[HN Gopher] What's the data Waymo car collects driving around SF...
___________________________________________________________________
What's the data Waymo car collects driving around SF all day?
Can't really understand what are all those sensors (e.g. why Tesla
ain't got them) and what's the diff from any other self-drivingish
company?
Author : talboren
Score : 7 points
Date : 2023-02-12 19:42 UTC (3 hours ago)
| BoorishBears wrote:
| I work in the AV field: Across all the real competitors (not
| Tesla) sensors are things like LIDAR, Cameras (visible spectrum
| and infrared), radio radars (yes radar already includes radio,
| but for laypeople LIDAR is confused with radar), microphones.
|
| -
|
| Tesla doesn't have them for two different reasons...
|
| Claimed reason: Humans have eyes and make driving work, so cars
| won't need more than that.
|
| Real reason: Tesla wants to make money by selling the package to
| the public today. It needs to be cheap enough to be mass produced
| today, and sold at an accessible cost.
|
| And it doesn't really matter to them if the current stack will be
| enough because regardless of what they claim, they're not averse
| to screwing over current owners and releasing a new stack and
| abandoning people who paid for an untenable one (note: I've been
| saying this since before they did exactly this with HW4 a few
| week ago...)
|
| The sensor stacks alone the actual AV players use are worth more
| than the average Tesla. That's why they're mostly initially
| aiming for ridesharing, where the company can own the vehicle.
| 988747 wrote:
| > The sensor stacks the actual AV players use are worth more
| than a vehicle.
|
| Not sure about that. LIDAR prices dropped to 1-2k USD.
| BoorishBears wrote:
| That's not how that works.
|
| Even if tomorrow morning a $1 LIDAR came out, it'd be a long
| time before you saw it in a vehicle and the costs around the
| integration into their stacks (even when amortized across the
| number of AVs being produced) would still leave them
| completely untenable for individual ownership.
|
| And if your thought is "well just increase the number of AVs
| produced", the designs being used also aren't aiming to be
| mass producible on the scale of a Tesla because again,
| they're not artificially hobbling their approaches with that
| in mind.
|
| -
|
| And it's not just LIDAR. Specialized sensors that can't even
| be sourced in those kinds of numbers, compute resources,
| infrastructure _for_ those resources (both on and _off_
| vehicle).
|
| Profitably producing cars is a game of pennies completely
| opposed to what AVs require. Trying to integrate solving AVs
| into that framework is pretty much resigning yourself to not
| doing either particularly well... unless you're only paying
| lip service to one side: hence why I use the terminology
| "real competitors".
|
| And to be clear, I don't think Tesla doesn't know better, or
| is incapable of _actually_ trying. It 's a very intentional
| choice. Traditional manufacturers would do terrible
| _terrible_ things to extract $7500 additional dollars for a
| software switch (they 've done that for a lot less!). EVs and
| AVs are a chance for them to dip into that, and Tesla's most
| impressive innovation has been getting ahead of the curve on
| it.
| joshu wrote:
| I think people don't realize how tough NHV is on
| lidars/sensors/etc - cars are a really difficult
| environment for computers and so on.
| foxandmouse wrote:
| Hey! a question I know something about :)
|
| Waymo is banking on Lidar being the solution to autonomous
| vehicles, the reason tesla doesn't use them is that Elon thinks
| Lidar won't work and that we need Ai for driverless cars (or more
| accurately a highly advanced AGI).
|
| At first, I thought Elon was being too radical, he's known for
| overcorrecting for the sunk cost fallacy, but after more time, I
| share his view.
|
| Personally, I think Waymo is a joke. Just because they came from
| Google they think they're too big to fail before even having a
| product for market. They've already lost 170 billion in just a
| year an a half. The CEO got charged with stealing tech and
| selling it to tesla, then got sued by tesla when he made a
| competitor to the stolen tech he sold them.
|
| Take it as you will, personally I'm pretty bias against google
| these days, Pichai has ruined the company for me. I feel he's
| turned google into Microsoft, there's nothing wrong with
| Microsoft, but we already have one of those, it was nice when we
| also had a google.
| edude03 wrote:
| Lidar point clouds, high resolution imagery and driver inputs
| mostly. The high level idea being you want to train the neural
| network so that when it sees x on the sensors it should do y
| thing the human did.
|
| As for why other self driving cars don't have lidar that's... A
| contentious topic. The argument for lidar is that it captures the
| environment in 3d unlike cameras that only capture in 2d and thus
| you can save on computation trying to convert 2d images into a 3d
| scene. On the other hand lidar is very expensive, has limitations
| like not being able to see through rain and maybe most
| importantly is believed to be unnecessary since "humans drive
| with just their eyes".
|
| My personal belief is its a matter of cost vs time to market -
| lidar makes locatization and object recognition a little bit
| easier and if for example it saves 1 year of development time at
| a cost of 100k per car produced it's worth it if you're only
| making a small fleet of cars anyway.
___________________________________________________________________
(page generated 2023-02-12 23:01 UTC)