We take a look at the various types and degrees of self-driving technology. We’ll look at what the future holds, and when we’ll get there.
For decades, futurists have been dreaming of robot chauffeurs to take the drudgery out of driving. Thanks to advances in modern AI, this is a prospect that’s nearer than many of us might have suspected. But are we on the cusp of a revolution in transport like the one we’re already seeing with Battery Electric Vehicles, or is this just another gimmicky money-pit for Silicon Valley executives?
What is a self-driving car?
It might seem obvious that a self-driving car is a car that drives itself. But there are actually several distinct increments that lead up to this point. On British roads, there are already cars that nudge themselves back into their lane when it looks like they’re veering outside of it.
There are others that perform an emergency stop automatically when a collision is imminent. In addition, there are vehicles that maintain a fixed, safe distance to the car in front.
How do self-driving cars work?
Self-driving cars do their work via a variety of different methods, but they’re all powered by artificial intelligence. We might think of them not as one single technology, but as a combination of scores of different technologies.
Many of the technologies that might ultimately make the car fully driverless have yet to be perfected — but even now we’re seeing improvements in safety as a result of machine intervention.
Levels of self-driving cars
In 2014, the Society of Automotive Engineers published their classification for current and future self-driving cars, in the snazzily titled ‘J3016’ Levels of Driving Automation. These levels are based on the amount of input required from a human driver. There are six of them in total (or arguably five, if you don’t count the first one).
Level 0: No Automation
If you’re driving a classic car (or anything that’s more than a few years old) then you don’t have any automation. This is a level 0 vehicle.
Level 1: Hands-on
This is where the human driver and the machine share control over the vehicle. Vehicles incorporating the aforementioned adaptive cruise control and lane-assist would fall into this category.
Level 2: Hands-off
A car in this category would be able to drive itself — but only under the supervision of a human driver. You’d have to be prepared to sit in the driver’s seat and take over when it looks like things are going wrong. Interestingly, the car might also monitor your face, and specifically your eyes, to make sure that you’re actually paying attention.
Level 3: Eyes-off
This is the level at which you’re allowed to take out your phone and respond to some emails, or to open up Netflix and catch up on the latest episode of Cobra Kai. You’ll get an announcement from the vehicle when it’s your turn to drive.
This level of self-driving car might be just the thing for traffic jams, as they pose little danger to life and limb and therefore can be handled by a machine without risk.
Level 4: Mind-off
This is the point at which you’re allowed to go to sleep. If the vehicle requires your attention, but can’t get it, then it will be able to safely pull over to the side of the road. This would be allowed only within pre-agreed areas.
Level 5: Fully Automated
Vehicles of level 5 don’t need a steering wheel, because they don’t need any input from a human driver. These are a long way away — but don’t be surprised if you see them in the next decade or two.
Self-driving car technology
This revolution in motoring is, when you look under the hood, actually composed of several different technological revolutions. Of these, one, in particular, stands out as especially significant.
AI self-driving cars
Self-driving cars are bristling with sensors. These include cameras, GPS, and Lidar (that’s ‘light detection and ranging’, a technology that allows the creation of detailed 3D maps of an environment). Data collected by these sensors can then be subsequently fed into an intelligent program able to interpret the data and act upon it.
These programs are a product of machine learning. They haven’t been programmed, as such — researchers have simply just provided a digital environment in which machines can make decisions, and selected the best machines in a kind of Darwinian process. You can see this process at work in the game ‘Trackmania’ here.
While the real world is more complicated than a video game, the principles remain the same. In fact, much of the machine learning takes place in virtual worlds — which allows millions upon millions of iterations of the same program to be refined before being let loose in a real-world vehicle.
Once automation becomes commonplace on real-world roads, developers will have a much richer reservoir of data available to inform new generations of the program. This means that improvements to the algorithm will happen more quickly, resulting in superior performance.
Benefits of self-driving cars
Of course, there wouldn’t be anywhere near this level of excitement if self-driving cars didn’t promise a number of significant advantages. Let’s run through a few of the major ones.
Self-driving cars remove the drudgery of driving the same route every day. While you’re being chauffeured around by a robot, you’re free to sit back, catch up on some reading, get on with some work, or, if you’re on a very long journey, get some sleep.
They have the potential to be vastly safer than ordinary cars. Machines of this kind aren’t limited by the sensory organs adapted by our tree-dwelling ancestors. Sensors can be positioned everywhere on the car, and there are no blind spots.
Moreover, self-driving cars can be networked in such a way that they’re constantly and instantaneously communicating information about where they are. Thus, collisions between self-driving cars could be rendered virtually impossible.
This is consequential when it comes to things like junctions. Driverless vehicles don’t need to worry about things like give-way lines, roundabouts or traffic lights — they can communicate and work all of this out on the fly, adjusting their speeds so that collisions become impossible.
Driverless cars can’t get tired or drunk or intoxicated in any other manner. They can make better decisions, faster. And there’s no such thing as ‘thinking time’ for an automated car when it’s time to perform an emergency stop.
New models for driving
There will also be practical knock-on consequences. When you own a car that can drive itself, there’s little sense in leaving it parked on your drive. It could be out in the town, earning you money by acting as a taxi.
When you want to go for a night out, you could simply arrange for your car to be there to pick you up at a certain time or place. And if it can’t make it, you could just hail a driverless cab. What difference does it make?
We might end up with an entirely different model of ownership that is more like a long-term subscription to a taxi service. With no human driver to pay, things would surely work out a great deal cheaper.
We might even find that the sheer number of cars on the road is vastly reduced — because only a minority of motorists need to be behind the wheel at any given time.
That means less demand overall, less burden on the environment, and better transport for everyone. What’s not to like?
Disadvantages of self-driving cars
Before we get too excited, there are a few downsides to driverless cars worth considering, too. In many cases, these are worth considering long before the technology itself arrives.
Technological advancement is always going to leave some people behind. And in many cases, resisting the change is a fool’s errand. Revolutions of this kind tend to create just as many jobs as they destroy. People who are made redundant by a machine can simply re-train as engineers and programmers.
Of course, this is all very well and good when the pace of change is gradual. But what if the driverless car becomes an overnight sensation? In the UK, there were around 268,000 HGV drivers between July 2020 and July 2021. During the same period, there were around 350,000 licensed taxi and private hire drivers on the road.
At a certain point, it will become inevitable that driverless trucks and delivery drones are going to eat into the labour market. Potentially, this transition might happen very quickly.
Some will find themselves unable to retrain, and they might have to seek the support of family or the state. The economic, cultural and social consequences are worth preparing for.
You may have no choice
When we get to the point where everyone on the road is driving a perfectly safe car that’s networked with everyone else’s car, owning a traditional machine is going to become very impracticable. A human-operated car might be the most dangerous one on the road — and things like insurance might become very complicated and prohibitive.
For many of us, driving is fun. But recreational driving might find itself increasingly restricted to days at the track, and enthusiast motoring clubs might eventually find themselves a dying breed.
The technology is unknown
Many of the algorithms used by self-driving cars are going to be proprietary (owned by somebody who is likely to charge for their use). But the uncomfortable truth at the heart of a lot of modern AI is that even the companies who own the algorithms don’t know exactly how they work.
They aren’t invented yet
This is a fairly major one: a technology that’s not yet developed isn’t one that we can easily take advantage of. There’s no predictable timeline when it comes to artificial intelligence. A good rule of thumb, however, is that improvements tend to arrive very suddenly for the general public.
You might occasionally become disgruntled when Amazon’s Alexa doesn’t understand what you’re saying – but the fact that it can understand you even some of the time is borderline miraculous, by the standards of just a few years ago.
Ethics of self-driving cars
The advent of self-driving cars poses a few ethical challenges. Some of these relate to the cars themselves, and the AIs that drive them. But there are also more diffuse consequences, like mass human unemployment, which should be considered.
There’s also the possibility that driverless cars might limit human agency — which basically means that drivers might start to lean and rely too heavily on the crutch they’re given.
When we drive a car, much as we might not think it, we’re making ethical decisions. When a pedestrian walks into the road ahead of us, we might slam on the brakes. By extension, designing an automated driver also involves ethical considerations.
Obviously, it’s not ethical to design a car that drives recklessly. But problems arise when we come to work out what the rules that govern driverless car behaviour should be.
A study of machine ethics published in Nature in 2018 revealed that our moral intuitions when it comes to driving vary tremendously depending on where in the world we are. In some countries, for example, society is more likely to assign a greater moral weight to decisions that save passengers rather than pedestrians.
So, programming cars with a sense of right and wrong might be even more complicated than it would first appear. It’s almost inevitable that at some point a self-driving car will make an egregious decision that will generate headlines, and potentially cast the entire industry into disrepute.
Of course, when we’re weighing the value of a self-driving car, we’re not comparing it against an idealised machine that always does the thing that would please everyone — we’re instead comparing it against what a fallible human driver might have done in the same situation.
According to driving safety charity Brake, a person is killed or seriously injured every twenty-two minutes on UK roads — which is a fairly low bar for even an imperfect machine to clear.
We should also remember that ‘trolley problem’ scenarios where the driver has to choose between two lives are incredibly rare — in the vast majority of cases, humans know how they should be driving, and simply fail to meet the standard. In this respect, ethics are less important than tech.
When will self-driving cars be available?
Given the current state of play when it comes to driverless cars, there’s reason to be excited. So how far away is this technology?
Do self-driving cars exist today?
The cars are fully driverless and bounded by a 50-square-mile area. If you’re in the area, you can hail a car and be taken on a ride by a machine. The reception by the locals has been somewhat lukewarm.
The purpose of rollouts like this is largely to collect training data that can be used to create subsequent generations of driving AIs — but the fact that so few of the residents of Chandler have got into the cars suggests that there might be cultural factors preventing widespread uptake.
What about the self-driving cars Tesla sells?
While Tesla might call their driving assist tech ‘Full Self-Driving’, this is something of a misnomer. It’s one that’s pretty consequential, too — one driver has been warned multiple times for being in the backseat of his car while it was in motion.
‘Full Self Driving’ is actually a version of the second J3016 level. While it will drive itself, you’ll need to supervise. It tends to get fooled by roadworks and other temporary obstacles – as you can see when this Tesla driver is forced to make an emergency stop 2 minutes into a test drive.
While there are a number of technical and cultural obstacles to the self-driving car becoming mainstream, it’s largely a question of when rather than if those obstacles will be overcome.
It’s a technology that promises, in the long run, to solve many of the problems we associate with road travel: it’s cleaner, safer, and more affordable — and you can use it to get back from the pub.
Unlike other proposed transport technologies like the Hyperloop, it uses the existing road infrastructure to do it. But while the level 5 self-driving car might not be with us for a decade or so, we’ll be able to enjoy incremental progress in the meantime, rather than waiting for a sudden revolution.