The Levels That Actually Matter
The industry measures autonomy in levels, from 0, no automation, to 5, full autonomy anywhere with no human needed. Most cars sold today sit at Level 2, driver-assist features like lane-keeping and adaptive cruise control that still require full driver attention. The handful of true robotaxi services operate at Level 4: fully autonomous, but only within a mapped, approved geographic area.
Why Level 3 Barely Exists
Level 3, where the car drives itself but a human must be ready to take over within seconds, turned out to be one of the hardest to make safe in practice, because handing control back to a distracted human on short notice is itself dangerous. Most companies have leapfrogged straight from Level 2 assistance to Level 4 robotaxis operating in tightly controlled zones instead.
How These Cars Actually See the Road
Autonomous vehicles combine several types of sensors: cameras for visual detail, radar for detecting speed and distance through rain or fog, and often lidar, which bounces laser pulses off surroundings to build a precise 3D map. Onboard computers fuse this data in real time, comparing it against detailed pre-built maps of the streets the car is certified to drive on.
What's Actually Working Today
- Geofenced robotaxis. Services operate fully driverless rides within specific, heavily mapped cities, expanding gradually block by block.
- Highway driver-assist. Increasingly capable hands-off systems on major highways, still requiring driver attention.
- Freight and yard automation. Autonomous trucking and warehouse-adjacent driving in more controlled, predictable environments than city streets.
What Still Makes City Driving Hard
Unprotected left turns, aggressive human drivers, construction zones that do not match the map, and split-second judgment calls around pedestrians remain genuinely difficult. These are not edge cases in a busy city, they are a normal Tuesday, and a system needs to handle thousands of them flawlessly to be trusted at scale.
Weather adds another layer: heavy snow can blind cameras and confuse lidar, and construction can outdate a map overnight. Progress here is steady but incremental, each new city a company expands into requires months of mapping and testing before driverless rides begin.
The Regulatory Piece
Technology alone does not determine the rollout pace; regulators approve autonomous operation city by city, often requiring a defined incident-reporting process and a period of supervised testing with a safety driver before fully driverless rides are allowed. That means the spread of robotaxis looks less like a single global launch and more like a slow, city-by-city expansion.
Key Takeaways
- Most cars on the road today are Level 2, driver-assist, not truly self-driving.
- True driverless robotaxis exist but operate only within specific, well-mapped cities, Level 4, not Level 5.
- Cameras, radar, and lidar work together, cross-checked against detailed pre-built maps.
- Unpredictable city situations, construction, aggressive drivers, bad weather, remain the hardest unsolved problems.
- Expansion is regulatory and city-by-city, not a single global switch-flip.
