Self-driving cars were supposed to be everywhere by 2020. Then 2022. Then 2025. The predictions kept moving while the technology kept improving — just not as fast, as universally, or as safely as the hype promised.
In 2026, autonomous vehicles exist. They are real, they are operating commercially, and they are nothing like the fully autonomous future that was advertised. Here is an honest status report.
The autonomy levels (what the numbers mean)
Level 0: You drive. No assistance. Level 1: One function assisted (adaptive cruise control). Level 2: Multiple functions simultaneously (Tesla Autopilot, GM Super Cruise). You must pay attention. Hands on wheel (or monitored). Level 3: Car drives in specific conditions; you can look away but must be ready to take over. Mercedes Drive Pilot (limited highways, Germany and Nevada). Level 4: Fully autonomous in defined areas/conditions. No human needed within geofence. Waymo, Cruise (before suspension). Level 5: Fully autonomous everywhere, all conditions. Does not exist. May not exist for decades.
Most consumer “self-driving” is Level 2. The robotaxis are Level 4. The gap between them is enormous.
Who is actually operating autonomously
Waymo (Alphabet/Google)
Status: Commercial robotaxi service in San Francisco, Phoenix, Los Angeles, and Austin. Expanding to Atlanta and Miami.
Fleet: 700+ vehicles (Jaguar I-PACE). Over 50 million autonomous miles driven.
How it works: Geofenced areas with extensive pre-mapping. Lidar, radar, cameras. No human driver. Remote operators can assist in edge cases but do not drive.
Safety data: Waymo reports significantly fewer serious incidents per mile than human drivers in their operating areas. Independent analysis (Swiss Re study, 2024) supports this claim for their geofenced operations.
Limitations: Only works in mapped cities. Struggles with construction, unusual weather, and unmapped areas. Expensive ($50K+ sensor suite per vehicle).
Tesla Full Self-Driving (FSD)
Status: Level 2 driver assistance marketed as “Full Self-Driving.” Available to Tesla owners as subscription ($99–199/month). Uses camera-only approach (no lidar).
Claims vs. reality: Tesla and Musk have repeatedly predicted full autonomy “within a year” since 2016. Has not materialized. FSD requires constant driver supervision. NHTSA has investigated multiple fatal crashes involving Autopilot/FSD.
Approach: Camera-only vision system trained on billions of miles of Tesla fleet data. Impressive in highway and suburban settings. Struggles with edge cases, construction, emergency vehicles, and scenarios requiring depth perception that lidar provides.
Regulatory pressure: Multiple NHTSA investigations, recalls, and lawsuits. California DMV accused Tesla of misleading marketing. FSD is not autonomous by any regulatory definition.
Cruise (GM)
Status: Operations suspended since October 2023 after a pedestrian dragging incident in San Francisco. GM has invested $10 billion. Resuming limited operations with safety drivers in 2025.
Lesson: One serious safety failure can halt an entire program. Public trust is fragile.
Others
Zoox (Amazon): Purpose-built autonomous vehicle (no steering wheel). Testing in Las Vegas and San Francisco. No commercial service yet.
Mobileye (Intel): Camera-based approach. Robotaxi pilots in Munich and Detroit. Conservative timeline.
Chinese players: Baidu Apollo, Pony.ai, WeRide operating robotaxis in Beijing, Guangzhou, and other Chinese cities with government support. Less public data on safety.
What works today
- Highway adaptive cruise + lane keeping (Level 2) — genuinely useful, reduces fatigue on long drives. Available in most new cars.
- Robotaxi in geofenced cities (Level 4) — Waymo in SF and Phoenix works for defined trips. Not perfect, but functional.
- Automated parking — many vehicles park themselves in controlled environments.
- Highway trucking automation — limited routes with hub-to-hub autonomous trucking (Aurora, TuSimple paused US ops).
What does not work today
- Full self-driving for consumers — no car you can buy drives itself everywhere
- Rural or unmapped areas — autonomous systems require pre-mapped environments
- All weather conditions — heavy rain, snow, and fog significantly degrade camera and lidar systems
- Construction zones — temporary lane changes, workers, and unmapped obstacles challenge all systems
- Ethical edge cases — the trolley problem is real engineering, not philosophy. Who is liable when an autonomous vehicle must choose between bad outcomes?
The safety debate
Pro-autonomy argument: Human drivers kill 1.35 million people globally per year. Even imperfect autonomous systems that are marginally safer would save thousands of lives annually. Waymo’s data supports this in their operating domains.
Anti-autonomy argument: One high-profile failure (Cruise pedestrian incident, Tesla Autopilot crashes) destroys public trust. Autonomous systems fail in unpredictable ways — humans fail in predictable ones. Mixing autonomous and human drivers creates new failure modes.
The honest position: Level 4 robotaxis in geofenced, mapped, favorable-weather urban environments appear safer than average human drivers in those same environments. Everything else remains unproven at scale.
Timeline (revised, again)
| Timeframe | Realistic expectation |
|---|---|
| Now–2028 | Level 4 robotaxis in more US cities (Waymo expansion). Level 2 in most new cars. |
| 2028–2032 | Level 4 in 10–20 major cities globally. Level 3 on highways (limited). First trucking routes. |
| 2032–2040 | Level 4 potentially common in urban areas. Level 5 still unlikely. Consumer ownership of autonomous vehicles possible in limited form. |
| 2040+ | Level 5 possible but uncertain. Depends on AI breakthroughs, regulation, and infrastructure investment. |
What this means for you
Buying a car today: Level 2 assistance (adaptive cruise, lane keeping) is worth having for highway driving. Do not pay premium for “Full Self-Driving” expecting autonomy — it is Level 2 with a misleading name.
Using a robotaxi: Waymo in SF or Phoenix is a genuine experience of autonomous transport. Try it. The technology works within its constraints.
Urban planning: Cities preparing for autonomous vehicles (dedicated lanes, updated infrastructure, modified parking requirements) will adapt faster than those treating autonomy as distant future.
The bigger picture: Autonomous vehicles will transform transport — but incrementally, city by city, condition by condition, over decades rather than overnight. The revolution is real. It is also slow, expensive, and constrained in ways the 2015 predictions ignored.
The self-driving car is not here. The self-driving robotaxi in specific neighborhoods is. The gap between those two statements is where the entire industry lives — building the future one geofenced mile at a time.
Lumen is edited by Leo Hartmann. Related: Solid-State Batteries · AI Tools for Creatives