Self-driving cars are already here, but not in the way most people expected. Fully driverless robotaxis operate commercially in three U.S. cities right now, and one automaker sells a car that legally drives itself on highways at speeds up to 40 mph. The technology exists on a spectrum, though, and the version most people picture (a car that handles every road, in every condition, with no human involvement) is still years away.
What’s Available Right Now
The self-driving landscape in 2024 splits into two categories: cars you can ride in and cars you can buy.
On the ride-hailing side, Waymo One runs a fully autonomous robotaxi service, with no safety driver behind the wheel, in San Francisco, Los Angeles, and Phoenix. Anyone can download the app and take a ride 24/7. These vehicles operate within defined service areas, navigating city streets, intersections, and highway on-ramps using a suite of cameras, radar, and laser-based sensors.
On the ownership side, Mercedes-Benz sells the only car a consumer can buy that legally takes over driving on public roads. Its Drive Pilot system, certified in Nevada and California, handles freeway driving in heavy traffic at speeds up to 40 mph. When it’s active, you can legally take your eyes off the road and use the car’s touchscreen for browsing or other tasks. That’s a meaningful legal distinction: if the system is driving and something goes wrong, Mercedes accepts responsibility, not you.
Tesla’s Full Self-Driving software, despite its name, is classified as a Level 2 system. Tesla itself has told California regulators that FSD “continues to firmly root the vehicle in SAE Level 2 capability,” meaning the driver must stay attentive and keep hands near the wheel at all times. It can handle many driving tasks impressively, but you remain legally responsible every second it’s engaged.
The Levels of Autonomy, Explained
The automotive industry uses a six-tier scale (Level 0 through Level 5) to describe how much a car can do on its own. The jump from Level 2 to Level 3 is where the legal and practical shift happens.
- Level 2: The car can steer, accelerate, and brake simultaneously, but you must monitor the road at all times. Most “autopilot” and “self-driving” features on the market today fall here, including Tesla’s FSD.
- Level 3: The car drives itself under specific conditions, and you can disengage mentally. You must be ready to take over when prompted, but you’re not required to watch the road. Mercedes Drive Pilot is the only consumer system at this level in the U.S.
- Level 4: The car handles all driving within its defined operating area with no human fallback needed. Waymo’s robotaxis operate at this level. They don’t have a steering wheel option for passengers.
- Level 5: A car that drives itself everywhere, in all conditions, with no restrictions. This does not exist yet, and no company has announced a firm timeline for delivering it.
Why Full Autonomy Is Taking So Long
The core problem isn’t typical driving. Self-driving systems handle sunny highways and well-marked city streets reasonably well. The challenge is everything else. A U.S. Department of Transportation report on autonomous vehicles and weather lays out the obstacles clearly: fog, heavy rain, and snow severely degrade cameras, radar, and laser-based sensors. Snow covering the road makes lane markings invisible to sensors. Wet and icy surfaces create reflections that produce faulty readings. Even sunny conditions can cause glare that compromises sensor accuracy.
Falling snow can trigger proximity sensors to beep continuously, registering phantom obstacles where none exist. Heavy rain degrades camera images to the point of uselessness. And the communication links that allow vehicles to talk to each other and to traffic infrastructure are disrupted by thunderstorms, hail, and dense fog. The DOT report estimated that developing reliable fail-safe approaches for severe weather was a seven-to-ten-year goal, and that timeline hasn’t shortened much.
Beyond weather, there are “edge cases,” the rare, unpredictable situations that human drivers handle instinctively but algorithms struggle with. A traffic cop waving you through a red light. A mattress falling off a truck. A cyclist signaling a turn with a hand gesture. Each of these requires the kind of contextual judgment that remains difficult to automate reliably.
Safety: Better or Worse Than Humans?
The safety record is genuinely mixed and depends heavily on where and how you measure. In Mountain View, California, Google’s self-driving test cars logged a police-reportable crash rate of 2.19 per million miles between 2009 and 2015, compared to 6.06 per million miles for human drivers in the same area. That’s roughly a third the crash rate.
But a broader analysis found that autonomous vehicles actually have a higher accident rate per million miles when you account for the fact that they mostly operate in favorable conditions: clear weather, well-mapped urban areas, moderate speeds. They avoid snowy roads, rural highways, and other challenging environments where human crash rates spike. Comparing a robotaxi cruising through sunny Phoenix to a human driver navigating a blizzard in Minnesota isn’t an apples-to-apples measurement. The technology performs well within its boundaries, but those boundaries are still narrow.
Where the Laws Stand
More than 20 U.S. states have passed laws explicitly allowing the deployment of highly automated vehicles. States like Florida, Texas, Arizona, California, Georgia, and Nevada permit Level 4 and Level 5 vehicles to operate without a licensed driver in the car. Others, like Kansas and Arkansas, require a licensed driver to be present during an initial trial period before full driverless operation is allowed.
This patchwork means the legal landscape varies dramatically depending on where you live. A robotaxi that operates freely in Phoenix may not have permission to cross into a neighboring state. And most of these laws were written to enable commercial fleet services, not private ownership of fully self-driving cars. The legal framework for buying a Level 4 vehicle at a dealership and registering it like a normal car is still largely untested.
The Cost Factor
Price has been one of the biggest barriers to putting self-driving technology in affordable cars. The laser-based sensors used in robotaxis cost between $1,500 and $6,000 per unit, and a single vehicle needs several of them. For consumer cars with partial automation (Level 2 or Level 3), sensor costs range from $600 to $750 each, with total sensor packages exceeding $1,500. Entry-level versions designed for basic driver assistance are cheaper, around $150 to $300 per sensor, and those prices are expected to hold steady in coming years.
This explains why the most capable self-driving systems currently exist in expensive luxury cars or commercial fleet vehicles, not in the cars most people drive. As sensor costs continue dropping and processing power improves, the technology will filter down to more affordable models. But a $25,000 car that drives itself on the highway is still a ways off.
Realistic Timeline for Consumers
If you’re wondering when you’ll be able to buy a car that truly drives itself everywhere, the honest answer is: not soon. Level 3 highway driving in specific conditions is available now, but only on one vehicle (the Mercedes EQS and S-Class) and only in two states. Robotaxi services will likely expand to additional cities over the next few years, giving more people access to driverless rides without owning the technology themselves.
Level 4 vehicles that handle most driving in most conditions could realistically reach consumers in the late 2020s to early 2030s, though they’ll likely come with geographic and speed restrictions. Full Level 5 autonomy, a car that replaces a human driver in every possible scenario, remains the hardest engineering problem in the industry. No company has solved the weather problem, the edge-case problem, or the cost problem well enough to promise a date. The technology is advancing steadily, but the gap between “works most of the time” and “works all of the time” is where the hardest years of development still lie.