History of Self-Driving Cars: DARPA Grand Challenge to Waymo

TL;DR: Self-driving cars didn’t come out of nowhere — they have a 25-year history that started with US government research challenges, leaped forward when Google launched its self-driving project in 2009, became publicly visible with Tesla’s 2014 Autopilot release, and reached commercial reality when Waymo launched the first paid driverless rides in Phoenix in 2018. By 2026, Waymo runs roughly 500,000 paid robotaxi rides per week across 10 US cities and Tesla’s FSD software is on roughly 6 million cars. The big technical questions about how to build the technology aren’t fully settled even now.
Why read: Understanding how we got here makes the present make sense. Tesla vs Waymo isn’t just a corporate rivalry — it’s a continuation of philosophical splits that go back to 2004.
Best for: Anyone curious about how autonomous vehicles came to be, anyone trying to understand current AV industry debates with context, anyone tracking AI applied to physical-world problems.
Skip if: You already know the DARPA Grand Challenges by heart. Daily AI fundamentals in our free Beginners in AI newsletter.

Twenty-two years ago a US defense agency offered a $1 million prize for anyone who could build a robot car that could drive across a stretch of desert by itself. Nobody won. Five years later Google decided autonomous cars were worth a serious bet and started a research project. Today that research project is Waymo, the world’s largest driverless robotaxi service.

Here’s how self-driving cars went from impossible to inevitable, year by year.

What started this whole thing?

The modern self-driving-car story begins with DARPA — the US Defense Advanced Research Projects Agency, the federal organization that funds high-risk research for the Department of Defense. DARPA wanted technology that could let trucks resupply soldiers in war zones without a human driver. They couldn’t buy it (it didn’t exist), so they decided to spark its creation by offering big cash prizes to anyone who could build it.

The three competitions DARPA ran — called the Grand Challenges — produced the engineers, the techniques, and the inspiration for almost every company working in self-driving today.

The DARPA Grand Challenges (2004, 2005, 2007)

Year	Challenge	Result
March 2004	150-mile course in the Mojave Desert. $1M prize.	Nobody finished. Best vehicle made it about 7 miles before getting stuck. Embarrassing for everyone — but the lesson was that this was harder than anyone thought.
October 2005	132-mile desert course. $2M prize.	Five vehicles finished. Winner: “Stanley,” a modified Volkswagen Touareg from Stanford University led by Sebastian Thrun. This is the moment self-driving cars went from “maybe someday” to “this is real.”
November 2007	Urban Challenge: 60-mile course on a closed Air Force base with simulated traffic. $2M prize.	Six vehicles finished. Winner: “Boss,” a Chevy Tahoe from Carnegie Mellon University. Five of the six finishers used LiDAR sensors from a company called Velodyne — LiDAR became the standard sensor for autonomous driving from that day forward.

What this matters for today: The 2007 Urban Challenge is where Sebastian Thrun’s Stanford team and the Carnegie Mellon team developed most of the techniques modern AV companies still use. The Google self-driving car project hired heavily from both teams. Waymo’s engineering culture traces back to those competitions. Aurora’s co-founder Chris Urmson led the Carnegie Mellon team that won in 2007.

Google enters the game (2009)

In 2009 Google co-founder Larry Page personally hired Sebastian Thrun to run a new internal project. Goal: build a real self-driving car, not for a contest, but as the foundation of an actual product. The project ran in secret for about 18 months, then went public in October 2010 when Google announced it had been testing autonomous cars on California public roads.

Why this mattered: until Google’s announcement, self-driving cars were a research curiosity. Once Google — the most valuable tech company in the world at that point — was working on it seriously, every other major tech and auto company had to pay attention.

By August 2012, Google’s test fleet had driven 300,000 miles on public roads without causing an accident. By 2016 the project was spun out as its own Alphabet subsidiary called Waymo. See our Waymo profile for the modern-day deep dive.

Tesla makes self-driving consumer-visible (2014)

While Google was building research vehicles, Tesla took a different approach: ship driver-assistance features on consumer cars and let real customers use them on real roads. The first Tesla Autopilot release was October 2014 — basic adaptive cruise control plus lane-keeping on highways. By 2016 Tesla had added “Full Self-Driving” capability to its product roadmap (the marketing language was controversial even then, given the system was still officially driver-supervised).

The Tesla approach — vision-only sensors (cameras, no LiDAR), end-to-end deep learning (one big AI rather than rules-and-modules), data collected from a giant fleet of consumer cars — was the opposite of how Google / Waymo built their system. The Tesla-vs-Waymo philosophical debate has shaped the industry ever since. See our Tesla FSD profile and Vision-Only vs Sensor Fusion debate.

The crowded middle (2014–2020)

The mid-2010s were when basically every car company and tech giant decided they needed an autonomous-vehicle program. The list got long fast:

• Cruise founded 2013; acquired by General Motors for $1B in March 2016.
• Uber ATG (Advanced Technologies Group) launched 2015 in Pittsburgh.
• Lyft Level 5 autonomous division started 2017.
• Aurora Innovation founded 2017 by Chris Urmson (ex-Google), Sterling Anderson (ex-Tesla), and Drew Bagnell (CMU). See our Aurora profile.
• Argo AI founded 2016, backed by Ford, then later Volkswagen.
• Zoox founded 2014; later acquired by Amazon in 2020 for $1.2B.
• Pony.ai founded 2016 (US/China); now operates Chinese robotaxi services.
• Baidu Apollo launched 2017 as an open-source AV platform.
• TuSimple founded 2015 focused on autonomous trucking.
• Embark founded 2016, focused on trucking.
• Plus (Plus.ai) founded 2016, trucking.
• Kodiak Robotics founded 2018, trucking.
• Mobileye — founded much earlier (1999) but acquired by Intel in 2017 for $15.3B specifically to anchor Intel’s AV strategy. See our Mobileye profile.
• Wayve founded 2017 in London by Alex Kendall and Amar Shah, pioneering end-to-end AI for driving. See our Wayve profile.

Three things drove this 2015–2020 boom: (1) Google/Waymo had proven autonomous cars could work, (2) Tesla had shown consumer demand was real, and (3) every major auto company assumed they had to play or they’d be left behind. Venture capital flooded in. The valuations were enormous.

First commercial driverless rides (Phoenix, December 2018)

December 5, 2018: Waymo launched Waymo One in the Phoenix metro area — the first commercial self-driving taxi service in the United States. Initially the cars still had a safety driver behind the wheel. By 2020, the Phoenix service became fully driverless for customer rides (no safety driver in the car).

This was the moment self-driving stopped being a promise and became a product. Real customers could request a ride and a real driverless car would show up.

The shakeout (2021–2023)

Self-driving was harder than the 2015–2020 boom assumed, and harder than the venture-capital valuations could justify forever. The shakeout came:

• 2020: Uber sold its ATG division to Aurora. Uber was done trying to build its own self-driving stack.
• 2021: Aurora went public via SPAC merger at a valuation around $13B.
• 2022: Argo AI shut down. Ford and Volkswagen, its backers, both decided they couldn’t justify the burn rate.
• 2023: Embark shut down. TuSimple delisted from NASDAQ after a difficult restructuring, eventually focusing on China-only operations.
• October 2023: A Cruise vehicle in San Francisco dragged a pedestrian who had been struck by a different vehicle. The incident became a public scandal; California regulators suspended Cruise’s permits; the company halted operations nationwide. Cruise has substantially reduced its ambitions since.

By the end of 2023 the AV industry looked very different from what people had imagined in 2018. The companies still standing were the well-funded ones with operational track records: Waymo at the top, then Aurora (in trucking), Mobileye (in ADAS chips), Tesla (in consumer cars). The dozens of mid-stage companies of the boom era had thinned dramatically.

The current era (2024–2026)

• 2024: Waymo expanded from Phoenix to San Francisco, Los Angeles, and Austin. The Tesla “We, Robot” event unveiled Cybercab. Tesla launched its Austin Robotaxi pilot in mid-2025.
• 2025: Aurora launched commercial driverless trucking between Dallas and Houston. Waymo continued geographic expansion. NHTSA opened investigations into Tesla FSD covering 2.88–3.2 million vehicles.
• February 2026: Waymo raised $16 billion at a $126 billion valuation. The Wayve+Uber partnership announced plans for public robotaxi trials in London during 2026.
• April 2026: Tesla released FSD v14.3.2, a unified end-to-end neural network. Mobileye reported Q1 revenue of $558 million (+27% YoY).
• May 2026: The industry now has multiple commercial driverless services at scale (Waymo One in 10 metros, Aurora trucking in Texas), with Wayve+Uber and Tesla Robotaxi expansion pending.

Why is this still hard?

If self-driving were a solved problem, every car would be self-driving by now. It’s not. The remaining hard problems:

• Edge cases. A child in a Halloween costume. A reflective puddle. A construction sign upside-down. A police officer waving you into the oncoming-traffic lane. The variety of weird-but-real driving situations is enormous, and AVs have to handle all of them.
• The safety bar is much higher than humans’. Society broadly accepts that human drivers cause ~40,000 US road deaths per year. It will not accept the same crash rate from AVs. The expected safety bar for AVs is significantly better than humans, not equal.
• Weather and degraded conditions. Snow, heavy rain, fog, dust, low sun, night with no streetlights — every AV company struggles in these conditions. Snow especially hasn’t been solved at Level 4.
• Vision-only vs sensor fusion isn’t settled. See our deep dive at Vision-Only vs Sensor Fusion.
• Regulation varies state-by-state. The FAA owns drone airspace; vehicle regulation is split between NHTSA federally and state DMVs locally. Each state can have different AV rules.
• Legal exposure. When (not if) an AV is in a fatal crash, the manufacturer faces liability that doesn’t exist for human-driver crashes. The legal framework for AV liability is still being built.

FAQ

Who invented self-driving cars?

The modern era of self-driving cars came from the DARPA Grand Challenges starting in 2004. The key technical pioneers were the Stanford and Carnegie Mellon university teams led by Sebastian Thrun and Chris Urmson respectively. Their work in 2004–2007 produced the engineers and the techniques that every modern self-driving company uses.

When did self-driving cars become real?

Depends on definition. Real commercial driverless rides for paying customers: December 2018 (Waymo Phoenix). Real driverless rides at scale (hundreds of thousands per week): 2024–2026 with Waymo expansion. Real driverless cars on consumer roads everywhere: not yet, and unclear when (depends on whether Tesla’s vision-only approach reaches Level 4 at consumer scale).

What killed Cruise?

An October 2023 incident in San Francisco where a Cruise vehicle dragged a pedestrian who had been struck by a different vehicle. The California DMV suspended Cruise’s permits. GM (Cruise’s parent) restructured the program multiple times since. Cruise no longer operates customer-facing rides as of 2026.

Why does Tesla matter so much in this history?

Tesla made the “ship driver-assistance to real consumers and gather data from millions of cars” approach work. Even people who think Tesla’s technical choices (vision-only, end-to-end neural network) are wrong have to acknowledge that Tesla’s ~6 million-car installed base produces vastly more real-world driving data than Waymo or any robotaxi service can collect. Whether the data advantage is enough to overcome Tesla’s sensor limitations is the unresolved question of 2026.

Where can I read more about each company?

We have detailed profiles of each major AV company: Waymo, Tesla FSD, Mobileye, Aurora, Wayve. The Vision-Only vs Sensor Fusion post covers the philosophical debate that runs through the whole industry.

Where can I read primary sources?

DARPA published official Grand Challenge documentation at darpa.mil. Waymo’s own history at waymo.com. The Stanford Stanley team published academic papers (Thrun et al., 2006). The Carnegie Mellon Boss team published academic papers (Urmson et al., 2008). Both are accessible via Google Scholar.

The bottom line

Self-driving cars are a 22-year overnight success. They started with a US-government desert-driving competition in 2004, found their breakthrough at the 2007 DARPA Urban Challenge, took serious shape when Google launched a research project in 2009, became commercial reality with Waymo One in 2018, and reached meaningful scale in 2024–2026 with Waymo’s expansion and Aurora’s trucking debut.

The hard problems aren’t fully solved. The Tesla vs Waymo debate isn’t settled. The next 24 months produce the deployment data that starts to answer the remaining open questions. But the era of “self-driving is impossible” ended around 2018. The era of “self-driving works in specific places” is right now. The era of “self-driving everywhere by default” isn’t here yet, and might not ever look quite the way the 2018-era visionaries promised.

For broader context: Waymo Explained, Tesla FSD Explained, Mobileye Explained, Aurora Innovation Explained, Wayve Explained, Vision-Only vs Sensor Fusion debate, Computer Vision in Autonomous Vehicles. Daily AI fundamentals in our free Beginners in AI newsletter.

Sources

• DARPA, DARPA Grand Challenge 10 Years Later and 2004 Grand Challenge documentation — primary federal source for the foundational competitions.
• Stanford Racing Team (Thrun et al.), “Stanley: The Robot That Won the DARPA Grand Challenge” (2006) — the academic paper documenting the 2005 winning team. Available via Google Scholar / arXiv.
• Carnegie Mellon Tartan Racing Team (Urmson et al.), 2008 Journal of Field Robotics paper on “Boss” — the academic record of the 2007 Urban Challenge winner.
• Waymo, waymo.com and waymo.com/research — primary source for Waymo’s history and ongoing operational data.
• Tesla, Tesla blog and news — primary source for Autopilot/FSD release history.
• NHTSA, Automated Driving Systems — primary federal regulatory reference.
• SEC EDGAR — primary financial source for SPAC-merger filings of Aurora, Embark, and TuSimple, plus 10-Ks of public AV companies.
• Wikipedia, History of self-driving cars — background reference for the broader timeline (cross-checked against primary sources).
• Sebastian Thrun, “What we’re driving at” (Google official blog, October 2010) — the public announcement of the Google self-driving project.
• R&D World, DARPA Grand Challenge at 20+ years — industry-press retrospective with primary-source quotes.

You May Also Like

• Waymo Explained
• Tesla Full Self-Driving (FSD) Explained
• Mobileye Explained
• Aurora Innovation Explained
• Wayve Explained
• Vision-Only vs Sensor Fusion