History of Self-Driving Cars: From 1980s Labs to 2026 Robotaxis

The phrase "self-driving car" conjures images of steering-wheel-free pods gliding through city streets. But the history of self-driving cars stretches back decades before any rider ever summoned a robotaxi from a smartphone. Understanding that timeline matters for today's EV buyers and riders: it explains why Level 4 autonomy works brilliantly inside geofenced urban zones, why your Tesla's FSD Supervised still demands your full attention, and why the gap between the two remains one of the hardest engineering problems in transportation.

This guide walks through the key milestones, then grounds the discussion in what you can actually use right now — where robotaxis operate, what they cost, how safe the data shows them to be, and where the hard limits still are.

The Early Decades: Remote Control, Radio Waves, and DARPA

The earliest demonstrations of "driverless" vehicles weren't autonomous at all — they were remote-controlled. In 1925, electrical engineer Francis Houdina demonstrated a radio-controlled car through New York City traffic, though it crashed shortly after. Throughout the 1950s and 1960s, RCA Labs and General Motors experimented with embedded highway wires that guided specially equipped cars, envisioning an infrastructure-dependent form of automation that never scaled beyond test tracks.

The modern era of self-driving cars began in the 1980s with Ernst Dickmanns at the Bundeswehr University in Munich, whose team equipped a Mercedes van with cameras and early computer vision, achieving autonomous lane-keeping and obstacle detection at highway speeds. Carnegie Mellon's Navlab project pursued a parallel path in the US, culminating in "No Hands Across America" in 1995 — a 2,800-mile drive from Pittsburgh to San Diego with autonomous steering for roughly 98% of the trip (the driver still managed throttle and brakes).

Then came DARPA. The Defense Advanced Research Projects Agency's Grand Challenge in 2004 offered a $1 million prize for any vehicle that could complete a 142-mile desert course autonomously. No entrant finished. The longest run covered about 7 miles before getting stuck on a rock berm. A year later, DARPA ran the challenge again — and five vehicles completed the 132-mile course, with Stanford's "Stanley" (a modified VW Touareg) taking the $2 million prize. The 2007 DARPA Urban Challenge pushed further, requiring vehicles to navigate a simulated urban environment with traffic laws, intersections, and other vehicles. That event produced the core technical talent that would go on to found or lead Waymo, Cruise, Aurora, and Argo AI.

From Research Labs to Robotaxi Fleets

Google launched its self-driving car project in 2009, recruiting key engineers from the Stanford and Carnegie Mellon DARPA teams. By 2015, the project had logged over a million autonomous miles in Lexus RX450h hybrids and custom "Firefly" prototypes. In December 2016, Google spun the project out as Waymo, named for "a new way forward in mobility."

Uber entered the space aggressively, poaching 40 researchers from Carnegie Mellon's robotics center in 2015 and launching a pilot ride service in Pittsburgh in 2016 using Ford Fusion hybrids with roof-mounted lidar. A fatal crash in Tempe, Arizona in March 2018 — in which an Uber test vehicle struck and killed a pedestrian crossing outside a crosswalk — became a watershed moment for the industry, triggering a temporary industry-wide pause and forcing a reckoning with safety-case methodologies.

Meanwhile, GM's Cruise Automation division (acquired in 2016) began testing Chevrolet Bolt EVs in San Francisco, and by 2022 had received approval from the California Public Utilities Commission (CPUC) to operate a commercial driverless ride service. Amazon acquired Zoox in 2020 for roughly $1.3 billion, and the company has since developed a purpose-built, bidirectional robotaxi with no steering wheel.

China's Baidu launched Apollo Go in 2020 and by 2024 was operating the world's largest robotaxi fleet by ride volume, primarily in Wuhan, Beijing, and Chongqing. Tesla, despite CEO Elon Musk's repeated timeline predictions dating to 2014, has kept its Full Self-Driving system firmly at SAE Level 2 — requiring constant driver supervision — though the company has begun limited testing of a driverless-capable platform.

Understanding SAE Automation Levels

The single most important framework for understanding self-driving history — and today's market — is the SAE J3016 standard, which defines six levels of driving automation. Confusing these levels is the root of most consumer misunderstanding about what cars can actually do.

The critical takeaway: every consumer vehicle you can buy in 2026 — including a Tesla with FSD Supervised — is Level 2. You are the driver. You are liable. Level 4 robotaxis like Waymo operate in a fundamentally different category: the vehicle is the driver, the company bears liability, and no human intervention is expected or possible from the rider's seat.

Where Can You Actually Use a Robotaxi Today?

Robotaxi availability has expanded meaningfully through 2025 and into 2026, but it remains geographically constrained. Here is where you can hail a fully driverless (Level 4, no human behind the wheel) ride as of mid-2026:

Cruise, which was once operating in San Francisco, Phoenix, and Austin, suspended all driverless operations in October 2023 after a serious incident in San Francisco in which a pedestrian was dragged by a Cruise vehicle. GM announced significant restructuring of Cruise in 2024-2025, pivoting toward personal-vehicle ADAS rather than robotaxi operations. As of mid-2026, Cruise is not running commercial driverless rides.

If you live outside these service areas, Level 4 robotaxis are not yet available to you — regardless of what you may read in press releases about "expansion plans."

How to Use a Robotaxi and What It Costs

Using Waymo One — the most broadly available service — works almost exactly like Uber or Lyft:

1. Download the Waymo One app (available in service-area app stores).
2. Create an account; some markets require you to join a waitlist before gaining full access.
3. Enter your pickup and drop-off locations within the geofenced service area.
4. The app dispatches a vehicle; you'll see its location, ETA, license plate, and a PIN to unlock the doors.
5. Enter the PIN on a screen mounted in the rear seat. The vehicle will not move until all doors are closed and seatbelts are fastened.
6. A screen shows your route. You can request music, adjust climate, add a stop, or pull over (in safe locations).
7. Payment is automatic through the app upon arrival.

Pricing is competitive with UberX and Lyft — often within a few dollars on the same route. The real cost advantage shows up when you factor in that there's no driver to tip (though Waymo does offer in-app tipping that goes to cleaning and support staff). For frequent riders in service areas, Waymo also offers subscription-style ride packages that can bring per-ride costs down further.

Safety Data: What the Numbers Actually Show

Robotaxi safety is the most consequential question in the history of self-driving cars, and it's where hype most often outpaces evidence. Here is what the published data indicates as of 2026:

Key context for interpreting these numbers:

• Operating domain matters. Waymo operates in geofenced, relatively well-mapped urban environments at moderate speeds. Comparing its crash rate to the national average (which includes rural highways, bad weather, and drunk driving) would be misleading. The Swiss Re/IIHS comparison uses same-geography benchmarks, which is more rigorous.
• Cruise's 2023 incident. In October 2023, a Cruise vehicle in San Francisco struck a pedestrian who had been thrown into its path by a human-driven car, then dragged the pedestrian approximately 20 feet while attempting to pull over. The California DMV suspended Cruise's deployment permit, and the company suspended operations nationwide. The incident raised questions about how autonomous systems handle multi-vehicle, high-complexity edge cases — and about transparency with regulators.
• NHTSA standing general order. Since 2021, NHTSA has required all manufacturers and operators of Level 2+ and Level 4 systems to report crashes within 24 hours under a Standing General Order. This has created the most comprehensive public crash database for automated vehicles, though data granularity and reporting consistency remain ongoing challenges.
• Severity vs. frequency. Robotaxis tend to be involved in low-speed, low-severity contacts (often rear-ended by human drivers in stop-and-go traffic) at higher rates per million miles than some benchmarks suggest, but at significantly lower rates for serious injuries and fatalities. The safety case for robotaxis rests primarily on eliminating impaired, distracted, and fatigued driving — the leading causes of traffic deaths.

Technology Stack: How Robotaxis Actually See the World

The history of self-driving cars is, in many ways, a history of sensor evolution. Today's Level 4 robotaxis use a multi-modal sensor suite — a fundamentally different approach from the camera-only system Tesla uses for its Level 2 FSD.

The camera-only approach (Tesla's vision system) argues that since humans drive with eyes alone, cameras plus neural networks should suffice. LiDAR proponents counter that redundancy and direct distance measurement are essential for safety-critical systems, especially given that camera performance drops in low light, heavy rain, and direct sun. Every commercially operating Level 4 robotaxi fleet in 2026 uses LiDAR as a core sensor.

Geofencing is the other critical technology that doesn't get enough attention. Level 4 robotaxis don't drive "anywhere." They operate within precisely defined operational design domains (ODDs) that specify approved roads, speed limits, weather conditions, and time-of-day constraints. If you request a ride that would take the vehicle outside its geofence, the app simply won't allow it. This is both the system's greatest strength (it only operates where it's been exhaustively validated) and its greatest limitation (scaling to new cities requires millions of miles of mapping and testing).

Regulatory Landscape: Who Approves Robotaxi Operations

In the United States, robotaxi regulation is a patchwork of federal and state authority:

• NHTSA (federal): Regulates vehicle safety standards (FMVSS). Has granted limited exemptions for vehicles without steering wheels or mirrors (e.g., Zoox, Nuro). Maintains the crash-reporting Standing General Order. Does not approve driving software or operations.
• State DMVs and PUCs: The real gatekeepers. California's DMV issues testing permits; the California Public Utilities Commission (CPUC) issues commercial deployment permits. Arizona takes a notably permissive approach — no state permit is required for testing or deployment, which is why Waymo launched commercially in Phoenix first. Texas, Nevada, and other states have their own frameworks.
• Local governments: Cities can impose additional requirements (pickup/drop-off zones, airport access agreements, data-sharing mandates). San Francisco's Board of Supervisors has at times pushed back against expansion, citing traffic disruption concerns.

This fragmented regulatory environment means that a robotaxi company may be fully approved in Phoenix, conditionally approved in San Francisco, and unable to operate at all in Miami — not because the technology differs, but because the regulatory pathway hasn't been established.

Current Limitations: What Robotaxis Still Can't Do

The history of self-driving cars is also a history of promises that arrived later than expected. Honest assessment of 2026 limitations:

The Road Ahead: What History Suggests About Timelines

If the history of self-driving cars teaches one lesson, it's that the last 10% of any autonomy problem is dramatically harder than the first 90%. The DARPA Grand Challenge went from zero finishers to five in one year — but going from "works most of the time in a desert" to "carries paying passengers safely in downtown San Francisco at night in the rain" took 17 years and billions of dollars.

For 2026 and beyond, the realistic trajectory looks like this: continued city-by-city expansion of Level 4 services, with each new city requiring 12-24 months of mapping, testing, and regulatory approval. Purpose-built vehicles (no steering wheel, optimized for rider comfort) will increasingly replace retrofitted production cars. Pricing will continue to converge with rideshare as fleet utilization improves. And Level 5 — the car that can drive you from a Manhattan garage to a rural Vermont cabin in a snowstorm — remains a research problem, not a product roadmap item.

For consumers, the practical question is not "when will self-driving be everywhere" but "is there a robotaxi in my city, and does the safety data support using it?" In Phoenix and San Francisco in 2026, the answer to both is increasingly yes.

Frequently Asked Questions