What if cars drove themselves?

A car needs a driver to steer, speed up and stop. Some new cars can help the driver.

What if all cars could drive themselves? What would they need to do?

Driving takes a lot of skill. A driver watches the road, reads signs and traffic lights, notices people and bikes, and decides in a split second when to steer, speed up or brake. Engineers are building cars that can do some of this by themselves.

So what if cars drove themselves? This is a thought experiment about a real technology that is still being developed. We'll look at how these cars sense the world, how they make decisions, and what experts think they might change, for better and for worse.

Automated driving combines perception (sensing the environment), prediction (anticipating what others will do), planning and control. Partial automation is already common in new cars; driverless operation exists only in limited areas.

This thought experiment asks what full automation would require and what it might change. It draws on the Insurance Institute for Highway Safety (IIHS) and Encyclopaedia Britannica, and labels what is established, what has been measured, and what is still a prediction or a debate.

Many cars today have helpers. Some can brake by themselves if they get too close to the car in front.

But a person still has to drive and pay attention.

Many new cars already have driver assistance features. Automatic emergency braking can stop the car if it's about to hit the vehicle ahead. Lane-keeping features warn the driver or gently steer if the car drifts. IIHS research found that forward collision warning plus automatic braking cut rear-end crashes in half.

But these features help a driver; they don't replace one. IIHS says truly driverless cars aren't for sale yet, and drivers will keep sharing the job for the foreseeable future.

Front crash prevention, lane departure warning, blind spot detection and rear crash prevention all show real-world crash reductions in IIHS and HLDI studies; forward collision warning with autobrake halved rear-end crashes (Cicchino, 2017), and pedestrian-detecting autobrake cut pedestrian crashes by 27% (Cicchino, 2022). In 2024 NHTSA finalized a rule requiring front crash prevention on nearly all new light vehicles by September 2029.

IIHS notes that most automakers offer partial (Level 2) automation, which never replaces the driver, and that it has not found a crash-reduction advantage for it beyond the crash-avoidance features themselves.

If cars drove themselves, people could ride without driving. Kids and older people might get around more easily.

But some experts worry about more cars on the road. Nobody knows for sure yet.

Supporters hope self-driving cars could make roads safer by removing human mistakes. Britannica notes they might also help children, older people and people with disabilities get around, and reduce the need for parking.

Critics point out possible downsides: people might travel more, adding traffic and pollution, and could choose to live farther from work. IIHS says that in theory fully automated driving could prevent the vast majority of crashes. Early results are encouraging: a 2026 IIHS study found that one company's driverless taxis were involved in 68 percent fewer police-reportable crashes than human drivers in the same cities. But those taxis drive only in set areas, so what would happen if every car drove itself is still a prediction, not a fact.

Projected benefits include crash reductions (IIHS: in theory, most crashes could be avoided with full automation), wider mobility for children, older adults and people with disabilities, and less need for parking (Britannica). Projected costs include more vehicle miles traveled, congestion, emissions and sprawl.

The net outcome depends on how safe the systems actually are, where and when they operate, and how people choose to use them. IIHS cautions that more automation does not guarantee more safety. Its 2026 study of one operator's Level 4 robotaxis found police-reportable crash involvement 68% lower, and injury crashes 81% lower, than human drivers in the same cities and years, and it calls for continued monitoring as deployment grows. Whether that would hold for every car, everywhere, is a prediction, not a result.

A self-driving car uses cameras, radar and lasers to look all around. A computer works out what to do.

It must spot cars, bikes, people and traffic lights, all the time.

Self-driving cars use several kinds of sensors at once. Cameras see lanes, signs and lights. Radar uses radio waves to measure how far away things are and how fast they're moving. Lidar, short for light detection and ranging, uses laser light to measure distances. GPS and detailed maps help the car know where it is.

Powerful computers combine all this into a 3-D picture of the surroundings and continually adjust steering, speed and braking. Using machine learning, the systems are trained on huge amounts of driving data so they can handle situations they weren't specifically programmed for.

Britannica describes the core loop: radar, GPS, cameras and lidar build a 3-D model of infrastructure, vehicles, pedestrians, traffic lights and signs; computers process it and continuously adjust steering, speed, acceleration and braking. Machine learning improves the algorithms from data, and AI lets the system decide without explicit instructions for every situation.

Connected-vehicle technology adds radio communication with other vehicles and infrastructure. The hardest problems are rare, unexpected events and knowing when the system is outside the conditions it was designed for.

Engineers sort cars into six levels, from 0 to 5. At level 0, a person does all the driving.

At level 5, the car could drive anywhere by itself. No car can do that yet.

Engineers use six levels, from SAE International, to describe how much a car drives itself. At Level 0 the human does everything. Levels 1 and 2 help with steering or speed, but the driver must stay alert the whole time. At Level 3 the car drives in certain conditions, but a person must be ready to take over. Level 4 cars drive themselves completely, but only in set areas. Level 5 would drive anywhere, in any conditions, and doesn't exist yet.

Today, driverless taxis at Level 4 carry passengers in some cities, but only inside set boundaries.

SAE levels, used by NHTSA: 0, no automation; 1, assistance with one function; 2, assistance with several, driver fully engaged; 3, conditional automation with a human on standby; 4, full control within a defined operating area, stopping safely if it reaches its limits; 5, full automation under all conditions.

IIHS notes there are no Level 5 systems on the road, and that everything available, in private cars or ride-hailing, is limited to specific roads and conditions. It also found that regular users of Level 2 systems often overestimate what they can do, a reminder that the human role matters at every level below 4.