Latest / Elon Musk Thinking / Elon Musk said At Teslas Autonomy Day.

Transcript

Elon Musk said At Teslas Autonomy Day. · Elon Musk Thinking

0:00
I can add something. The general principle here is that any part of this
0:03
could fail and the call will keep driving. So you could have cameras fail,
0:07
you could have power circuits fail, you could have one of the Tesla
0:11
full self driving computer chips fail, car keeps driving. The probability of this
0:15
computer failing is substantial, lower than somebody losing consciousness. That's the key metric,
0:20
at least in order of magnitude. At first, it seems improbable.
0:23
How could it be that Tesla, who has never designed a chip before,
0:26
would designed the best chip in the world. But that is objectively what has
0:30
occurred. Not best by a small margin, best by a huge login.
0:34
It's in the cars right now. Old Tesla's being produced right now have this
0:37
computer. We switched over from the Nvidious solution for sn X about a month
0:43
ago, and you switched over Model three about ten days ago. All cars
0:47
being produced have the have all the hardware necessary, compute and otherwise for full
0:52
self driving. I'll say that again. Old Tesla cars being produced right now
0:56
have everything necessary for full self driving. All you need to do is improve
1:00
the software and later today you will drive the cars with the development version of
1:04
the improved software and you will see for yourselves. We finished this design like
1:08
maybe one and a half two years ago and began design of the next generation.
1:12
We're not talking about the next generation today, but we're about halfway through
1:15
it. All the things that are obvious for our next generation chip. We're
1:18
doing the strategy here and started basically three level three years ago was designable to
1:23
computer that is fully optimized and aiming for full self driving. Software that is
1:29
designed to work specifically on that computer and get the most out of that computer.
1:33
So you have tailored hardware that is a master of one trade. Self
1:36
Driving and Video is a great company, but they have many customers and so
1:41
as they apply their resources, they need to generalize solution. We care about
1:45
one thing, self driving, so that it was designed to do that incredibly.
1:49
The software is also designed to run on that hardware incredibly well. And
1:53
the combination of the software in the hardware I think is unbeatable. We' we're
1:56
going to explained to you today is that lighter is a fulls errand and anyone
2:00
relying on light R is doomed. Doomed expensive sensors that are unnecessary. It's
2:05
like having a whole bunch of expensive epeties, but one appendix is bad.
2:08
We only want to put a whole bunch of them. That's ridiculous. You'll
2:12
see if you're driving for an hour in a city and you had a solution
2:15
hypothetically that it was a kilo, what you'd lose four miles a Model three,
2:20
So if you're only going, say twelve miles an hour, then that's
2:23
there would be a twenty five percent impact on range in city. It's basically
2:27
power is the power. The power of the system has a massive impact on
2:30
city range, which is where we think most of the robotaxi market will be.
2:35
So power is extremely important. We don't want to spalok too much about
2:37
the next generationship, but it's it'll be at least, let's say, three
2:40
times better than the current system. And I think if if somebody started today
2:46
and they were really good, they might have something like what we have right
2:49
now in three years, but in two years we'll have something three times better.
2:53
The thing that's I think a very powerful sustainable advantage for us is the
2:57
fleet. Nobody has the fleet. Those weights are constantly updated and improved based
3:00
on billions of miles driven. Tesla has one hundred times small cars with the
3:05
full self driving hardware that everyone else combined we have by the end of this
3:08
quarter, will have five hundred thousand cars worth the full eight camera set up
3:12
fall ultrasonic. Someone will still be on hardware too, but we still have
3:15
the data gathering ability, and then by a year from now, we'll have
3:19
over a million cars with full self driving computer hardware everything. It's just a
3:23
massive data advantage. It's similar to how like the Google search engine has a
3:27
massive advantage because people use it and people are programming effectively programmed Google with the
3:32
queries and the results. The simulator, we have quite a good simulation too,
3:38
but it just does not capture the long tail of weird things that happen
3:40
in the real world. If the simulation fully captured the real world, that
3:44
would be proof that we're living in a simulation. I think, yeah,
3:46
it doesn't. I wish, but simulations do not capture the real world.
3:51
The real world is really weird and messy. You need the cars on the
3:53
road simulation. You're fundamentally grading your own homework. If you know that you're
3:58
going to simulate it, Okay, you can definitely solve for it. But
4:00
as Andrew saying, you don't know what you don't know. The world is
4:03
very weird and has millions of corner cases. And if somebody can produce a
4:09
self driving simulation that accurately matches reality, that in itself would be in a
4:14
monumental achievement of human capability. They can't. There's no way. Everyone's training
4:19
the network all the time, is what it amounts to. Whether autopoler is
4:24
on orall. The network is being trained. Every mile that's driven for the
4:27
car that's hard or tour above is training the network. The crazy thing is
4:30
the network is predicting paths it can't even see with incredibly high accuracy. It
4:35
can't see around the corner, but saying the probability of that curve is extremely
4:40
high. So that's the path, and it nails it. You will see
4:42
that in the cars today. We're going to turn on augmented vision so you
4:45
can see the lane lines and the path predictions of the cars overlaid on the
4:48
video. The car is an inference optimized computer. But we do have a
4:54
major program at Tesla, which we don't have enough time to talk about today,
4:57
called Dojo. That's a superpower full training computer. The goal dojo will
5:01
be to be able to take in vast amounts of data and train at a
5:05
video level and do unsupervised massive training of vast amounts of video with the dojo
5:11
program dojo computer. That's for another day. We're just being more conservative right
5:15
now, and then as we gain a higher higher confidence will allow users to
5:19
select a more aggressive mode. That'll be up to the user. But in
5:23
the more aggressive modes, in trying to merge in traffic, there is a
5:27
slight no matter how many, there's a slight chance of like vendor bender,
5:30
not a serious accident, but you basically will have a choice of do you
5:33
want to have a non zero chance of a vendor vender on freeway traffic,
5:38
which is unfortunately the only way to navigate it LA traffic. The car can
5:42
operate if it's completely disconnected from the fleet. It just it uploads the training
5:46
that's better and better as the free gets better and better. So simply if
5:50
you're just connected from the fleet from that point onwards, it would stop getting
5:54
better, but it would still function fine. The computer power in the full
5:57
Sell driving computer is incredible, and it maybe we should have mentioned that if
6:01
it had never seen that road before, it would still have made those predictions
6:05
provided it was a road in the United States. You're all going to dump
6:09
lighter. That's my prediction. Walk my words. I should point out that
6:12
I don't actually superheate lighter as much as may sound. But at SpaceX,
6:15
SpaceX Dragon uses light r to navigate to the space station and dock. Not
6:19
only that, we SpaceX developed its own lighter from scratch to do that,
6:25
and I bear ahead of that effort personally because in that scenario, lighter makes
6:29
sense. And in cars it's freaking stupid. It's expensive and unnecessary, and
6:33
as Andrew was saying, once you saw a vision, it's a worthless so
6:36
you have expensive hardware that's worthless on the car that we do have a forward
6:41
radar which is low cost and is helpful, especially for occlusion situations. So
6:46
if there's like fog or dust or just snow, the radar can see through
6:48
that. If you're going to use active photon generation, don't use visible wavelength
6:54
because once with passive optical you've taken care of all visible way stuff. You
6:58
want to if you want, I want to use a wave link that is
7:00
inclusion penetrating like radar. So liner is just active photon generation in the visual
7:04
spectrum. If you're going to do active photon generation, do it outside the
7:09
visual spectrum in the radar spectrum. So like at three point eight millimeters versus
7:12
four hundred one hundred antime, we're going to have much better occlusion penetration.
7:15
And that's why we have a forward radar, and then we also have twelve
7:18
ultrasonics for near field information in addition to the eight cameras and the forward radar.
7:25
You only need the radar and the forward direction because that's the only direction
7:27
going real fast. So we've gone over this multiple times, always sure we
7:30
have the right sense of suite shure we add anything more. No, we
7:32
have four hundred and twenty five thousand cars with hardware tuned beyond, which is
7:36
means they've got all eight cameras right, the radar and ultrasonics, and they've
7:41
got at least a video computer, which is enough to essentially figure out what
7:45
information is important what is not, compress the information that is important to the
7:48
most salent elements and uploaded to the network for training. So it's a massive
7:53
compression of real world data. I suppose it could possibly be used for something
7:57
besides self driving. We've been too focused on self drive. So as we
8:01
get that really nailed, maybe there's going to be some other use for millions
8:05
and then tens of millions of computers with hardware three or full sub driving computer.
8:09
Yeah, maybe there would be. It could be. Maybe there's some
8:11
sort of eight angle here as possible. There are a number of important signals
8:13
as under saying lanelines are one of those things, But one of the most
8:16
important signals is drive space. So what is drivable space and what is not
8:20
drivable space? And what actually really matters the most is drivable space more than
8:24
lanelines. And the prediction of driveable space is extremely good, and I think,
8:30
especially after this upcoming winter, will be incredible. It will be like,
8:33
how could it possibly be that good? That's crazy. It's extremely important
8:37
that things not be rigidly tied to GPS because GPS eric can vary quite a
8:41
bit, and the actual situation for a road can vary quite a bit,
8:48
so that reconstruction that could be a detour, and if the car is using
8:52
GPS as primary, this is a real bad situation. Is asking for trouble.
8:54
It's fine to use GPS for like tips and tricks, so you can
8:58
drive your home neighborhood better than a neighborhood in some other country or some other
9:01
part of the country. So you know your own neighborhood well and you use
9:05
like the knowledge of your neighborhood to drive with more confidence, to maybe have
9:11
counterintuitive shortcuts and that kind of thing. But the GPS overlay data should only
9:16
be helpful, but never primary. If it ever primary, it is a
9:18
problem. High precision GPS maps and lanes are a really bad idea. The
9:24
system becomes extremely brittle, so any change like this might any change to the
9:28
system makes it it can't adapt. So if it locks onto GPS and high
9:33
precision lane lines and does not allow a vision override, In fact, vision
9:37
should be the thing that does everything, and then like lane lines are a
9:41
guideline, but they're not the main thing. We briefly balked up the tree
9:45
of high precision lane lines and then realized that was a huge mistake and reversed
9:48
it out. It's not good. There's three steps to self driving, because
9:52
it's being future complete. Then there's being future complete to the degree that where
9:56
we think that the person in the car does not need to pay attention and
9:58
then there's at a reliability level, where we've also convinced regulators that is true.
10:03
So it's there's three levels. We expect to be set feature complete and
10:05
self driving this year, and we expect to be confident enough from our standpoint
10:09
to say that we think people do not need to touch the wheel look out
10:13
of the window sometime probably around a second quarter of next year, and then
10:18
we start to expect to get regulatory approval at least in some jurisdictions for that
10:22
towards the end of next year. That's roughly the timeline that I expect things
10:26
to go on, and probably for trucks, the tuning will be approved by
10:30
regulators before anything else, and you could have, maybe if you're a long
10:33
haul doing long haul freight, you could have one driver in the front and
10:37
then have four semis trailing behind in a platooning manner. And I think that
10:41
probably the regulators will be quicker to approve that than other things. If in
10:45
order to have a self driving car or robotaxi, you really need redundancy throughout
10:48
the vehicle at the hardware level. Starting in as it was October twenty sixteen,
10:52
all cars made by Tesla have redundant power steering, So we've redundant motors
10:58
and the power steering, so any one failure of if I'm the motor fails,
11:01
the car can still steer. All of the power and data lines have
11:03
redundancy, so you can sever any given power line or any data line and
11:07
the call will keep driving the auxiliary power system. Even if the main pack
11:11
you lose complete power in the main pack, the car is capable of steering
11:15
and braking at using the auxiliary power system, so you can completely lose the
11:18
main pack and the car is safe. The whole system, if from a
11:22
hardware standpoint, has been signed to be a robot taxi since basically October twenty
11:26
sixteen, so when we roll that hardware autopowered version two, but we do
11:31
not expect to upgrade cars made before that. We think it would actually cost
11:35
more to make a new car than to upgrade the cars, just to give
11:39
you a sense of how hard it is to do this. Unless it's designed
11:41
in, it's not worth it. So we've gone through the future of self
11:43
driving. Whereas it's hardware, it's vision, and then there's a lot of
11:48
software, and there's the software problem here should not be Minimi. It's a
11:52
massive software problem that managing vast amounts of data, training against the data,
11:56
how do you control the car based on the vision. It's a very difficult
11:58
software problem. So going after going over just like teslim massive plan. Obviously,
12:03
we've made a bunch of forward looking statements, as they call it.
12:05
And but let's go through some of our other forward looking statements that we've made.
12:09
No way back when we created the company, we said we build the
12:11
Tesla Road, Sir. They said it was impossible and that even if we
12:15
did build it, nobody were buy it. It was like universal opinion was
12:18
that building an electric car was extremely dumb and would fail. I agree with
12:22
them that probably failure was high, but that this was important. So we
12:26
built the Tesla Roads, going into production in two thousand and eight and shipping
12:30
that car it's not collector's item. Then we would build a more affordable car
12:33
with the Model S. We did that. Again, we were told that
12:35
it's impossible. I was called a fraud and a liar and it's not going
12:39
to happen. This is all untrue, Okay. Famous last words now is
12:43
we were into production with the Model S and twenty twelve exceeded all expectations.
12:46
There is still in twenty nineteen no car that can compete with the Model s
12:52
of twenty twelve. It's seven years later. Were waiting, so we'd build
12:56
on an affordable car, maybe highly affords affordable, more affordable. With the
13:00
Model three we built the Model three, we're in production, said we'd get
13:03
over five thousand cars a week for Model three. At this point, five
13:05
thousand cars week is a walk in the park for us. It's not even
13:09
hard. So we do large scale solar, which we did through the Sol
13:11
City acquisition, and that we develop into play the solar roof, which is
13:15
going really well. We're now on version of three of the Solo tail roof,
13:18
and we expect to will out production of the Solo Tawer roof significantly later
13:22
this year. I have it on my house and it's great. And I
13:26
make the power roll and the power pack, and we made the power role
13:30
and power pack. In fact that the power pack is now deployed in massive
13:33
grid scale utility systems around the world, including the largest operating battery projects in
13:39
the world that at above one hundred megawatts, and in the next or probably
13:43
the next year two years at the most. We expect to have a gigawatt
13:46
scale battery project completed. So all these things I said we would do them,
13:52
we did it, so we'ld do it. We did it. We're
13:52
going to do the rover taxi thing too. Only criticism and it's a fair
13:56
one. And sometimes I'm not on time, but I get it done and
14:00
the Tesla team gets it done. So what we're going to do this year
14:03
is we're going to reach a combined production of ten thousand a week between SX
14:07
and three. Feel very confident about that, and we feel very confident about
14:09
being feature complete with self driving. Next year, we'll expound the product line
14:13
with Model Y and Semi, and we expect to have the first operating robotaxis
14:18
next year with no one in them next year. It's always difficult to like
14:22
when things are on an exponential at an exponential rate of improvements. It's very
14:26
difficult to wrap one's mind around it because we're used to extrapolating on a linear
14:30
basis. But when you've got massive amounts of as the hardware, massive amounts
14:33
of hardware on the road, the cumulative data is increasing exponentially. Software is
14:39
getting better at an exponential rate. I feel very confident in predicting autonomous rover
14:43
taxis for Tesla next year. Not not an old jurisdictions because we won't have
14:48
regulatory approval everywhere, but I'm confident we'll have at least regulatory approval somewhere literally
14:52
next year. So any customer will be able to add or remove their car
14:56
to the Tesla network, so expect this to operate. Is like a combination
14:58
of maybe the Uber and Airbnb model. So if you own the car,
15:03
you can add or subtract it to the Tesla network, and Tesla would take
15:07
twenty five to thirty percent of the revenue. And then in places where there
15:11
aren't enough people sharing their cars, we would just have dedicated to Tesla vehicles.
15:15
Well, when you use the car, we'll show you our right sharing
15:16
app, so you're able to be able to summon the car from the parking
15:20
lot, get in and go for a drive. It's really simple to just
15:22
take the same Tesla app that currently have. We'll just do it. We'll
15:26
update the app and add some in Tesla or commit your car to the fleet,
15:30
so see that some in your car, or add some many Tesla,
15:33
or add your add or subtract your car to the fleet. You'll be able
15:35
to do that from your phone. So we see potential for smoothing out the
15:39
demand distpution curve and having the car operates at a much higher utility than an
15:43
amal car it operates so like typically the use of a car is about ten
15:48
to twelve hours a week, so most people will drive one and a half
15:50
to two hours a day, typically ten to twelve hours a week of total
15:54
driving. But if you have a car liken operate autonomously, then most likely
15:58
you could probably most likely you'd have that car operate for a third of the
16:02
week or longer. So they're a hundred hours in a week, so probably
16:04
you've got something on the order of fifty five to sixty hours a week of
16:07
operation, maybe a bit longer. So the fundamental utility of a vehicle increases
16:12
by a factor of five. So you look at this from a macroeconomic standpoint
16:17
and say, just if this was like some if we were operating some big
16:19
simulation, if you could upgrade your simulation to increase the utility of cars by
16:23
a factor of five, that would be a massive increase in the economic efficiency
16:29
of the simulation. Just gigantic, we'll do Model three S three and excess
16:33
taxis. But we made an important change to our leases, so if you
16:36
lease a Model three, you don't have the option of buying it. At
16:38
the end of the lease, we want them back. If you buy the
16:41
car, you can keep it, but if you lease it, you have
16:44
to give it back. And as as I said, where in any locations
16:47
where there's not enough supply for sharing, TESL will just make its own cars
16:51
and add them to the network in that place. So the current cost of
16:53
a Model three Robotaxi is less than thirty eight thousand dollars. We expect that
16:56
number to improve over time resigning the cars. The cars currently being built are
17:02
all designed for a million miles of operations. The drive units designed, find
17:06
and test and validated for a million million miles of operation. The current battery
17:08
pack is about maybe three hundred to five hundred thousand miles. The new battery
17:14
pack that probably going into production next year is designed explicitly for a million miles
17:18
of operation. The entire vehicle well, it's designed to operate for a million
17:22
miles with minimum maintenance. So we'll actually be adjusting tire design and really optimizing
17:26
the car for a hyper efficient robotaxi, and at some point you won't need
17:30
steering wheels or pedals and we'll just delete those. So as these things become
17:34
less and less important, we'll just delete parts just they won't be there.
17:37
If you say, like probably two years from now, we make a car
17:41
that has no steering wheels of pedals, and if we need to accelerate that
17:44
time, we can always just delete parts easy. You know, probably say
17:47
long term three years robotaxis with eliminated parts, maybe it ends up being twenty
17:52
five thousand dollars or less. And you want a super efficient car. So
17:55
the electricity consumption is very low. So we're currently at four and a half
17:59
miles per kill whatever, but we will improve that to five and beyond.
18:02
And there's just really no company that has the full stack integration. We've got
18:06
the vehicle design and manufacturing, we've got the computer hardware in house, we've
18:10
got the in house soft development and AI, and we've got by far the
18:14
biggest suite. It's extremely difficult, not impossible paths, but extremely difficult to
18:18
catch up when Tela has one hundred times more miles per day than everyone else
18:22
combined this is these todays. Is the cost of running a gasoline car or
18:26
the average cost of running a car in the US. This is taken from
18:30
Triple A. So it's currently about sixty two cents a mile. Half thousand
18:33
miles from fifty million vehicles adds up to two trillion a year. These are
18:36
literally just taken from the Triple A website. Cost of ride sharing is the
18:40
pointing to uber left is two to three dollars a mile. The cost run
18:42
a robotaxi, we think less than eighteen cents a mile, and dropping like
18:48
this is current, This would be current, this current cost, future costs
18:51
will be lower. You say, what would be the probable gross profit from
18:53
a single robotaxi. We think probably something on the order of thirty thousand bells
18:57
per year, and we expect that we're literally designing. We're designing the cars
19:02
the same way that commercial semi trailer semi trucks are designed. Commercial semi trucks
19:06
are all designed for a million mile life, and we're designing the cars for
19:08
a million my life as well. So phenomenal dollars that would be a little
19:12
over three hundred thousand dollars over the course of eleven years might be higher.
19:15
I think these consumptions are actually relatively conservative, and this assumes that fifty percent
19:19
of the miles driven art does nothing are not useful, so this is only
19:22
at fifty percent utility. By the middle of next year, we will have
19:26
over a million Tesla cars on the road with full self driving hardware feature complete
19:30
at a reliably level that we would consider that no one needs to pay attention,
19:34
meaning you could go to sleep and from ashtandpoint if you fast forward a
19:37
year, maybe a year and three months at it. But next year,
19:41
for sure we will have over a million robo taxis on the road. The
19:45
fleet wakes up with an over the air update. That's all it takes.
19:49
You say, what is in their personal value of Roverotaxi. Probably on the
19:52
order of a couple hundred thousand dollars, So buying a Model three is a
19:56
good deal. Questions our production rates Generally, if you look at a compound
20:00
annual production rate since twenty twelve, which is like the that's our first full
20:03
year of model Model as production, we went from twenty three thousand vehicles produced
20:08
in twenty thirteen to around two hundred and fifty thousand vehicles produced last year,
20:14
So in the course of five years we increased output by a factor of ten.
20:18
I would expect that something similar occurs over the next five or six years.
20:22
As for sharing sharing versus, I don't know. The nice thing is
20:25
that essentially customers are fronting us the money for the car. That's great.
20:29
Sure, we expect the solving for the snake charger. It's pretty straightforward.
20:33
It's from a vision prop standpoint, it's like a known situation. Any kind
20:37
of known situation with vision is like a charge port. It's a trivial.
20:41
Yeah, the cars were just automatically park and automatically plug in, there would
20:45
be no humans of supervision required. Yeah. During what was a pricing that
20:49
we just threw some numbers on there, I think definitely plug in order pricing
20:52
you think makes sense. We just randomly said okay, maybe a dollar.
20:56
And the thing is there's like on the order of two billion cars and trucks
21:00
in the world, So robotaxis will be an extremely high demand for a very
21:03
long time. And from my observation thus far is that the auto industry is
21:07
very slow to adapt. Like I said, there's still not a car in
21:10
the road that you can buy today that is as good as the model S
21:12
was in twenty twelve. So that suggests a pretty slow rate of adaptation for
21:18
the car industry, and so probably a dollar is conservative for the next ten
21:22
years because people to think there's like, actually not enough appreciation for the difficulty
21:29
of manufacturing is insanely difficult. But a lot of people I talked to think,
21:32
if you just have the right design, you can instantly make as much
21:36
of that thing as the world wants. This is not true. It's extremely
21:40
hard to design a new manufacturing system for new technology. Audi's having major problems
21:45
manufacturing Eutron, and they are extremely good at manufacturing. And if they're having
21:48
problems, what about others. So on the order of two billion Coston trucks
21:52
in the world, on the order of about one hundred million units per ure
21:56
of production capacity of vehicles, but only of the old design, it will
22:00
take a very long time to convert all of that to full self driving cars.
22:04
And they really need to be electric because the cost of operation of a
22:07
gasling diesel car is much higher than an electric car. It's any robotax that
22:11
isn't electric will absolutely not be competitive. There's a close that we put into
22:15
our cars. I think it was about three or four years ago. They
22:17
can only be using the Tesli the network, yes, but it's like the
22:19
app store that you can add, only add or remove them through the tails.
22:22
The network and then tells it gets revenue share. I guess you could
22:26
operate a rental car fleet, but I think this is very unwieldy. I
22:30
think there will be a phone home thing where if the car gets stuck,
22:33
it'll just phone home to Tesla and ask for a solution. Things like being
22:36
pulled over for by athlete softshare. That's easy, fresh program and that's not
22:40
a problem. It will be possible for somebody to take over using the steering
22:44
wheel, or at least for some period of time, and then probably down
22:47
the road we'll just cap the steering wheel so there's no steering control. We'll
22:51
just take steering wheel off, put a cap on that were literally just unbolt
22:53
the steering wheel and put a cap on where the steering wheel handle currently is.
22:56
I think there'll be a transition period where people will be able to take
23:00
over and should be able to take over from the rover taxi, and then
23:03
once regulators are comfortable with us not having the steering wheel, we'll just delete
23:07
that. And for cars that are on the are in the fleets, obviously
23:11
with the permission of the owner. If it's owned by somebody else, we
23:14
would just take the steering worl off and put a cap where the steering world
23:17
currently attaches. In future, will the probabity of the steering world being taken
23:19
away in the futures one hundred percent. Consumers will demand. It does not
23:22
mean prescribing a point of view about the world. This is me predicting what
23:26
consumers will demand. Consumers will demand in the future that people are not allowed
23:30
to drive these two ton death machines. There were amphibians, but then pretty
23:33
much that things just become like land creatures. It'll be a little bit of
23:37
an amphibian phase. I think are a mistake. We actually had ht maps
23:41
for a while. Actually can that because you either need hd mapps, in
23:44
which case, if anything changes about the environment, the car will break down,
23:48
or you don't need HTAM apps, in which case why you're wasting your
23:49
time during hd mapps. The hd maps saying like the two main crutches that
23:56
are that should not be used and will be obviously fall and foolish are light
24:00
aer and amaps. If you need a geofest area, you don't have real
24:03
self driving. I think we're actually going to want to push our sort of
24:06
standar range plus battery more than our long range battery because the energy content in
24:11
the long range pack is to send higher kilo what hours. So essentially you
24:15
can make third more cars if you just if they all sort of standard range
24:19
plus instead of the long range pack. It's ones like around fifty kilo what
24:22
hours the other ones around seventy five kilo what hours. So we're actually probably
24:26
gonna binance our sales intentionally towards the smaller battery in order to have a higher
24:30
volume of what basically you want to the obvious thing to maximize the number of
24:34
autonomous units or the number of maximize the output that will subsequently result in the
24:40
biggest autonomous leap down the road. So we're making doing a number of things
24:42
in that regard, but it's just on our for today is meeting the million
24:45
mile of life is basically just about getting the cycle life of the pack to
24:52
you need. Basically on the order let's say you've got a basic math,
24:55
if you've got a two undred and fifty mile range pack. You're gonna need
24:57
four thousand cycles. Very achievable. We already do that with our stationary storage
25:03
some stay stationary storage solutions like Powerpack. We're ready to play power Pack with
25:07
four thousand cycle life capability. Really fundamental message that consumers should be taking today
25:11
is that it's financially insane to buy anything other than a Tesla. They were
25:18
like owning a horse in three years find if you don't own a horse,
25:22
but you should go into it with that expectation. If you buy a car
25:25
that it does not have the hardware necessary for full self driving, it is
25:27
like buying a horse, and the only car that has the hardware nessary for
25:30
full self driving as a Tesla for people should really think about their approaches any
25:33
other vehicle. It's basically crazy to buy any other car than a Tesla.
25:37
We need to make that convey that argument clearly, and we will have to
25:41
today. Yeah, we talked to regulators around the world all the time.
25:45
As we introduce additional features like navigator and autopilots, this requires like regulatory approval
25:49
on a perju jurisdiction basis. I think fundamentally regulators in my experience are convinced
25:55
by data. So if you have a massive amount of data that shows that'omy
26:00
is safe, they listen to it. They they may take time to digest
26:03
the information. Their process may take a bit of time, but they have
26:07
always come to the right conclusion on what I've seen. I think like probably
26:11
yeah, like Tesla owned robot taxis would be in dense open areas along with
26:15
customer vehicles, and then as you get to medium and low density areas,
26:19
it would chand to be more that people own the car and occasionally lend it
26:25
out. Yeah, there are a lot of edgecases in Manhattan and say downtown
26:27
San Francisco, but those are and there are variousties around the world that have
26:32
a challenging open environments. But we do not expect this to be a significant
26:36
issue when I say future complete IDOL work in downtown San Francisco and downtown Manhattan
26:41
this year. Right now, AI neural nets are used really for object recognition,
26:45
and we're still basically just using it as still frames, so identifying objects
26:49
and still frames and tying it together in a perception path planning layer thereafter the
26:55
But what's happening is steadily is that the neural net is eating into the software
26:59
base more and more, and so over time or expect the neural net to
27:03
do more and more. Now, for a computational cost standpoint, there are
27:07
some things that are very simple for a HIC and very difficult for a neural
27:11
net, and so it probably makes sense to maintain some level of heuristics in
27:15
the system because they're just computationally a thousand times easier than a neural net.
27:21
Neural net is like a cruise missile, and if you're trying to swat a
27:23
fly, just use a fly swatter, not a cruise missile. So bit
27:26
over time, I would expect that it moves really to just training on against
27:32
video and then video in car steering and pedals out or basically video in that
27:37
lateral longitudinal acceleration out almost entirely. That's what we're going to use the Dojo
27:41
system for. There's no system that can currently do that. Essentially, the
27:44
car is going to do what a human would do. It can think with
27:47
humans like basically a camera on a slow gimble and it's a remarkable that people
27:51
are able to drive the car in the way that they are, because if
27:53
you know, you can't look at all directions at once. The car can
27:56
literally look in all directions at once with multiple cameras, so cumbs are able
28:00
to drive just by looking this way, you're looking that way. They're actually
28:03
stuck in their driver's seat. They can't really get out of the driver's seat.
28:06
So it's like kind of one camera on a gamble and is able to
28:08
drive. A conscientious driver can drive with very high safety. The cameras in
28:11
the cars have a better vantage point than the person, so they're like up
28:15
in the b pillar or at in front of the review mirror. They've they've
28:19
really got a great vantage point. So if you're turning onto a road that's
28:22
got a lot of high speed traffic, you can just do what person is
28:26
just like gradual turn a little bit. They've got fully into the road that
28:29
the camera see what's going on and if things look good, and then the
28:32
rear cameras don't show on any oncoming traffic, or if you go and if
28:34
it looks sketchy, you can just pull back a little bit, just like
28:37
a person. The behavior is like remarkably, it starts to become remarkably lifelike.
28:42
It's like quite eerie. Actually, the car just starts behaving like a
28:45
person. Well, we publish the accidents per mile every quarter, and what
28:49
we see right now is that water pilot is about twice as safe as a
28:52
normal normal driver on average, and we expect that to increase quite a bit
28:56
over time. Like I said, in the future, it will be consumers
29:00
will want to outlaw saying they will succeed or am I saying I agree with
29:03
this position, But in the future, consumers will want to outlaw people driving
29:07
their own cars because they're unsafe. And if you think of like elevators used
29:11
to be operated on a big lever go up and down the floor. There's
29:14
like a big relay and yet elevator operators, but then periodically they would get
29:18
tired or drunk or something and then turn the lever at the wrong time and
29:22
sever somebody in a half. So now you do not have to elevator operators.
29:26
And we're quite alarming if you went into an elevator that had a big
29:30
lever that could just move between floors arbitrarily. So there's just buttons. And
29:34
in the long term, again not a value judgment saying I want the world
29:37
to be this way, I'm saying consumers will most likely demand the people are
29:41
not allowed drive Carson follow up, Can you share with us how much Tesla's
29:45
spending on auto pilot or autonomous technology by order of magnitude on an annual basis?
29:49
Thank you. It's basically our entire expense structure. Between now and when
29:52
the robot taxis are fully deployed throughout the world, the sensible thing for us
29:57
is to maximize rate and drive the company cashlow neutral. Once the rover taxi
30:00
fleet is active, I would expect to be extremely casually positive. Who is
30:04
liable for an accident? Is it Tesla? Or does it mean if the
30:08
vehicle has an accident and harms, probably Tesla is probably Tesla? Yeah,
30:12
very good, right, thing too is just make sure there are very few accidents