Latest / eMotors: Electric Revolution / Wayve: London's Vision for Autonomous Driving
Transcript
- 0:09OK, let's unpack this with an image that, well, it almost
- 0:12defies common sense. Especially if you've ever tried
- 0:15to drive in the UK. Picture this, a Ford Mustang
- 0:18Maki, an electric car, completely driverless, and it's
- 0:22weaving its way through the organized chaos of central
- 0:26London traffic. And we're talking about pure
- 0:28chaos. Tight, centuries old streets.
- 0:31Pedestrians who just appear out of nowhere.
- 0:34Delivery cyclists who treat traffic laws as suggestions.
- 0:37And double Decker buses everywhere.
- 0:38Exactly. And yet this car, powered by Auk
- 0:41company called Wave, is handling all of it without a human
- 0:45touching the wheel. That scene isn't just a cool
- 0:47demo though, it signals a really monumental shift in autonomous
- 0:50vehicle technology. This isn't your parents self
- 0:53driving car. No, this is what the industry is
- 0:54calling embodied AI or AV2 Point O.
- 0:58And that is the heart of our deep dive today.
- 1:01Our goal for you, the listener, is to really get a quick but
- 1:03thorough grasp of what this AV 2 point O tech is, how it
- 1:07basically throws out the old rule book, what we'll call AV
- 1:10One point O, and why global giants like Uber, NVIDIA and
- 1:14Microsoft have poured over a billion dollars into wave.
- 1:17It's a huge bet on them winning this robo taxi race.
- 1:21For a decade, the story was all about systems that only really
- 1:24worked in you know, these perfectly gridded American
- 1:28cities, right? Wave chose one of the toughest,
- 1:31most unpredictable environments on Earth to prove their model
- 1:34works. Which is why the proof is so
- 1:37important. So let's jump straight to that,
- 1:39the real world proof ABBC reporter recently took a ride in
- 1:44one of these wave equipped Mustangs.
- 1:46And they didn't take it easy on the car.
- 1:48No, the route went through genuinely complex situations.
- 1:51Sudden traffic jams. Temporary traffic lights popping
- 1:53up. Double parked cars, forcing it
- 1:55to maneuver into oncoming traffic.
- 1:57All the stuff that makes you grip the steering wheel.
- 1:59And the outcome was perfect. The human safety driver who is
- 2:02always in these cars for now, never once had to intervene.
- 2:06And that is the key metric, isn't it?
- 2:08Especially in London, where the driving culture is, let's say,
- 2:12aggressive A. Little impatient.
- 2:14Yes, Wave System showed it could handle these really subtle,
- 2:17nuanced events. OK, so give us an example of
- 2:20that nuance. What did it actually do?
- 2:22Well, the reports mentioned a few things.
- 2:24It correctly predicted the path of a food delivery cyclist who
- 2:28just cut across three lanes of traffic.
- 2:30Which happens every 30 seconds in London.
- 2:32It's constant and it navigated around parked cars then merged
- 2:36back into the flow of traffic seamlessly.
- 2:39And crucially, it avoided pedestrians who are just
- 2:42completely ignoring the crosswalks and everyday reality.
- 2:46Here's the fascinating part, the real nuance.
- 2:49The reports all stressed that it didn't drive like a confident
- 2:52London cabbie. No, not at all.
- 2:54It drove with the cautious hesitation of a nervous learner
- 2:57driver. It would pause for an extra half
- 3:00second where a human would just gun it.
- 3:02And while that might be a little annoying if you're behind it,
- 3:05that caution points to some very high level decision making.
- 3:08There's this one moment that really stood out.
- 3:10The pedestrian. The blind pedestrian, yeah,
- 3:12first with a cane, stepped out unexpectedly from between two
- 3:15parked cars, right into the car's path.
- 3:18And the a. The AI saw the cane, registered
- 3:21the movement, slowed down and adjusted its course before the
- 3:25human safety driver could even physically react and hit the
- 3:27brake. That's incredible.
- 3:29It tells you it's not just following a set of pre
- 3:31programmed brittle rules. It's reacting to the world in a
- 3:36way that suggests it's generalizing, predicting.
- 3:40It showed high adaptation and safety, even if it lacked that
- 3:43human confidence for, you know, maximum traffic efficiency.
- 3:47And this level of proven caution is exactly what the UK
- 3:50Government wants to see as they get ready to green light this
- 3:52tack. Absolutely.
- 3:54Right now, these cars need human oversight, but there's a clear
- 3:57regulatory path in place. And that clarity attracts
- 4:00investment, right? Of course, the UK is planning
- 4:03legislation to enable fully self driving vehicles, possibly
- 4:06without a safety driver at all from the second-half of 2027.
- 4:10That's. A hard date.
- 4:11It is. It provides A framework that
- 4:13encourages companies to be aggressive, like Uber's plan for
- 4:16a Level 4 robo taxi pilot in London, which Wave is part of,
- 4:20targeted for 2026. OK, so let's pull back the lens
- 4:23a bit. Let's look at the technology
- 4:25that makes waves so different. How can they handle the chaos of
- 4:28London when other systems struggle even in Phoenix?
- 4:31This is the critical technical take away for you.
- 4:34For the last decade, AV one point O the tech pioneered by
- 4:37companies like Waymo relied on a very brittle, very expensive
- 4:42architecture. What made it so brittle?
- 4:44It was two things really. First, these incredibly detailed
- 4:47high definition 3D maps of every single Rd. the car would ever
- 4:51drive on. South, the car goes off the mat.
- 4:53It stops. It's geofenced and second it
- 4:56relied on thousands, sometimes millions of hand coded rules.
- 5:00Like if you see this specific type of construction barrel at
- 5:03this exact angle, then do this specific maneuver.
- 5:06That sounds impossibly expensive to scale.
- 5:08If you want to go from London to Paris, you have to remap
- 5:11everything and then recode all the rules for how Parisian
- 5:14scooter drivers behave. You've got it.
- 5:16It just doesn't scale and it often required all this bulky,
- 5:19expensive sensor equipment like Lidar, which drove up the
- 5:22vehicle. Cost AV2 point O waves approach
- 5:25flips that on its head. So how do they get around the
- 5:27mapping and the rule coding trap?
- 5:29They use a foundation model AI. It's trained end to end on just
- 5:33massive amounts of raw camera and sensor data.
- 5:36So instead of being programmed with rules, the model learns the
- 5:39physics and the causality of driving just from watching.
- 5:42It learns to generalize. Exactly, just like a human, you
- 5:45don't need a high def map of a road you've never seen to be
- 5:48able to drive on it safely. Which brings us back to that
- 5:50term from the intro Embodied AI. We're all getting used to
- 5:54cognitive AI. Things like ChatGPT, which is
- 5:57brilliant at predicting the next word in a sentence.
- 6:00Right, it operates in an abstract world of language.
- 6:02So what happens when you put that kind of predictive power
- 6:05into something physical like a car?
- 6:08That's embodied AI. It's about giving AI models the
- 6:11ability to interact with and influence the physical world in
- 6:15real time. The model has to build an
- 6:16internal predictive model of the world.
- 6:19It has to understand inertia, friction, the intentions of that
- 6:22pedestrian. And make life or death decisions
- 6:25based on that understanding. So why is driving the best first
- 6:28application for this? It's two things unparalleled
- 6:32data scale and the intense demand for safety.
- 6:35Autonomous cars generate petabytes of video and sensor
- 6:38data. That scale of real world
- 6:41physical data is actually way bigger than all the language
- 6:44data used to train models like ChatGPT.
- 6:47I never thought of it that way. And the safety critical nature
- 6:49of driving just forces you to innovate.
- 6:52If your language model hallucinates, you know it's
- 6:54annoying. If your car's AI model
- 6:56hallucinates, that's a crash. That really clarifies it.
- 7:00And Wave Co, founded by Cambridge alumnus Alex Kendall,
- 7:03has been pushing this idea since 2017.
- 7:05They were really swimming upstream back then.
- 7:08Oh. Completely.
- 7:09It was a deeply contrarian approach.
- 7:11All the big money was being poured into brute forcing those
- 7:13AV one Point O rule based systems.
- 7:15It's kind of call back to old projects like LVN in 1989 which
- 7:19tried this but just didn't have the compute power or the data.
- 7:22The technology just wasn't there yet.
- 7:24Not even close. Wave is capitalizing on that
- 7:26convergence of cheap cameras, powerful GP use, and massive
- 7:29data sets that exist today. Well, that contrarian bet
- 7:32certainly seems to have paid off.
- 7:33Wave just secured a $1.05 billion funding round.
- 7:38One point O 5 billion. I mean, that's not just a good
- 7:41investment, it's the largest AI fundraise in UK history.
- 7:45Their total funding is over $1.3 billion.
- 7:48Now that much money tells you the strategy is sound.
- 7:52And that the partners are highly strategic.
- 7:54Let's breakdown that ecosystem because it explains everything
- 7:57about their commercial plan. First, you have Uber.
- 8:00Right. Uber's business model depends on
- 8:02drivers, which means high labor costs, low margins.
- 8:06They desperately need autonomy. But they learned the hard way
- 8:09that building your own AD tech stack is just astronomically
- 8:13expensive. So Wave offers them a shortcut.
- 8:16Exactly. A licensable, scalable software
- 8:18brand. That's what makes the 2026
- 8:20London Robo taxi pilot possible. Uber gets the autonomy without
- 8:24the massive R&D spend. Then you have the tech partners
- 8:27NVIDIA and Microsoft. Waves models need a phenomenal
- 8:30amount of computing muscle, both for the training which happens
- 8:32in the cloud and for the inference the real time decision
- 8:35making inside the car. And that's a direct win for
- 8:38NVIDIA. It is.
- 8:39It guarantees huge demand for their most advanced GPU's and
- 8:42their driveway platform. They're cementing their role as
- 8:46the AI engine of mobility. Wave needs their chips to run
- 8:50these complex models in a car. And Microsoft provides the cloud
- 8:54backbone for all that training. Wave runs on Azure, so this
- 8:58investment deepens Microsoft's footprint in a huge AI vertical.
- 9:02It's actually, you know, moving things in the physical world.
- 9:05It's a textbook case of mutual dependency.
- 9:08But the most important point for you to understand, the thing
- 9:10that clarifies their whole goal is Wave's business model.
- 9:15They are not operating taxis. They're not competing with Uber.
- 9:17They're enabling Uber. They are a pure software
- 9:20provider. They plan to sell or license
- 9:23their AI driver to car makers and fleet operators globally.
- 9:27They just want to be the intelligence behind the metal.
- 9:29And their first commercial product is a brilliant move.
- 9:31It's not full Level 5 autonomy. No, they plan to launch an L2
- 9:35plus driver assistance system first.
- 9:37Something like a really, really advanced cruise control.
- 9:40Which is so clever. It lets them generate revenue
- 9:42right away. It builds trust with consumers.
- 9:44And most importantly, it lets them collect millions of miles
- 9:48of diverse real world driving data from their customers, which
- 9:52they then feedback into their main AI model to make it even
- 9:55better. So they're using a current
- 9:57product to fund the R&D for the future product, all while
- 10:01harvesting the data they need to perfect it.
- 10:03That's a sustainable path to scale.
- 10:06And all of this connects back to the bigger picture, the societal
- 10:08impact. We're talking about a projected
- 10:10market worth over $100 billion. But the real motivation is
- 10:15bigger than money. The sources really highlight the
- 10:17human cost of driving. Around 1.2 million people die in
- 10:21traffic accidents globally every single year.
- 10:24If autonomous driving can reduce or even just partially eliminate
- 10:28that number, the moral mandate for this technology is, well,
- 10:31it's massive. But the challenges are still
- 10:33huge, and they're not just technical.
- 10:36No, it's two filled. First they have to shed that
- 10:38nervous learner driver caution while staying perfectly safe.
- 10:42The AI needs to learn the right amount of aggression to merge
- 10:45into heavy traffic without causing a backup.
- 10:48And then there's the cultural hurdle, public acceptance.
- 10:51Especially in the UK, people are skeptical.
- 10:54London cabbies are already calling them fairground rides.
- 10:57And you have to win those people over with flawless performance
- 11:00day in and day out. But wave strategy is built for
- 11:03that long game, isn't it? It is.
- 11:05The old AV one point O approach was about these constrained Geo
- 11:09fenced solutions. Expensive to deploy Fragile
- 11:12inside their little sandbox. Waves AV 2 Point O is the
- 11:15scalable solution. It's designed to get gradually
- 11:17better everywhere all at once through over the air updates
- 11:21using cheaper, more standard cameras and sensors.
- 11:23And as we said, that's only possible now because the
- 11:26computing power from companies like NVIDIA is finally here.
- 11:29You can run these massive AI models right inside the car.
- 11:32The software intelligence is replacing the brute force of
- 11:35extensive hardware. It really is a compelling moment
- 11:38in technology. This has been a tremendously
- 11:40insightful deep dive. We've looked at the real world
- 11:43success of Waves Embodied AI in London.
- 11:46We've clarified that fundamental shift from AV 1.0 to Mapless end
- 11:50to end learning in AV 2.0. And we've mapped out that
- 11:54powerful billion dollar ecosystem that's backing this UK
- 11:57pioneer. So to leave you with a critical
- 11:59thought to consider as you watch this space evolve.
- 12:02Wave has the money and the breakthrough tech, but their
- 12:05plan is to be a pure software provider.
- 12:08Can they actually carve out a viable market niche, or will
- 12:12they inevitably get squeezed by the industry behemoths?
- 12:16Because on one side you have Tesla, which is using a similar
- 12:19end to end approach but has a massive head start with billions
- 12:22of miles of real world driving data from its customers.
- 12:25And on the other side, you have Waymo, which completely
- 12:27dominates the established geofenced AV one Point O market
- 12:31and is already making significant revenue.
- 12:33It's the essential question, isn't it?
- 12:35Is being the brain for hire a sustainable model when you're
- 12:38two biggest competitors own either the entire software
- 12:41infrastructure or the real world operational fleet?
- 12:44It's a classic battle of scaling strategies we'll just have to
- 12:47watch closely. That is a fascinating strategic
- 12:50question to Mull over as this technology continues its rapid
- 12:53advance. Thank you so much for joining us
- 12:55on this deep dive into autonomous mobility and embodied
- 12:59AI. We always enjoy helping you
- 13:00simply unpack these complex sources and distill what matters
- 13:05most. If you enjoyed gaining this
- 13:06knowledge quickly and thoroughly, we invite you to
- 13:09follow the channel to stay informed on our next exploration
- 13:12into the world's most interesting subjects.
- 13:14Until next time.