Skip to content

Podcast

Waymo and the Rise of the Robotaxis | to Build a Driver | 1

Business Wars

Source ↗ ← All highlights
  • Passenger Trapped In Looping Robo-Taxi
    • Mike Johns experienced a robo-taxi looping a parking lot and he couldn’t stop it from inside the car.
    • He was shaken and vowed never to ride a robo-taxi again, illustrating passenger trust issues. Transcript: David Brown It’s December 2024, and in Phoenix, Arizona, Mike Johns is trapped inside a driverless taxi. Johns Waymo has arrived at his destination, the airport terminal, but instead of stopping to drop him off, the vehicle keeps looping through the parking lot over and over. The robo-taxi is now on its sixth lap of the lot, and Johns can’t open open Waymo representative tells Johns to try using the stop command on the Waymo app. I don’t have an option to control the car. Oh, my gosh. After several minutes, the representative finally finds a way to stop the ride. Johns bolts for the terminal, shaken, and vowing never to ride in a robo-taxi again. But in a growing number of U.S. Cities, passengers don’t have this choice. A tap in an app might summon a human driver or a car with no driver at all. And leading this driverless revolution is Waymo, Google’s 20-year quest to build not a car, but a driver. Waymo says its vehicles have now driven more than 100 million autonomous miles. That’s the equivalent of traveling to the moon and back 200 times. But this journey has cost Google’s parent company, Alphabet, billions, with profits still somewhere over the horizon. And the competition, Amazon Zoox, and China’s Baidu is closing in fast, betting that speed, not caution, will win the race. But is this the dawn of a trillion-dollar industry? (Time 0:00:09)
  • Technical Challenges Extend Timeline
    • Building a car that sees and thinks like a human proved far harder than engineers expected.
    • Technical edge and regulatory, insurance, and trust hurdles all slowed commercialization. Transcript: David Brown You know, driverless taxis have been promised for years, and now that promise is finally starting to take shape. Today, in cities like Phoenix, Los Angeles, and Shanghai, people can hail rides with no one behind the wheel. Technology that once felt like science fiction has slipped into the everyday, in some places at least. But the journey has been long. It’s been costly and full of wrong terms. And the next stretch may prove even more challenging. Building a car that can see and think like a human has proven to be far harder than even Google’s engineers imagined. Mapping streets is one thing. Teaching AI how to react to emergency vehicles, spot chain-link fences, or respond to a construction worker waving traffic through the wrong lane, that’s another matter. Even something as ordinary as a plastic bag drifting across the road can confuse the vehicle’s cameras. And then there are the human hurdles. Regulators aren’t sure who to ticket when a driverless taxi breaks the rules. Insurance companies can’t agree on who’s to blame when an accident occurs, and many passengers still aren’t ready to trust an empty driver’s seat. One recent poll found just 9% of people said they trust a self-driving car, while 66% actively fear them. Yet, despite all the uncertainty, the money keeps flowing. Billions of dollars have already been poured into bringing these vehicles to market. (Time 0:02:46)
  • Ghost Rider: The Audacious Motorcycle
    • Anthony Lewandowski built Ghost Rider, an autonomous motorcycle for DARPA’s Grand Challenge.
    • The bike fell hundreds of times but showcased bold differentiation and hustle. Transcript: David Brown It’s 2002 at the University of California, Berkeley. In a cluttered lab, a 22-year graduate student hunches over a desk scattered with circuit boards and toy parts, with a phone wedged between his ear and shoulder. His name is Anthony Lewandowski. He’s Belgian, ambitious, and already known around campus as the kid who never sleeps. On a shelf behind him sits his latest creation, a small Lego robot that sorts Monopoly money by color. Sounds trivial, but that little robot has just won him a national robotics prize and a reputation as the campus prodigy who can build anything out of almost nothing. Lewandowski’s on the phone with his mother in Brussels. She’s just read an article about a new U.S. Government competition, a million-dollar challenge that sounds almost impossible. The contest is run by a U.S. Government agency, and it’s offering a prize of $1 million for the first vehicle that can drive itself across the Mojave Desert. The contest has an exhilarating romantic name, the Grand Challenge. Lewandowski pauses, a driverless car. To most people, it sounds like science fiction. To Lewandowski, a bright engineering student who still believes that with enough intelligence and dedication, anything is possible, well, it sounds like an opportunity. The competition is the start of a journey that will pull him from Berkeley’s labs into the heart of Silicon Valley and, eventually, into one of the fiercest rivalries in tech history. The agency behind this million-dollar competition is DARPA, the U.S. Department of Defense’s secretive engine of innovation. Formed after the Soviet launch of Sputnik, DARPA’s mission is to ensure America is never again caught technologically flat-footed. The Internet began as a DARPA project, and the agency has played a huge role in other technologies, too, such as satellite navigation, voice assistants like Siri, and even the graphical Interfaces that help make personal computing usable. Now, DARPA has turned its attention to driverless cars. The push comes from the war in Afghanistan, where U.S. Troops are being killed by roadside bombs hidden along convoy routes. Congress wants to remove soldiers from harm’s way wherever possible, and it has tasked DARPA with an ambitious goal, build autonomous military vehicles by 2015. The grand challenge is DARPA’s attempt to jumpstart this effort. It’s a desert race open to anyone who can build a vehicle that drives itself nearly 150 miles across the Mojave. No remote controls, no human staring from afar. Each vehicle must sense the world, make decisions, and drive on its own. Like many others, Lewandowski first considers building a car. But one day, riding through the California hills, a pack of motorcyclists whips past him, which sparks an idea. A motorcycle would be harder to automate than a car. Far harder. But it would also set Lewandowski apart. DARPA wants bold ideas, and surely no one else is attempting to automate two wheels instead of four. If he can make a bike balance and steer itself through the desert, he won’t just be another entrant. He’ll be unforgettable. This is a masterclass in differentiation. Lewandowski knows he’s an underdog with less money and fewer resources than the big teams. So instead of trying to beat them at their own game, he changes the game entirely. He picks a project so audaciously difficult that even if it fails, he’ll succeed in separating himself from the herd. (Time 0:04:51)
  • First DARPA Race: Spectacular Failures
    • In the 2004 DARPA Grand Challenge nearly every entrant failed, with Lewandowski’s bike falling at the start.
    • Carnegie Mellon’s Sandstorm made the furthest progress at 7.4 miles, showing early promise. Transcript: David Brown It’s March 2004, and in the Mojave Desert, the countdown to the start of the first Grand Challenge is underway. The starting line looks like something out of Mario Kart. Fifteen experimental vehicles are lined up in the dirt. Cars, trucks, and a single motorcycle. Each one promising to drive itself across nearly 150 miles of rough desert. Ghost Rider, Lewandowski’s autonomous Yamaha, is the only bike in the field, and it looks wildly out of place even among the hulking four-wheelers beside it. A beige Humvee named Sandstorm idles nearby. It was built by a team from Carnegie Mellon University in Pittsburgh, including a young roboticist named Chris Urmson. Unlike Lewandowski’s camera-based system, the Carnegie Mellon crew relies on something new, LIDAR. That stands for Light Detection and Ranging, and it’s a sensor that fires thousands of laser pulses a second, measuring how long each takes to bounce back. LIDAR allows Sandstorm to build a three-dimensional map of the world in real time. The Marshal waves a flag and the robotic vehicles lurch forward. Chaos ensues. One truck veers into a barrier. Another plows into a ditch. A third clips a fence post and flips over. Lewandowski’s Ghost Rider wobbles for a few heroic seconds, then tips over right at the starting line. By the end of the first mile, one vehicle is on its side and another’s tangled in barbed wire. By mile seven, only Carnegie Mellon Sandstorm is still moving. But soon, even the Humvee finds itself stranded on a mound of dirt, its wheels spinning helplessly. By the time the race is over, not a single vehicle has finished the course. Sandstorm made it the furthest, 7.4 miles. (Time 0:10:05)
  • LIDAR Became A Defining Sensor Choice
    • LIDAR emerged as a powerful way to map environments in 3D, giving teams a sensing advantage.
    • That sensing choice shaped early technical approaches and future commercial bets. Transcript: David Brown A beige Humvee named Sandstorm idles nearby. It was built by a team from Carnegie Mellon University in Pittsburgh, including a young roboticist named Chris Urmson. Unlike Lewandowski’s camera-based system, the Carnegie Mellon crew relies on something new, LIDAR. That stands for Light Detection and Ranging, and it’s a sensor that fires thousands of laser pulses a second, measuring how long each takes to bounce back. LIDAR allows Sandstorm to build a three-dimensional map of the world in real time. The Marshal waves a flag and the robotic vehicles lurch forward. Chaos ensues. One truck veers into a barrier. Another plows into a ditch. A third clips a fence post and flips over. Lewandowski’s Ghost Rider wobbles for a few heroic seconds, then tips over right at the starting line. By the end of the first mile, one vehicle is on its side and another’s tangled in barbed wire. By mile seven, only Carnegie Mellon Sandstorm is still moving. But soon, even the Humvee finds itself (Time 0:10:40)
  • Stanley Wins And Draws Google In
    • Stanley, built by Sebastian Thrun’s team, won the 2005 Grand Challenge and caught Larry Page’s attention.
    • Page invited Thrun to Google and pushed to pursue self-driving cars aggressively. Transcript: David Brown They’re eager to prove that last year’s crashes were just the beginning of something bigger. But they’re not the only ones with something to prove. Their toughest competition is Stanley, a gleaming VW Touareg built with corporate backing from Volkswagen’s research labs in Palo Alto. At the center of Stanley’s team stands Sebastian Thrun, a 38-year Stanford professor with a boyish smile. He’s a German-born roboticist obsessed with machine learning and has spent years teaching computers to understand the real world. Volkswagen sees the project as a way to prove that its engineers can compete with Silicon Valley’s brightest minds. For AM General, which still supports the Carnegie Mellon team, it’s about demonstrating that its Humvees can handle anything. And for DARPA, this race is still an experiment, one it hopes will lead to military vehicles ferrying weapons and supplies without the need for human occupants. It’s science with big stakes. The prize pot has now doubled to $2 million. But whoever makes a well-functioning autonomous vehicle first won’t just win the cash. They’ll potentially own the future of transportation. The race begins. Engines roar, dust clouds bloom, and one by one, the vehicles head into the desert. This time, something extraordinary happens. Instead of collapsing after a few miles, nearly every team’s robot vehicle makes real progress. Machines that just 18 months earlier wobbled and crashed now carve perfect turns, break on slopes, and weave around boulders. All but one of the 23 entrants beat the seven-mile record set by Sandstorm last year. Five complete the entire course. But the prize comes down to two contenders, Sandstorm and Stanley. Sandstorm finishes second. Stanley crossed the line just 11 minutes faster. Among the spectators lining the Nevada Hills is one Larry Page, the co-founder of Google. He’s not just here for fun. He’s come to scout talent. The morning after the DARPA Grand Challenge, Sebastian Thrun is a hero. His modified Volkswagen, Stanley, has just driven 132 miles of desert terrain without human intervention, a breakthrough that’s caught the eye of Google’s co-founder. Page asks Thrun to visit Google’s headquarters in Mountain View, California. It’s the first of dozens of meetings at Googleplex, in which the pair discuss and debate the promise of self-driving cars, and how they just might change the world. To Page, this technology isn’t just a stunt. (Time 0:16:30)
  • Reframing: Build A Driver, Not A Car
    • Larry Page reframed the problem: don’t map the world, fix the driver inside the car.
    • That vision prioritized building an onboard ‘driver’ as Google’s strategic north star. Transcript: David Brown To Page, this technology isn’t just a stunt. It’s the next step in Google’s corporate mission to organize the world’s information. He believes that the most chaotic, dangerous information system on Earth is traffic. Autonomous cars, he says, could eliminate human error on the roads and reduce crashes by as much as 90%. And he wants Thrun to help Google make that happen. Thrun listens with mounting irritation. The DARPA races took place on a controlled desert course, where there were no pedestrians, no delivery vans blocking the lane, no road work, no traffic lights. But the real world is different. It’s unpredictable, messy, alive. Thrun tells Page that trying to build a robot that can handle this chaos is madness. Page doesn’t blink. He asks for a technical reason not to pursue autonomous driving. Thrun loses his temper, yelling, it can’t be done, goddammit! This right here is the friction between a manager and a visionary. A manager looks at the current capabilities and sees roadblocks. A visionary like Paige looks at the ultimate destination and ignores the road entirely. Paige isn’t interested in better. He’s interested in different. He’s reframing the problem from how do we map the world to how do we fix the driver. And usually, the person asking the impossible question is the one who ends up owning the future. Thrun leaves the meeting fuming. (Time 0:19:25)
  • Larry’s 1K Goal Accelerates Testing
    • Larry Page set the “Larry’s 1K” target to log 1,000 complex autonomous miles fast.
    • The team broke the problem into 10-mile segments to iterate toward the goal. Transcript: David Brown Google co-founder Larry Page has given them a simple but brutal target, and he’s put real money behind it. If the Project Chauffeur team can successfully have an autonomous car drive 1,000 miles of complex California roads, or 100,000 miles total within two years, Page has promised the Engineers enormous bonuses. The milestone becomes known as Larry’s 1K. Thrun, Urmson, and Lewandowski start by breaking the task into pieces. Instead of one long route, they design a patchwork of 10-mile segments, each one chosen to challenge the project’s driving software in a different way. They choose a stretch of freeway to assess its merging and lane-keeping, a climb into the Santa Cruz Mountains to teach it how to break on hills, and a run through the famously windy Lombard Street in San Francisco, involving eight hairpin bins packed into a single city block to see if their machine can manage chaos. Ridges, tunnels, intersections, roundabouts, each segment becomes a new test of judgment, timing, and perception. For a human, these moments are instinctive. For a machine, they’re puzzles to be solved. This is the classic answer to the question, how do you eat an elephant? Perhaps you’ve heard this one before. One bite at a time. See, if you give a team a massive vague goal like build a robot driver, chances are they’ll freeze, but if you break it down, you turn an impossible mountain into a series of climbable Steps. These small winds keep the momentum going, especially when the finish line seems like it’s still miles away. Day after day, the Project Chauffeur team loads up their modified Toyota Priuses and sends them out onto California’s back roads, with their spinning LiDAR domes capturing 3D maps Of the world around them. Each time they tweak the car’s software, the car manages to drive a little farther, before handing control back to the human in the driver’s seat. The gains are incremental. (Time 0:25:29)
  • Stealth Tests And First Public Road Trial
    • Engineers secretly tested modified Priuses around Mountain View, once boxed in by human-driven convoy for safety.
    • A public road trial on Central Expressway marked their first successful real-traffic test. Transcript: David Brown At night, the engineers work late, taking their modified Toyota Priuses around the streets near the Google campus. The cars hum through the dark with spinning LiDAR domes on their roofs, alien shapes gliding past convenience stores and sleeping suburban homes. Officially, Project Chauffeur doesn’t exist. Secrecy is paramount. The handful of people who do know about it refer to it only by codename. The rest of Google has no idea that the multi-billion dollar company they work for is quietly teaching cars to drive themselves right outside its own parking lot. But every trip beyond Google’s campus brings the risk of exposure. One night, near Shoreline Boulevard, a five-minute walk from Google HQ, two engineers pull over to adjust a sensor. They crouch beside the modified Prius and hook the car up to their laptop through cables that snake across the curb. But then, blue lights flash behind them. A Mountain View police officer steps from his cruiser. He looks from the open laptop to the spinning laser dome on the vehicle and demands to know what the two men are doing. For a moment, neither engineer speaks. Google’s top secret project is about to be blown open by a night patrolman. Then, one of the engineers straightens up, smiles, and in Russian-accented English, calmly tells the cop they work for Google. The officer studies them, shrugs, and says, and drives away. The engineers exhale. Google’s big secret is still intact for now. By the spring of 2009, Google’s self-driving project is finally ready to face the real world. For months, the team has been testing their self-driving systems in empty parking lots and on deserted roads. But now, they’re ready for their first public road trial on the busy Central Expressway that runs straight through Silicon Valley. Chris Urmson sits in the driver’s seat. He’s not driving. He’s here as a backup, ready to take the wheel if anything goes wrong. To keep things safe, Urmson has enlisted a few fellow Google engineers in their own cars. They line up ahead and behind the test vehicle, forming a protective convoy across every lane of the expressway, boxing out other traffic so the automated Prius can glide down the road In peace. For a few glorious minutes, it works. The car holds its lane, changes speed, and behaves exactly as programmed. For the first time, Google’s self-driving system has handled real traffic on a real Silicon Valley highway. (Time 0:28:15)
  • Harsh Road Tests Lead To 1K Success
    • A tense coastal test made the Prius brake sharply near cliffs but never crossed lines.
    • By 2011 the team met the 1,000-mile target and earned Larry Page’s bonuses. Transcript: David Brown That same spring, a silver Toyota Prius hums north along Highway 1, hugging the edge where the California coastline plunges into the Pacific. Inside the car sits Chris Urmson and Anthony Lewandowski. To their right, a sheer cliff drops to the rocks below. The stakes for this road test are higher, an escalation of the short trips they’ve been doing with their autonomous car for the past few months. Irmsen keeps his hands an inch from the wheel, jaw tight. Lewandowski grins like a kid at a roller coaster launch. Relax, it’s only our lives on the line. That’s supposed to make me less nervous. Behind them, in a human-driven car, a teammate hunches over a laptop, monitoring live readings from the Prius, ready to remotely adjust its position if the software drifts off course. But staying in lane is only part of the challenge. The Prius’s onboard LiDAR scans cliffs, guardrails, and passing traffic, building a three-dimensional map of everything around them. Then, the computers beneath it turn all of this data into judgments, like when to brake, when to steer, and when to trust the road. The Prius speeds towards a blind bend. Urmson’s fingers twitch, anticipating the brake. But nothing happens. Then, the car slams to a stop so late it pitches both men forward against their belts. Oof. A little eager on the timing, don’t you think? Lewandowski grins. Hey, at least it noticed the cliff. With the bend tackled, the Prius surges forward again, reaches the next curve, and breaks just as abruptly. It’s a ride that’s jerky, unnerving, and miraculous. For all the lurching, the robot never crosses the line. The ride is rough, but resolute. By 2011, the Project Chauffeur team has met the Larry 1K Challenge, logging 1,000 miles of complex autonomous driving across California. (Time 0:31:25)
  • Moonlighting Sparks Internal Conflict
    • Anthony Lewandowski founded Odin Wave and began moonlighting on LiDAR outside Google.
    • Chris Urmson saw this as disloyal and urged Google HR to remove him from the project. Transcript: David Brown It’s one year later, and the tension inside Google’s self-driving division has reached its breaking point. Officially, Project Chauffeur is thriving. But behind the glass walls, the mood has soured beyond repair. Anthony Lewandowski wants to move faster. He believes the team has spent too long perfecting prototypes instead of putting real cars on the road. Outside Google, Tesla’s autopilot is stealing headlines and clocking millions of semi-autonomous miles. Uber, General Motors, and Ford are all building their own driverless programs. Lewandowski feels Google is losing ground, and he’s not waiting for permission to act. He’s founded a new company, Odin Wave, a startup that’s developing its own LiDAR sensors, the laser eyes at the heart of Google’s self-driving vehicles. Lewandowski insists this work is separate from Google, but the overlap is obvious. He’s also trying to get his colleagues at Project Chauffeur to join him. When word reaches Chris Urmson, it’s the final straw. For Urmson, Lewandowski’s moonlighting isn’t just disloyal. It’s a direct threat to Google’s most sensitive technology. He fires off an email to Google HR, urging that Lewandowski be removed from the project immediately. The partnership at the heart of Google’s self-driving revolution is broken. (Time 0:42:20)
  • The Data Copy That Ignited A War
    • Lewandowski copied roughly 14,000 proprietary files from Google onto an external drive before resigning.
    • He then launched Otto and later sold it to Uber, igniting a major legal and corporate conflict. Transcript: David Brown For months, Anthony Lewandowski has argued with his bosses, fought with colleagues and watched as his dream of commercializing self-driving cars has drifted further out of his control. Now, he’s reached a decision. He double-checks he’s not being watched, then slides a portable external hard drive into his workstation. On the screen, directories flick open. LiDAR schematics, circuit designs, calibration data, proprietary mapping files. He drags the documents onto his hard drive. Line by line, gigabytes of confidential data from Project Chauffeur begin to copy over roughly 14,000 files, including instructions for calibrating and tuning Google’s custom LiDAR. Lewandowski watches the progress bar inch forward, his eyes darting to the door. When it’s done, he ejects the drive, pockets it, and powers down his monitor. (Time 0:46:31)
  • Uber Buys Otto To Buy Time
    • Uber bought Otto for about $680 million and used it to demonstrate autonomous truck deliveries.
    • The acquisition accelerated Uber’s capabilities and deepened rivalry with Google. Transcript: David Brown In August 2016, Uber buys Otto, just 10 months after Lewandowski started the business. The deal is reportedly worth $680 million, and Uber doesn’t waste any time proving the commercial potential of its new acquisition. A few months later, Uber makes the world’s first autonomous truck delivery by transporting 50,000 cans of Budweiser 120 miles. Inside Google’s self-driving team, jaws drop. Lewandowski has barely walked out the door, and now he’s joined their biggest rival, taking the expertise he helped build with him. For Uber’s CEO, Travis Kalanick, buying auto is a masterstroke. Google may have the science, but Uber has the market share. And now, thanks to Lewandowski, it has the brains, too. For Google, it’s a gut punch. Its prodigy has crossed the aisle and given a leg up to the very company racing against them. Let’s look at the strategy here. Uber is late to the party and they know it. They have two choices. Spend five years trying to build this technology from scratch or write a massive check to buy the team that already knows how to do it. This is the classic buy versus build calculation. Travis Kalanick figures that in a winner-takes market, time is actually more expensive than cash. So he didn’t just buy a truck company. (Time 0:47:53)
  • Waymo Emerges As A Standalone Player
    • Google spun Project Chauffeur into Waymo to become a standalone mobility business under Alphabet.
    • The field quickly turned competitive as Uber, Tesla, GM, and others raced to commercialize autonomy. Transcript: David Brown Two months later, in a sleek glass auditorium at Google HQ, a new face steps on stage. John Krafchek, a former Hyundai executive, adjusts his microphone and smiles at the packed room of journalists. Behind him, one word fills the giant screen. Waymo. Over the past year, Google’s self-driving project has changed its shape and personnel. First, Lewandowski left to start auto. A few months later, Chris Urmsen left Google as well. Now, the project is changing its name as the experiment finally becomes a business. Kraft Check tells reporters the team is spinning out from Google as a standalone company. Waymo will still be by Google’s alphabet, but it will have its own identity and mission. The new name, he explains, means a new way forward in mobility. After a decade of research, Google’s moonshot has finally reached its destination. Waymo is ready to bring robo-taxis to the streets. But it won’t have the road to itself because Uber, armed with Otto and Anthony Lewandowski, has just pulled into the fast lane. (Time 0:49:39)