How far off are self-driving cars?

Five years ago I worked with Auto Trader to produce a report on the car of the future. It  was fairly successful, getting picked up by a lot of the national press and seeing me pop up as spokesperson for the report on BBC News.

While at the FT Future of the Car conference last week (from where you will find lots of quotes in this piece), I took a look back at some of the predictions we made. We got a lot right. I’d stand by our broad characterisations of the 2020s as the last decade of the combustion engine, the 2030s being a period of almost full electrification, and the 2040s being about autonomy.

Some people might challenge that last piece. Surely autonomous cars are here already? Well, that all depends what you mean by a self-driving car

Here Comes The Science Bit

Autonomous Vehicles or AVs are classified according to a scale. A scale that is slowly expanding. But to keep it simple, let’s stick to the six original levels.

• Level 0: Your traditional car with no assistance

• Level 1: A single automated system, such as cruise control

• Level 2: Partial automation, also known as ADAS (Advanced Driver Assistance Systems). E.g. cars with lane assist, braking assist and adaptive cruise control.

• Level 3: Conditional Automation. The vehicle can pilot itself in some conditions (weather, light, traffic, location) but not all. And the driver must be ready to intervene if needed.

• Level 4: High Automation. The car pilots itself in most situations and can detect and intervene even when things go wrong. The driver becomes a passenger but can override if needed.

• Level 5: Full Automation. No human interaction required in any circumstances.

So where are we today?

Many new cars today are at Level 2 or somewhere between there and Level 3. The big difference between Level 2 and Level 3 (and perhaps the reason we have weird classifications like “Level 2++) is - according to Eric Bach of Lucid - at Level 3 the liability transfers from the driver to the manufacturer.

Tesla cheerleaders take note: its products pitched as “Full Self Driving” and “AutoPilot” are Level 2 systems. If something goes wrong, you are liable. As Siddartha Khastgir, head of safe autonomy at the University of Warwick points out, “Tesla does not offer autonomous driving. It offers assisted driving.” He also highlights that misrepresentation of autonomous capabilities is a crime under the UK Automated Vehicles Act 2024. Make of that what you will.

The closest we have to full autonomy right now is so called RoboTaxi services operating in cities around the world, such as San Francisco and Phoenix in the US, and Wuhan in China, with the likes of Waymo and WeRide. These are Level 4 vehicles. They can operate fully autonomously, but only in a limited geographical area and not in all conditions - estimated by Uber’s Andrew McDonald as 98-99% of the time.

If you count this as full autonomy, then yes, we have self-driving cars today. I don’t. And I don’t think most consumers would either.

How would you define an autonomous vehicle?

Ask people what they think a self-driving car should do, and they’ll give you a pretty universal answer. They want to get in, tell the car where they want to go, and for it to take them there. Whether they’re drunk or sober, fully awake or half asleep, staring out the window or watching a film. They expect it to do that in the conditions that they can drive in now: day or night, rain or shine. Of course, humans can’t drive safely in all conditions (even when they think they can). But it’s more than 98/99% of the time.

Level 4 autonomy, limited by location, only available as a taxi, and not functioning in all weathers, does not fulfil that popular definition.

Especially given one rather unexpected revelation from Uber’s McDonald:

“Running autonomous vehicles, you learn all the little ways that your Uber driver breaks the law every day to improve the customer experience.”

In other words, you have to do more walking to meet a RoboTaxi because it won’t pull up on double yellows, reverse down a one way street, or hold up traffic to pick you up or drop you off as close as possible to your destination.

Level 5 is an important milestone because it fulfils the popular definition of a self-driving car. Though it still won’t break the law for you.

When do we get to Level 5?

According to Maria Alonso from the World Economic Forum (WEF), their latest forecasts say that the “vast majority” of personal cars will be Level 2 or Level 2+ (a distinction I won’t cover in this article) by 2035. 10% will be Level 3 and 4% will be Level 4. 

Alonso says that technology providers have told her that there are no more engineering barriers to Level 5. I’m not 100% convinced about that, based on conversations I’ve had with some experts. But even if it’s true, a lot of other barriers remain.

One of them is cost. Adding the sensing and processing capabilities to a car costs money, though that is coming down. While a fully equipped AV platform for a RoboTaxi application might have cost $150,000, Baidu in China recently announced a $37,000 car with all the required functions.

Another is regulation. There isn’t consistent regulation - if there is regulation at all - internationally. And it might not even be at a national level. Cities, regions, might have their own regulations. 

Then there’s insurance. Once we get to Level 5, the insurance industry has to be willing to underwrite insurance for every manufacturer. In fact, they have to be willing to do it from Level 3 up, if the car has an accident while in an autonomous mode.

There is consumer acceptance. Do people trust a car to drive them around?

And for RoboTaxis, there’s the business model. Who is going to underwrite the cost, and accept the risk, of operating a lot more autonomous vehicles? Uber’s McDonald estimated the cost of covering a city like London with sufficient vehicles to run a service to be in the billions.

So, when?

We will overcome all of these barriers. We will have RoboTaxi services that cover more and more cities until they start to merge and take on motorways to get you between them. The hardware will get cheaper and the software more sophisticated until that 98-99% becomes 99.99%. And if you still want to own your own vehicle when there are ubiquitous autonomous mobility services, you will be able to own one that needs no other input than a voice command.

But while there will be small examples and pockets of more advanced progress, I don’t think either of those things happen at universal scale until the 2040s. And that belief is only reinforced by those WEF numbers.

So five years on from the Auto Trader Car of the Future report, I still think it is the 2040s that will be the decade of self driving.