The world was supposed to be riding in driverless cars by now. That was the vision laid out by tech leaders and automakers a decade ago.
But in 2025, almost everyone is still gripping the wheel, stuck in traffic. Most headlines are talking about fatal crashes and lawsuits.
So what happened?
Is full self-driving technology a false promise?
What happened to the 2020 robotaxi boom?
During a press call in 2016, Elon Musk proudly said that a Tesla would drive itself across the US by the end of 2017.
In 2015, Uber predicted full autonomy was just around the corner.
The 2015 headline was: “Are driverless cars the future of Uber?” In 2020, it was: “Uber Gives Up on the Self-Driving Dream”.
Ford once promised that it would launch a fully autonomous vehicle in 2021. None of these things happened.
The hype was not without substance. Billions were poured into autonomous vehicle (AV) startups.
Argo AI, which was backed by Ford and Volkswagen, hired thousands of engineers before being shut down in 2022.
Apple spent years and reportedly over $10 billion trying to build a self-driving car before scrapping the project in early 2024.
GM also scrapped its Cruise project in late 2024, after spending $10 billion since acquiring it 8 years earlier.
At the end of the day, technological ambition and real-world complexity will stand in the way of every promising, life-changing innovation.
Making a car stay in its lane on a highway is not that hard. But giving a vehicle the judgment and decision-making ability of a human driver in a chaotic urban environment is something else entirely.
The complexity of unpredictable events that require human intuition have stalled even the most advanced players.
Who’s actually doing it now?
As of mid-2025, only a handful of companies are deploying real autonomous services, and nearly all of them are limited in scope.
Alphabet-owned Waymo is the clear frontrunner. It has launched driverless robotaxi services in five US cities, including Phoenix and San Francisco.
But even Waymo’s service operates in carefully geo-fenced zones with good weather conditions and controlled traffic patterns.
There are no Waymo cars driving themselves through snowstorms or chaotic rural highways.
In China, Baidu’s Apollo Go and WeRide are rolling out robotaxis in major cities like Beijing, Shanghai, and Shenzhen. Pilot zones have expanded to 20 cities.
Goldman Sachs expects China to have over 500,000 robotaxis by 2030. But here too, the rollout is managed tightly by local governments. Safety, liability, and data regulations are still being drafted.
Tesla remains the most controversial name in the space. For a company claiming to be using a “Full Self-Driving” system, that system is not really autonomous.
It’s an advanced driver-assistance system (Level 2), requiring full driver supervision.
Tesla plans to launch a driverless robotaxi service in Austin this month, using a small fleet of modified Model Ys.
The system will be remotely supervised, but the company hasn’t demonstrated a fully autonomous vehicle that can operate independently across diverse road conditions.
The US still operates under a self-certification model where companies claim their vehicles meet standards.
In contrast, the EU uses type approval, where regulators independently verify safety.
But perhaps that’s changing, as the European Commission now considers easing AV rules in exchange for tariff relief in trade talks with the US.
It’s a remarkable policy shift, one that suggests economic pressure may be reshaping regulatory priorities faster than safety can keep up.
Why is it still so hard?
Technically speaking, the problem is not just software. It’s the entire system architecture. A true Level 4 or Level 5 system needs to operate safely even when something fails.
This is known as fail-operational design. It requires redundancy at every level: hardware, sensors, compute units, and decision layers.
Most automakers use a blend of sensors: cameras, radar, and increasingly lidar. Lidar uses lasers to map the environment in 3D and can function in low light or poor weather.
Tesla, however, uses a vision-only approach. Musk argues that human drivers use only vision, so cars should too.
Critics say that’s a flawed comparison. Humans compensate for lack of visibility using context, experience, and intent. Machines don’t have that.
This controversy was re-ignited after a Tesla Model Y using Full Self-Driving (FSD) fatally crashed into a pedestrian in Arizona in late 2023.
The car failed to react to stopped vehicles and sun glare. Footage obtained by the Tesla vehicle was used by Bloomberg to demonstrate the incident.
Tesla claimed the cameras were not blinded, thanks to a photon-counting breakthrough, but the footage told a different story. The vehicle did not brake until it was too late.
The National Highway Traffic Safety Administration opened a defect investigation after the crash.
Tesla has recalled more than 2 million vehicles to update software and mitigate driver misuse.
The agency is now reviewing whether that fix was enough.
In China, regulators are moving in the opposite direction. Following a fatal crash by a Xiaomi SU7 in March, Beijing has decided to “hit the brakes.”
The Ministry of Industry and Information Technology announced plans for new safety rules.
There’s growing pressure to create a unified legal framework that addresses insurance, liability, and the problem of over-the-air software updates that alter how vehicles behave after they leave the factory.
Is the business even worth it?
The robotaxi model faces a brutal cost challenge. Waymo’s sensor suite on its Jaguar I-Pace costs around $9,300 per vehicle.
Tesla’s vision-only stack costs about $400. That’s a 23x difference. Tesla’s approach is cheaper and easier to scale, but critics argue it’s less safe.
Robotaxi companies must also contend with vandalism, city regulations, and operational costs.
General Motors’ Cruise venture was paused indefinitely after one of its cars dragged a pedestrian in San Francisco.
The trucking sector may be more viable in the near term. Comapnies like Aurora and Torc are testing hub-to-hub automation on highways, where conditions are more predictable.
McKinsey projects that 13% of US trucks will be autonomous by 2035.
Europe is also looking in the same direction. PWC projects that by 2030, about 30% of new trucks in Europe could be autonomous on fixed routes. It’s a simpler problem than city streets.
Insurance is another unresolved issue. When a car is partly controlled by software and partly by a human, who’s responsible in a crash?
China is trying to develop a national system, while the US leaves it to states and insurers.
Analysts believe it could take five to ten years before pricing models are fully established.
So will full self-driving ever happen?
The answer is probably yes; but not in the way people once imagined.
The dream of a car that picks you up anywhere, drives anywhere, and never needs your help is not going to arrive in the next five years.
Maybe not even in the next ten.
We are not heading toward full autonomy. We are headed toward highly capable, tightly constrained autonomy.
Robotaxis in controlled zones. Trucks on predictable routes.
Cars that can drive themselves on highways but hand control back in cities. That’s not nothing. But it’s not what was promised.
Ambitious projections and promises should be taken with a grain of salt. It was 2015 when most companies were ecstatic about FSD tech.
The AV industry over-promised and under-delivered for the past decade.
The hype pulled in capital, headlines, and billions in R&D.
It also fed unrealistic public expectations. Today, engineers are solving hard problems. Regulators are waking up. And companies are adjusting to reality.
Full self-driving won’t be a switch flipped overnight. It will be a slow rollout of use-case-specific systems with tight rules, limited conditions, and lots of humans still involved.
That’s where we are now, and that’s where we’re likely to stay.
The post Is full self-driving real or just hype? Promise, progress, and what’s on the road appeared first on Invezz
