Latest / Elon Musk Podcast / Tesla FSD or Wayve? Hands-On San Francisco Drive Review & ADAS Insights (2025)

Transcript

Tesla FSD or Wayve? Hands-On San Francisco Drive Review & ADAS Insights (2025) · Elon Musk Podcast

• 0:03
  Your ultimate authority for daily Elon Musk news.
• 0:06
  Exploring the world's biggest ideas with your host Will
• 0:09
  Walden. There's something new every day.
• 0:14
  Thanks to our amazing community members like you, we've reached
• 0:17
  the top. 15 of. Spotify's video podcast.
• 0:20
  The top 10 audio podcasts on both.
• 0:23
  Apple and Spotify for. The tech category.
• 0:25
  So you. All make this possible.
• 0:27
  If you want to support us more, check out our.
• 0:29
  Patreonthatspatreon.com/stage. 0 News so.
• 0:33
  We can keep this free and open for you to enjoy.
• 0:39
  Something interesting happened the other day.
• 0:41
  I was looking through our stats on Spotify and Apple Podcasts.
• 0:45
  And I noticed that about 50. 5% of you are not subscribed to the
• 0:50
  show. That means 45% of you are
• 0:53
  subscribed and I really? Do appreciate your support the
• 0:56
  other. 55% of you are awesome, but I'm going to ask you for a
• 1:00
  favor. Could you please hit the?
• 1:02
  Subscribe button. It'll take you one second, but I
• 1:05
  promise you 10 years of this podcast.
• 1:07
  For free, no paywalls. I'm not going to charge you
• 1:10
  anything ever, but I'm going to give you 10 years of this show
• 1:13
  for free. I've already been doing it for
• 1:14
  five years and I plan on doing it for 10 more.
• 1:17
  And the only way that we can continue doing this.
• 1:19
  Is with your. Support so one.
• 1:21
  Second of your. Time to hit the subscribe button
• 1:24
  right now. Would help the show.
• 1:26
  Tremendously. Thank you so much.
• 1:31
  Late last week, Business Insider rode in a Ford Mustang Mach E
• 1:35
  fitted with a company called Wave.
• 1:39
  It's the EV2 point O. They did a nearly hour long
• 1:45
  drive through San Francisco with no safety monitor takeovers
• 1:48
  beyond parking. And what waves EV 2 point O is,
• 1:53
  it's basically Tesla's full self driving.
• 1:55
  They did a review of this. So today I'm going to break down
• 1:58
  what that test says about Waves approach compared to Tesla, how
• 2:01
  it compares to Tesla's full self Driving, and what drivers and
• 2:05
  automakers can actually expect from the company.
• 2:09
  It'll lay out the reporting in plain terms and show where the
• 2:11
  technology delivers and where the gaps still sit.
• 2:16
  And I keep the focus on concrete behavior on the actual streets.
• 2:19
  From this review, I'm not going to give you my opinion because I
• 2:22
  didn't drive the vehicle, but I can tell you from this article
• 2:27
  that there are some definite differences.
• 2:29
  Now does WAVES approach give automakers a faster path to a
• 2:34
  capable driver assist system than trying to match Tesla mile
• 2:38
  for mile and build their own from scratch?
• 2:43
  Because that would be intense, a lot of money, and also would
• 2:47
  take a lot of time. Could they just swap this in to
• 2:50
  their new vehicles and have full self driving from the get go?
• 2:55
  Because Wave likes they're going to license this, they're going
• 2:59
  to license a hardware agnostic assisted driving stack to car
• 3:03
  makers rather than run a robo taxi service.
• 3:05
  They're going to give it to everybody else.
• 3:06
  They're going to make money from everybody else.
• 3:09
  So what does that mean in practice?
• 3:12
  Why does it change the business math for OEMs?
• 3:14
  Here we're going to analyze the system's behavior, the training
• 3:18
  recipe that moved it from the UK to US roads, and the timing
• 3:23
  claims to set expectations for when drivers might see this in
• 3:26
  showrooms. You know, I'm going to be
• 3:30
  covering automated driving systems on this channel, so if
• 3:34
  you're into that, make sure to hit the subscription or follow
• 3:37
  button. We talk about how they're built,
• 3:40
  how they get deployed and then we're going to jump into that
• 3:44
  right after this quick break. Now let's start with this test.
• 3:49
  Business Insider rode in a wave equipped mock E during evening
• 3:54
  rush hour around the Moscow and Center and Soma.
• 3:58
  The car handled jaywalkers, door opening drivers, and blocked
• 4:02
  intersections without a human corrections during this route.
• 4:06
  The real world behavior sets a baseline for comparing Wave's
• 4:10
  assisted driving approach to Tesla's FD that anyone can buy
• 4:15
  today. Now Wave's targets are the
• 4:19
  assisted driving market that automakers can ship widely, not
• 4:23
  a geofence robotaxi product like Elon Musk and Tesla are doing
• 4:28
  right now. The company pitches software
• 4:31
  that can drop into many vehicles and sensor stacks.
• 4:34
  So you have to have sensors on your vehicle.
• 4:36
  You can't just add the software and their hardware.
• 4:38
  And this approach gives OEMs the option to sell advanced driver
• 4:42
  assist without adding new hardware lines to the bill of
• 4:45
  materials. They just go to this
• 4:46
  manufacturer, they go to Wave and they say, hey, we need auto
• 4:50
  driving for a vehicles hook us up now that promise of no
• 4:55
  additional hardware cost is the hook that Wave uses in talks
• 4:58
  with manufacturers. The sister use a end to end
• 5:02
  model that learns driving directly from data rather than a
• 5:06
  rule book of hand coded behaviors.
• 5:09
  The vehicle perceives the scene through cameras and compare it
• 5:12
  with radar or lighter when available, and the model outputs
• 5:16
  control for steering and speed in one pipeline that adapts as a
• 5:20
  trains on more examples. Now that architecture matters
• 5:24
  because it let's wave keep one code base that can fit any car
• 5:30
  now. This San Francisco demo set
• 5:33
  between SAE levels two and three.
• 5:36
  The safety operator kept hands close but did not intervene
• 5:41
  except to park at the very end. And now the car braked hard two
• 5:45
  or three times while inching through heavy traffic, which
• 5:47
  matches the twitchy, low speed behavior that we've all seen in
• 5:51
  other automated systems. These small habits tell you the
• 5:55
  driver remains part of this loop.
• 5:58
  And we have to make sure that people are in these vehicles
• 6:01
  when they're still testing, have to be able to get to that wheel
• 6:04
  if needed. Now the comparison point
• 6:06
  everyone asks for is Tesla's FSD.
• 6:09
  It's the top of the line. Consumers can buy a Tesla and
• 6:13
  activate FSD right now. And Tesla's running robotaxi
• 6:17
  trials in Austin and they have a safety monitor riding in the
• 6:20
  front seat with them. Now that availability advantage
• 6:24
  gives Tesla real world miles and a user base that can fund
• 6:28
  constant iteration. And also any other vehicle with
• 6:30
  FSD. Millions of vehicles.
• 6:33
  Wave says mass market cars with its software likely arrive later
• 6:37
  in the decade though, which sets a different deployment curve.
• 6:42
  Elon and Tesla are already there.
• 6:45
  Wave needs to wait a little while till later on in the
• 6:48
  decade. So two or three years from now
• 6:51
  they're going to be deploying. Will vehicle manufacturers,
• 6:56
  hardware and software manufacturers let Wave in or
• 7:00
  will companies build their own? I don't know.
• 7:05
  It's it's really up in the air right now, but Wave's training
• 7:08
  story deserves a little bit of attention here.
• 7:11
  The company says it adapted Auk train driver to US roads with
• 7:16
  about 500 hours of US specific data.
• 7:19
  Now that's a small number compared to Tesla's claim of 6
• 7:22
  billion real world miles collected from its fleet of
• 7:25
  vehicles. Everybody that has a Tesla,
• 7:27
  they're gathering your data right now.
• 7:29
  They're spying on you and gathering your data.
• 7:31
  That's what they that's why they have you in that car.
• 7:33
  They need your data. But it points to an efficiency
• 7:36
  rather than a race to aggregate the largest pile of data.
• 7:41
  This is kind of a bet that targeted data and strong
• 7:44
  simulation, and they're hoping that it can move a model into
• 7:49
  new domains with less time and costs and train this model
• 7:54
  without billions of miles. Now.
• 7:59
  What kind of partnerships will they have?
• 8:02
  What kind of scale would they have?
• 8:04
  Wave has already announced a deal with Nissan in April to
• 8:08
  bring its assisted driving into mass produced vehicles.
• 8:12
  Tesla said it's in talks to license FSD to major automakers,
• 8:16
  but they have no named partners yet.
• 8:18
  Automakers will compare costs, integration timelines,
• 8:22
  liabilities and how each stack behaves in their own cars before
• 8:26
  they commit. Of course they're going to be
• 8:27
  doing testing on their own. They're not just going to take
• 8:29
  somebody's word for it. You know, if Tesla has billions
• 8:33
  of miles driven and Wave only has thousands, even if it's
• 8:39
  hundreds of thousands, that's not enough.
• 8:43
  So unless they have this incredible software stack that
• 8:49
  is just unbeatable, I don't see any manufacturers going with
• 8:52
  Wave or not many manufacturer going with Wave over Tesla at
• 8:55
  this point. Does come down to price too.
• 8:58
  So the licensing strategy influences sensor choices and
• 9:02
  compute budgets too. It really depends.
• 9:04
  You may have certain subscription models for this
• 9:10
  type of full self driving. You could have like a low level
• 9:13
  and mid level and high level like anything else for any
• 9:16
  subscription model. And wave positions as hardware
• 9:19
  agnostic across camera only builds and richer stacks with
• 9:22
  radar or Lidar so they can do anything.
• 9:25
  Now that flexibility let's any OEM choose a lower cost camera
• 9:29
  configuration for driver assist or a higher redundancy package
• 9:32
  for a higher automation feature in premium trims.
• 9:37
  A Tesla single maker vertical stack gives it tight hardware
• 9:40
  software control. But a licensable solution that
• 9:43
  fits many platforms lowers frictions for traditional OEMs.
• 9:48
  Tesla would control the hardware and the software, but if they
• 9:52
  license it to other hardware manufacturers, then they
• 9:57
  wouldn't be the Apple of vehicles anymore.
• 10:00
  They would become more like the Android of vehicles.
• 10:05
  A lot of different models, a lot of different makes, a lot of
• 10:07
  different manufacturers doing a lot of different things.
• 10:10
  And if Elon and his Tesla homies can figure this out and be good
• 10:15
  with all of those, that is going to be ridiculous.
• 10:19
  But that's what Wave is trying to do before Elon gets a hold of
• 10:21
  him. So performance through this,
• 10:26
  through this test is how we're going to drop, how we're going
• 10:30
  to judge this right, the trust of this.
• 10:32
  The Business Insider writer said a passenger might struggle to
• 10:35
  tell Waves behavior from Tesla's on the route that they actually
• 10:38
  took. That's a sign that modern end to
• 10:42
  end systems have converged on similar lane keeping, pass
• 10:45
  selection and gap acceptance behaviors in traffic that
• 10:49
  follows familiar patterns. I mean, we have a Honda Civic
• 10:53
  and it has you stay in your lane.
• 10:55
  You know, they have that lane keeping gap acceptance in
• 10:59
  traffic. It'll slow down, speed up on its
• 11:03
  own, you know, with cruise control.
• 11:05
  And that's enough for a lot of people.
• 11:07
  I don't really need much more than that.
• 11:08
  I don't need my car to drive for me.
• 11:10
  I actually enjoy driving. So for me, that's enough for a
• 11:14
  long trip to stay in your lane, you know, and signal me when I
• 11:18
  fall asleep or what, you know, when I'm going out of the lane
• 11:21
  to signal me, OK. But the other people, they might
• 11:25
  need that assistance, especially people that it's difficult for
• 11:31
  them to drive for some reason. You know these familiar patterns
• 11:36
  that Tesla does gap acceptance. Keep in your lane shift lanes
• 11:42
  smoothly. The differences show up in edge
• 11:44
  cases, intervention rates over millions of miles and how
• 11:49
  systems recover from ambiguous scenes.
• 11:52
  Now this timing is a little bit far for me.
• 11:58
  Wave points to fully autonomous tests in London with Uber in
• 12:02
  spring of 2026, which will be great.
• 12:04
  And while brought ADAS licensee into production vehicles could
• 12:08
  land as soon as 2027 and might not be soon enough.
• 12:12
  Tesla continues public trials and sells supervised features
• 12:16
  right now, and that split explains why you can ride an FSD
• 12:20
  today, but still cannot buy a car from a third party with
• 12:23
  Waves stack yet in the business case for OEMs hinges on total
• 12:28
  cost to shipment and maintain. So Wave argues that integration
• 12:33
  happens on existing sensors and chips.
• 12:35
  So anything that's out there already that reduces CapEx and
• 12:39
  shortens timelines, If OEMs want to get their own hardware, they
• 12:42
  can and they can just plug in Wave.
• 12:46
  Tesla's model would give an OEM mature software with a huge data
• 12:49
  engine behind it. Also requires alignment on
• 12:52
  update cadence, feature Rd. maps in branding, decisions that
• 12:56
  matter to a car maker's identity.
• 12:58
  Think about that. How many car makers, how many
• 13:00
  automakers are ditching car play, Apple's car play, because
• 13:06
  they want to build their own system because Apple Car play is
• 13:09
  not their brand. They want to take control of
• 13:12
  their brand within the vehicle, not only on the outside with
• 13:16
  their logos in their shiny stickers, things like that.
• 13:22
  They don't want that anymore. They want everything interior
• 13:25
  too. And sure, cars have been doing
• 13:27
  this for a long time. Vehicle makers have been doing
• 13:29
  this for a long time. But this way they have control,
• 13:32
  complete control over your software.
• 13:35
  Elon and Tesla, if you do full self driving, you might have to
• 13:40
  use their software for everything in your vehicle.
• 13:43
  And would that be a bad thing? I don't think so.
• 13:47
  You know, if you have a platform and say if you're Ford and you
• 13:51
  have a platform that works for you, but Elon offers you full
• 13:55
  self driving, billions of hours of data, billions of miles of
• 14:01
  data, and all you have to do is put a Tesla system in your car
• 14:06
  and do a little Tesla branding in there somewhere.
• 14:09
  Would you do that? That's going to be a tough
• 14:11
  question for these automakers and you know, that's up to them.
• 14:15
  If the price is right and they know they can have a superior
• 14:18
  product to Tesla, they want to keep that market share because
• 14:21
  Tesla's blowing them out of the water.
• 14:23
  That might be a good way to partner with Tesla.
• 14:25
  You get some of that what they call the rub, you know, from
• 14:28
  from basically your cool buddy comes along with you on a trip
• 14:33
  or a date or something comes along with you and like or not
• 14:37
  on a date. That's weird.
• 14:38
  But if they come along with you and you're hanging out with a
• 14:40
  bunch of people, like you get that rub from the cool person,
• 14:42
  right? Tesla's the cool guy.
• 14:45
  And of course, Ford isn't as cool as Tesla to tech Bros and
• 14:48
  to people like early adopters. So maybe Ford could get a little
• 14:52
  bit of that rub. Tesla, I don't know.
• 14:54
  It's up to them. Is that business case going to
• 14:56
  be worth it now? Each option trades control for
• 15:00
  speed in different ways. The road test details have also
• 15:04
  given us a clean read on a user experience.
• 15:06
  For Wave, the car waited for a vehicle blocking an intersection
• 15:09
  at a green light, which shows conservative conflict resolution
• 15:13
  system handled jaywalkers and during hazards with appropriate
• 15:17
  pauses and path adjustments, which signals solid short
• 15:21
  horizon planning and the breaking spikes in creeping
• 15:24
  traffic show tuning work that still remains to be done.
• 15:30
  It's a marginal thing. They can tweak that very easily.
• 15:36
  The take away is pretty simple, though.
• 15:38
  Waves Demo shows competent assisted driving behaviour in a
• 15:42
  complex urban setting under supervision.
• 15:44
  Tesla offers a product you can buy now with ongoing trials that
• 15:48
  push into high autonomy with a safety monitor.
• 15:53
  Waves licensing story and domain adaption claims give automakers
• 15:58
  an alternative route to ship useful assistance without owning
• 16:04
  the full vertical stack. They can plug it in, plug and
• 16:07
  play on their vehicles. They have full self driving
• 16:10
  because of Wave. It's a licensable, hardware
• 16:14
  flexible driver assist system that performs cleanly in San
• 16:19
  Francisco and Tesla. They hold the advantage right
• 16:22
  now. They ship FST today and they're
• 16:26
  building with data at scale. Hey, thank you so much for
• 16:32
  listening today. I really do appreciate your
• 16:34
  support. If you could take a second and
• 16:36
  hit the subscribe. Or the follow.
• 16:38
  Button on whatever podcast platform that you're listening
• 16:40
  on right now. I greatly appreciate it.
• 16:43
  It helps out the show tremendously and you'll never
• 16:45
  miss an episode. And each episode is about 10
• 16:48
  minutes or less to get you caught up quickly.
• 16:51
  And please, if you want to support the show even more, go
• 16:54
  to patreon.com/stagezero and please take care of yourselves
• 17:00
  and each other and I'll see you tomorrow.