Imagine for a moment that the motor car had never existed. I know this requires some imagination because almost everything about our modern world has been constructed around it, but humour me, indulge me if you will. Just keep reading.

It is 2026 and I have invented something. Naturally, because I’m a working-class engineer and therefore apparently incapable of simply enjoying an invention without building prototypes and turning it into a development programme, I decide that I must put my baby into production and I need investors. After a year of crowd funding and disappointing lottery tickets, I miraculously get selected for Dragons Den and my pitch goes something like this:

“Well,” I say, “it’s a machine that members of the public sit inside with their children.”

Peter says “Fine”.

“It weighs about two tonnes.”

Deborah says “OK”.

“And it can travel at 70 mph.”

Touker writes something.

“Actually considerably faster than that, but we’ll tell people not to.”

Evan, behind the screen, sucks air through his teeth, and Theo just stares at me.

They all want to know who is going to operate these machines, and I explain that practically anybody can, providing they reach the age of seventeen, (fifteen if there is somebody else in the car with them who knows how to drive), pass a relatively straightforward practical and theoretical examination, and haven’t done anything sufficiently stupid beforehand to convince the authorities that giving them control of two tonnes of rapidly moving metal would be a particularly bad idea.

After that, we’ll generally leave them alone for the next sixty years. There will be no regular competency assessment. No recurrent training. No simulator sessions. No annual demonstration that they still understand what all the road signs mean.

They may drive when tired, angry, distracted, hungry, late for work, arguing with their spouse, screaming at the children in the back seat or wondering why they’ve just passed their destination despite Google maps saying its 10 minutes away, on their mobile phone, on their lap, beside the Coke can held between their legs.

At which point Peter quietly asks, “How do you intend to stop them crashing into each other and killing everybody?”

I smile and say, “This is the really clever bit.”

“Paint.”

“We are going to paint a white line down the middle of the road.”

“I’m out!”

Just think about that for a minute. Two cars, each weighing roughly two tonnes, approach one another at 60 mph. The closing speed is 120 mph. There is no barrier between them. There is no rail guiding them. There is no automated system physically preventing either vehicle from crossing into the path of the other, in fact in many cases, absolutely nothing between four tonnes of rapidly approaching metal except several inches of white paint and an unwritten agreement between two complete strangers that neither of them is going to do anything stupid for the next three seconds. They might think about it, but hopefully they don’t act on those all too human impulses.

Then they pass. And another one comes. And another. And another.

We do this thousands of times a day.

I don’t know what the modern HSE would make of the original risk assessment for the car, assuming anybody had considered writing one, but I have a sneaking suspicion, given my 30 years subjected to the Functional Safety Inquisition that “mitigation: paint a line on the floor” might have generated a few comments. And yet, astonishingly, it works. Which brings me to something that I think we rather unfairly overlook.

Human Beings Are Good at Driving

We spend an enormous amount of time talking about how bad people are at driving. And people are bad at driving. Some people are spectacularly bad at driving. Spend half an hour on any busy road, and you will encounter somebody doing something so inexplicably stupid that you find yourself staring at them through the windscreen wondering whether they won their driving licence in a raffle, whilst imagining you’re in an X-Wing as you line up a reticule shaped mark on the windscreen with the idiot’s car ahead, press your thumbs deep into the steering wheel and blow them off the road; or maybe that’s just me…

But that’s not really the interesting part.

The interesting part is that, statistically speaking, practically everybody you passed on the way to work this morning successfully managed not to hit you. Nor did you hit them. Think about what your and their brains did during that journey.

A car approaches a junction on your left and you look at it. You don’t calculate its velocity. You don’t calculate its deceleration. You don’t establish its trajectory relative to your own and calculate a time-to-collision. You just look at it and think: They’re stopping. Except sometimes you look at exactly the same situation and think: They’re not stopping.

Why?

Something was different. Perhaps the car was travelling slightly faster. Perhaps its deceleration profile was different. Perhaps the driver wasn’t looking towards you. Perhaps its wheels were turned slightly. Perhaps it had crept six inches further over the give-way line than cars normally do. Whatever it was, your brain noticed it, compared it with your life’s worth of accumulated experience and quietly suggested that taking your foot off the accelerator might be a good idea. And you probably weren’t consciously aware that any of this happened.

Then there’s the cyclist ahead. They’re travelling in a straight line; except they’ve moved six inches towards the centre of the road and there’s a pothole coming up. You don’t necessarily consciously think: Cyclist has identified pothole and is preparing lateral manoeuvre. You just give them some room.

The pedestrian approaching the crossing looks as though they’re going to cross. The child beside the road worries you more than the adult standing next to them. The van ahead is wandering slightly within its lane and something tells you the driver may have dropped their phone. The car behind is too close. The car in front has braked. The car three cars ahead have braked, which is even more interesting as you watch to see whether the other two cars have noticed, and whether you can be the last one to brake, in a mental game of chicken with only you playing. The traffic lights have been green for quite a while, so although they’re still green, you’re already anticipating that they may change. And while doing all of this you’re steering, maintaining speed, checking mirrors, reading signs, listening to the radio and to the engine and what you think may be the driver side front  wheel bearings, which in turn is getting more difficult to make out over the sound of what you think is the hole in the flexi joint in your exhaust, finding your turning, avoiding potholes and possibly calmly explaining to your partner in the passenger seat that you are not tired and you know where you are going.

And then we cheat. Us human types have another trick which makes autonomous driving considerably harder than it looks. We don’t always follow the rules.

Consider a narrow road with parked cars on your side. According to the code, the obstruction is on your side, therefore you should give way to the approaching vehicle. So you stop. The approaching driver stops. You look at them. They look at you. They flash their headlights. You go. That’s it. Deal done. Marvelous. Two complete strangers have just negotiated a temporary modification to the normal operating rules of the road in perhaps two seconds without exchanging a single word.

But, even with everything I highlighted in the last episode, try writing that requirement at the system level.

The autonomous vehicle shall… Well. Shall what?

• Proceed when another vehicle flashes its headlights? Definitely not, because people flash their headlights for all sorts of reasons.
• Proceed when another driver has stopped? Not necessarily.
• Proceed when another driver has stopped, there is sufficient space, the vehicle appears to be yielding, the route is clear and the autonomous system has sufficient confidence that the other driver’s intentions…

The System Engineers out there can already see where this is going. Somewhere in a conference room there are forty-seven PowerPoint slides, twelve engineers, a couple of bored managers, sixteen laptops (somebody always brings 2 laptops), and somebody from Functional Safety wearing a T-shirt saying “Everything Worked, Everybody Died, Who You Gonna Call? S.O.T.I.F.” happily ruining everybody’s afternoon by asking what happens when the other driver changes their mind.

Welcome to autonomous driving.

I’m digressing somewhat here given that this episode was primarily about why today’s cars need to beep, but I’m going to run with it for a second or two.

The autonomous car doesn’t just have to learn to drive, it has to learn us, and that isn’t easy. It has to distinguish between the pedestrian standing beside the road who isn’t going to cross and the apparently identical pedestrian who is. It needs to recognise that the cyclist ahead is about to move around a pothole. It needs to understand that the car approaching the junction is technically required to stop but apparently has no intention of doing so.

It needs to recognise roadworks, faded markings, missing markings, white markings that look black in the sun, black tar that looks like white in the sun, temporary signs, mud, snow, rain, glare, shadows, emergency vehicles, animals, debris and the infinite variety of stupid things that people manage to do with their cars, and it has to do this sufficiently well that we trust it with our kids.

This is the bit I think gets lost whenever somebody announces that mass production autonomous vehicles are just around the corner because somebody’s latest prototype has driven another million miles. Driving a million miles isn’t necessarily the difficult bit, finding the one situation in those million miles where something happened that nobody had anticipated, understanding why your system made the decision it made, fixing it without breaking something else and then demonstrating with sufficient confidence that your fix works in every relevant variation of that scenario, is where things start becoming interesting. Like I said in my first episode, recording millions of miles of data is only useful if somebody has the time and resources to understand all the occasions when it went wrong.

Six years later I haven’t changed my mind. In fact, I haven’t changed my mind in the 11 years since I presented this at an automotive conference in Germany. Autonomous cars need us out of the loop and off the roads.

We are trying to train cars to interpret an environment whose visual language was never designed for machines.

Imagine a world where there are only autonomous cars on the road, things get much easier. What we recognise in today’s cars and roads is, to a surprising extent, a graphical user interface. Nearly everything we currently bolt, paint, illuminate or print onto the road, and everything you see inside the car, is effectively a GUI.

In that future world we would have gotten rid of (apologies for the long list, but I’m making a point here – skim it when it gets boring):

• Painted lane lines – no need to visually tell a computer where its lane is if the permissible trajectories are digitally defined.
• Centre lines – the cars already know the road geometry and where opposing traffic will travel.
• Lane arrows – straight ahead, left-turn-only, right-turn-only, etc.
• Hatched areas and chevrons – largely instructions to us humans about where not to drive.
• Give-way lines.
• Stop lines.
• Painted STOP/GIVE WAY markings.
• Box junctions.
• Bus-lane markings.
• Painted speed limits.
• Cat’s eyes/reflective road studs as visual lane guidance.
• Most roadside delineator posts and reflectors.
• Speed-limit signs.
• Repeater speed signs.
• STOP signs.
• GIVE WAY signs.
• One-way signs.
• No-entry signs.
• No-right-turn/no-left-turn signs.
• Mandatory-direction arrows.
• Keep-left/keep-right signs.
• Priority signs.
• Minimum-speed signs.
• Overtaking-prohibition signs.
• Clearway/no-stopping signs.
• Weight, height and width restriction signs – the routing system already knows the vehicle’s dimensions and permitted route.
• Low-bridge warnings.
• Steep-gradient warnings.
• Sharp-bend warnings.
• Junction-ahead warnings.
• Roundabout-ahead warnings.
• Crossroads warnings.
• Road-narrows warnings.
• Slippery-road warnings, at least as signs; the condition itself would instead be communicated digitally or detected by the vehicle.
• Traffic-light-ahead signs.
• Direction signs.
• Motorway destination boards.
• Exit signs.
• Route-number signs.
• Distance-to-destination signs.
• Street-name signs, as far as vehicle navigation is concerned.
• Tourist direction signs, again as far as the vehicle is concerned.
• Diversion signs – diversions become routing-data updates.
• Temporary roadworks direction signs intended for drivers.
• Variable-message signs telling drivers about congestion, accidents or journey times. The vehicles already know.
• Average-speed-camera signs.
• Speed cameras themselves, potentially. A properly governed autonomous vehicle simply doesn’t choose to break the law.
• Average-speed camera systems.
• Much conventional traffic enforcement aimed at driver behaviour – speeding, red-light running, illegal turns, lane violations become software/routing constraints rather than rules you’re hoping somebody obeys.
• Traffic lights, potentially one of the biggest changes. At an autonomous-only intersection, vehicles could negotiate crossing trajectories and timings directly rather than everybody sitting stationary waiting for a coloured lamp to give them permission.
• Many conventional roundabouts could potentially be redesigned. Their present geometry is partly a mechanism for forcing humans to negotiate priority safely.
• Pedestrian traffic lights. Pedestrians would be channelled above or below the road, never across it.
• Traffic-calming measures. Speed humps, chicanes and artificial narrowing exist largely because humans cannot be trusted to obey the law.
• Rumble strips intended to alert inattentive drivers.
• Some crash barriers used primarily to mitigate predictable human mistakes might become less necessary, although barriers protecting against mechanical failure and dangerous terrain would absolutely remain.
• Motorway countdown markers approaching exits.
• Lane-closure arrows and temporary lane-control signs. Instructions could be transmitted directly to vehicles.
• Manual traffic control at roadworks. The temporary road geometry could simply become part of the network’s routing information.
• Traffic police directing vehicles at junctions, in ordinary circumstances.
• Human-readable parking restriction signs.
• Parking bay markings, potentially. A vehicle could position itself within digitally defined parking coordinates.
• Parking meters and pay-and-display machines. Payment could happen automatically between vehicle/account and infrastructure.
• Ticket machines at toll roads.
• Toll booths.
• Most physical tolling infrastructure visible to drivers.
• Motorway gantries carrying driver information, although gantries might remain for sensors/communications equipment.
• Roadside advertisements designed specifically to catch drivers’ attention become rather pointless and arguably undesirable if nobody is looking out of the car window anymore.

Without all that out of the way, and with a sparkly new autonomous highway in place, autonomous car development is relatively straight forward. Until then, its going to be a slow burn.

However, getting back to the cars of today and the “Bloody Beeping” my mechanic was complaining so bitterly about, which apparently I am personally responsible for along with the utter lack of room to do anything in the engine bay. Modern cars beep at you. I should know, I’ve spent a lot of time developing and introducing most of the features that monitor, brake, nudge, steer and beep.

They beep because you haven’t put your seatbelt on. They beep because you’ve wandered towards a white line. They beep because they think you’re tired. They beep because you’ve exceeded the speed limit. They beep because you’re approaching something too quickly. They beep when you’re low on fuel. They beep when your tire pressure is low. Some tug the steering wheel. Some vibrate it. Some brake. Some watch your eyes to make sure you’re watching the road and not your phone and then beep.

And people complain about this constantly, even the developers. Sometimes with justification, because some of these systems can be annoying, and there are few things more irritating than having a car confidently misunderstand a situation that you understood perfectly well several seconds ago.

But the majority of people are looking at it the wrong way, in my humble opinion.

These safety systems aren’t appearing because the “bureaucrats in Europe are bored and feel like picking on drivers”, or that the regulators suddenly decided that driving had become dangerous. Driving was always dangerous. The regulators have simply caught up.

If the motor car genuinely appeared for the first time today, I struggle to believe that we would approve the road transport system in anything resembling its current form, but that doesn’t necessarily mean we’d ban the car.

We tolerate dangerous things all the time because they provide sufficient benefit. Aircraft are dangerous. Industrial machinery is dangerous. Electricity is dangerous. The sea is particularly dangerous and, having spent a substantial amount of time sailing on it, I can confirm that nobody has yet managed to put a barrier around that.

What we do is regulate the risk. The problem with cars is that we discovered how useful they were before we started thinking seriously about just how absurd the underlying proposition was, and then it was too late. We built houses around them. We built towns around them. We built industries around them. We built economies around them. You can’t simply announce on Monday morning that, having completed the risk assessment, we’ve decided cars are a bit dangerous, and everybody needs to hand their keys in by Friday. So instead, we’ve spent more than a century retrofitting safety onto the original idea. Brakes got better. Tyres got better. We added seatbelts. Then airbags. Then ABS. Then stability control. We improved crash structures, roads, barriers and junctions. We introduced drink-driving laws, speed limits, driving tests, MOTs and increasingly stringent vehicle regulations. Then electronics became sufficiently capable to start monitoring what the driver was doing.

And now the car is starting to intervene.

Lane Departure Warning (LDW). Lane Keep Assist (LKA). Emergency Lane Keep Assist (ELKA). Autonomous Emergency Braking (AEB). Blind Spot Monitoring (BSM). Intelligent Speed Assistance (ISA). Driver Attention Monitoring (DAM), the list goes on.

I’ve spent a large proportion of my career developing all of the above engineering acronyms into production, and they are all ultimately attempts to solve the same problem:

Human beings are good at driving. Right up until the moment when we’re not.

I like cars. I’ve spent a large chunk of my working life developing systems for them, and I don’t particularly want anybody to take mine away, (at least not until I can swap it for a Level 5 limo). Cars provide an extraordinary amount of personal freedom and independence. I can walk outside, get behind the wheel and drive hundreds of miles without asking anybody’s permission, checking a timetable or explaining where I’m going.

Today they are easily the most technically sophisticated item most of us will ever own and it’s that technical sophistication I want to dwell on for a bit, because a modern car is not really a single product at all, but the end result of an almost absurdly complicated engineering exercise involving thousands of people, hundreds of suppliers, dozens of engineering disciplines and several years of trying to persuade an extraordinary collection of mechanical, electrical, electronic and software systems, many of which were designed by people who have never met each other and quite possibly live on different continents and speak different languages, to behave as though they had all been conceived by one person, at the same time, for precisely the same purpose. By the time a new car reaches a showroom, perhaps five thousand people will have had some involvement in creating it: marketeers, solicitors, designers, engineers, software developers, aerodynamicists, metallurgists, manufacturing specialists, test drivers, safety engineers, suppliers and countless others responsible for components most owners will never know exist. And amongst all of that effort we somehow also expect the thing to be beautiful.

Whether you realise it or not, we have a unique, deep, even profound relationship with our cars, and perhaps all the regulation, the NCAP tests, the safety systems, the cameras, the radars, the lidars, the driver monitoring and, yes, the “bloody beeping”, are part of the price we have to pay to preserve it. Because the alternative isn’t necessarily a wonderfully simple car without all the annoying safety systems. Taken to its logical conclusion, the alternative is that at some point society decides the casualty rate associated with letting millions of relatively untrained people control two-tonne machines is no longer acceptable and actively denies many of us access to the road network.

So perhaps we should be slightly more tolerant of the engineers and regulators trying to prevent that from happening. Even when the car beeps at us. Because underneath the radar, cameras, airbags, ABS, ESC, AEB, LKA and every other acronym we’ve bolted onto the modern vehicle, the fundamental operating principle of the road network remains almost exactly as ridiculous as it was a hundred years ago.

Two strangers. Two cars. Sixty miles an hour. Heading directly towards one another.

And the primary safety system separating them is still…

Paint.

When you put it like that, perhaps the surprising thing isn’t that people occasionally crash. Perhaps the surprising thing is that we ever thought this was a good idea in the first place.

And perhaps the most extraordinary thing of all is that, thanks to an astonishing combination of human perception, prediction, cooperation, engineering and more than a century of accumulated experience, most of the time it actually works.

Coming soon:

Read these too:

How to Design an Autonomous Car – Part 7

How to Design an Autonomous Car – Part 1:  A Brief Intro

How to Design an Autonomous Car – Part 2:  Levels of Driving Automation 

How to Design an Autonomous Car – Part 3:  Getting Started

How to Design an Autonomous Car – Part 4:  Systems Engineering

How to Design an Autonomous Car – Part 5:  Atomic Requirements

The rise of Autonomous Car and the Death of the OEM

Can an Autonomous Car be safer than a Cargo Ship?

Regulation 79.02 and what it means to you

ADAS Market projected to reach $56bn by 2023

World first launch with off-the-shelf ADAS solutions