO-type inference without LIDAR will see NOTHING. The upper half of the US will lose FSD for months at a time. I feel like they dodge this question.
Upvote
-7
(7
/
-14)
This isnt how the real world works though. Many companies will do training with emergency services (ES) to help them react to new threats and issues. My company which builds windfarms will train with ES on how to respond to calls about our turbines and substations. This is common and ES needs to adjust to new real world changes they cannot stay stuck in one point in time.
Upvote
33
(34
/
-1)
Check your bingo cards, we've got the "this is bullshit just driving engagement" guy who is here, engaging with the content he so desperately wants everything to think is pandering nonsense.
Upvote
14
(20
/
-6)
I'm surprised at the lack of mention of the sensor differences. Architecturally, I'm confident Google is very far ahead here; the people I know on the software teams are researchers at the forefront of developing the underlying technology, and they've been using neural nets for a decade or so now.
But you do need lots and lots of high-quality data, with lots and lots of edge cases, in order to train these models and increase confidence into the several-nines levels. Tesla has a fundamental, self-inflicted problem there.
They chose to remove the non-camera (lidar and radar) sensors, with the idea that "if humans can do it with just their eyes, so can cars".
It's a very shallow argument. For one thing, humans can't do it well and kill other humans frequently. That's accepted for historical reasons, but that level of fatalities and injuries would never be allowed by regulators for a new technology. Just look at how little leeway we give airplanes.
The other problem with that argument is that humans have insanely good eyeball cameras. The hdr processing is so significantly better than anything that would fit on a car at a reasonable price as to be incomparable. And even with that, it's really just unsafe to drive in certain whiteout, fog, and sunlight conditions, not to mention dark nights on poorly-lit roads.
Google has a ton of useful data in these scenarios, Tesla has a bunch of noise. These are the long-tail problems that take years to solve, and even then are only solvable with useful data. Tesla will never get that data without sensors that can operate in difficult conditions. No radar or lidar also means no data advantage over humans in terms of precision at a distance; with radar you get speed/distance data and with lidar you get point cloud data.
And I'm not sure where they are at now, but last I heard they did not have a dedicated team generating synthetic data because they had more data than they could use (due to data processing architecture and resources) without it. Problem is, synthetic data is far more efficient because each point is a scenario you care about, whereas the filtering to be done on real-world data from millions of cars is only tractable when you have a software company like Google as a backstop for these pipelines (and even there, the costs would raise eyebrows; they already do at the 10k/day level).
All this to say, I think "four years behind" in terms of measurable safety (miles without injury in a fully driverless scenario) is a huge overestimate. Unprotected lefts are a really hard problem (for humans as well!), but they're a problem that's largely compute-constrained, which is just naturally easier as processors get faster. I'm curious how long it will take before there's a serious unprotected left accident from a Tesla when the sun is close to the horizon and the cameras are useless -- fsd might just proactively disengage in that scenario, in which case I'd say that already shows how far they have to go to catch up to 2018 Waymo.
Tesla is a long way off, and I'm very dubious that they have a real path forward with the current hardware decisions -- something that will be hard to go back on thanks to the extremely fragile ego at the helm.
But you do need lots and lots of high-quality data, with lots and lots of edge cases, in order to train these models and increase confidence into the several-nines levels. Tesla has a fundamental, self-inflicted problem there.
They chose to remove the non-camera (lidar and radar) sensors, with the idea that "if humans can do it with just their eyes, so can cars".
It's a very shallow argument. For one thing, humans can't do it well and kill other humans frequently. That's accepted for historical reasons, but that level of fatalities and injuries would never be allowed by regulators for a new technology. Just look at how little leeway we give airplanes.
The other problem with that argument is that humans have insanely good eyeball cameras. The hdr processing is so significantly better than anything that would fit on a car at a reasonable price as to be incomparable. And even with that, it's really just unsafe to drive in certain whiteout, fog, and sunlight conditions, not to mention dark nights on poorly-lit roads.
Google has a ton of useful data in these scenarios, Tesla has a bunch of noise. These are the long-tail problems that take years to solve, and even then are only solvable with useful data. Tesla will never get that data without sensors that can operate in difficult conditions. No radar or lidar also means no data advantage over humans in terms of precision at a distance; with radar you get speed/distance data and with lidar you get point cloud data.
And I'm not sure where they are at now, but last I heard they did not have a dedicated team generating synthetic data because they had more data than they could use (due to data processing architecture and resources) without it. Problem is, synthetic data is far more efficient because each point is a scenario you care about, whereas the filtering to be done on real-world data from millions of cars is only tractable when you have a software company like Google as a backstop for these pipelines (and even there, the costs would raise eyebrows; they already do at the 10k/day level).
All this to say, I think "four years behind" in terms of measurable safety (miles without injury in a fully driverless scenario) is a huge overestimate. Unprotected lefts are a really hard problem (for humans as well!), but they're a problem that's largely compute-constrained, which is just naturally easier as processors get faster. I'm curious how long it will take before there's a serious unprotected left accident from a Tesla when the sun is close to the horizon and the cameras are useless -- fsd might just proactively disengage in that scenario, in which case I'd say that already shows how far they have to go to catch up to 2018 Waymo.
Tesla is a long way off, and I'm very dubious that they have a real path forward with the current hardware decisions -- something that will be hard to go back on thanks to the extremely fragile ego at the helm.
Upvote
26
(28
/
-2)
EDIT: It's one thing to advise/train ES on new, fixed, structures in their operating area that may pose special problems/threats/whatever. It's quite another to expect ES and roadworkers etc. to have to deal with any number of different AVs, each with their own disparate way of having ES handle them.
No, it is absolutely unreasonable to have as many different systems as there are companies rolling out AVs... In an emergency, do you want first responders responding, or working out which marque of AV happens to be around (or worse, marques) - then remember which app, or which number to call, or what esoteric hand signals to make, or which beacon device to deploy. (Or dispatchers loading umpteen different apps/sites/etc. to cover the field of all possible AVs that their responders might encounter).
No - there needs to be one standard that all AVs must respond to.
Last edited:
Upvote
1
(7
/
-6)
I'll never get into a driverless Tesla taxi.
At least, not in the next X years.
At least, not in the next X years.
Upvote
-2
(4
/
-6)
If there is a taxis service that will send a small, single seat, electric car that is cheaper and more efficient that my general purpose gas powered car, it's not operating in my area.
The point of a taxis service is that if you cut out the driver, you can make a very small and efficient car perfect for a commuter that runs on a battery. I'd happily take a single seat commuter car instead of buying a large gas powered car that sits in a parking lot for 98% of its existence.
Hell, just wander into any city if you are confused on this point. Uber only makes ordering a taxi easier without introducing any real cost savings, and that alone was enough for Uber to suddenly become a large portion of city traffic and cause more urban dwellers to ditch their cars.
If you can make taking a single seat electric taxi to work cheaper than taking your 4 door gas powered car, you just tapped into a business a few tens of millions strong in the US, and a hundreds of millions (billions?) strong world wide.
Upvote
10
(13
/
-3)
Good thing Tesla has said single seater, cheap EV.
Oh they don't? Well, I'm sure they can cook one up.
What? They fired their entire new product team?
I think the solution to the problem you're describing, urban residents needing to move around within that urban environment, already exists. It's called public transit.
Upvote
15
(22
/
-7)
Drive for 10 minutes in Massachusetts, and you'll know why we're all begging for non-human drivers in cars...
In all seriousness, THAT, for me at least, is the dream here: drivers that will ALWASY follow the rules, be MUCH safer than humans, and be able to bring travel into a place where it can happen much faster and safer than with humans involved. You've seen automation lines, where the robots are doing things SO much faster than humans ever could, because of their capabilities? That's what I'm hoping for.
In a perfect world, we'd have NFC-equipped cars with communication standards, along with a identifying road network AND hyper-localized condition reporting, but we will never live there, so I'm just hoping for something in the right direction.
Upvote
33
(35
/
-2)
Merica.
We need to create a list of pregnant women to track them to protect the fetus.(yes actually suggested by a law maker)
If the pregnant person gets run over by a beta testing car? Acceptable sacrifice to the capitalism.
Upvote
15
(18
/
-3)
- Status