If you follow tech-related news, you’ll probably have stumbled upon the topic of “self-driving cars” (or “autonomous cars“) more than once lately. Maybe you dismissed it as you read the headline, thinking “this is just a Sci-Fi pipe dream, like flying cars or jetpacks“, or you just thought it wasn’t important enough. Or maybe you read a single article but you didn’t get enough context to know what the hell is this all about.
I’m gonna try to explain, as simply as possible, what are self-driving cars and how do they work.
Why do we need self-driving cars?
Humans are bad at driving. Even if you consider yourself a good driver (by human standards), you aren’t. Humans get distracted, they get drunk and try to drive home, their night vision is deficient, their reflexes worsen with age (and they aren’t that great even when they’re young), they get sleepy, etc. Just in 2015, 35.000 people died on car accidents in the US, and more than 2 million were injuried. That’s on one year, in one country. If it’s of any consolation, those numbers were much higher about 20 years ago, before security measures like airbags and ABS became common on most of the new cars.
Those are a lot of deaths. Using public transportation is much safer, but that’s not always a viable option (for example, if you live in the suburbs), and even if it is, it’s not as convenient as owning a car. Self-driving cars would combine the convenience of a car, with the added security of having someone (or something) much more experienced and capable than you at the wheel. Eventually, if all the cars in the roads are self-driving, accident rates could get close to zero.
That’s why we “need” self-driving cars. It’s a matter of saving lifes. But there are many other perks apart from not getting killed in a moving steel cage. Owning a self-driving car would be like having a private choffeur, 24 hours a day. I’ll ramble about these perks in my next article, but let’s just say that the future looks pretty interesting.
What is a self-driving car?
It’s a car that can operate without human intervention. Well, duh. Yeah, sorry to dissapoint you, but the basic definition is as simple as that. However, that leaves it open to a lot of interpretation and ambiguity. Under what circumstances must the car be able to operate autonomously? How much time can the car operate this way before the driver has to take over? To clarify those questions, SAE (Society of Automotive Engineers, a US-based standards organization) has defined 6 levels of autonomy. Thus, a car “autonomy” can be defined by the highest level it implements.
- Level 0: At most, the car just issues warnings (such as colision detection, or ultrasonic parking sensors), but it can’t take over navigation control in any way.
- Level 1 (“hands on“): In specific circumstances, the car can take partial control of navigation. Examples of this cars are the ones that have basic “cruise control“, where the car can keep a constant speed in the highway by itself, but the driver must control steering. Another example would be “parking assistance“, where the car handles steering during the maneuver but the driver has to control speed. Pretty much every car manufactured in the last 6-8 years has some form of “Level 1″ autonomy (at least as an optional feature).
- Level 2 (“hands off“): The car can take total control of navigation in some specific scenarios, but the driver must be prepared to retake control at any time. For example, in a highway, that would mean that the car itself must be able to decelerate to avoid colisions, automatically change lanes to overtake, and automatically steer the wheel to follow the road lanes. Some high-end cars, such as Tesla and Mercedes, have “Level 2” autonomy, usually marketed as “[big word] cruise control” or “autopilot“.
- Level 3 (“eyes off“): Like “Level 2“, but the car is prepared to react safely on an emergency, so the driver can take his eyes off the road. The car may still prompt the driver to retake control in specific situations though, and the driver must be ready to do so (so, he can’t take a nap).
- Level 4 (“mind off“): Like “Level 3“, but the driver doesn’t need to be prepared to retake control at all. If the car encounters an unexpected situation, it must be able to overcome it or safely abort the trip (park the car) if the driver doesn’t respond. That means, that in this case the driver can safely take a nap, but he still may need to drive manually to reach his destination.
- Level 5 (“wheel optional“): All the other levels focus on specific circumstances, like highways or parking. Level 5 means that there’s no human intervention required in any way, in any circumstances. The car is prepared to handle any situation that a human could (city environments, rural roads, highways, parking lots, bad weather, etc). In fact, Level 5 cars may very well be sold without a steering wheel or mirrors.
How does a self-driving car work?
I don’t want to get too technical, so I’ll keep this short. Basically, a self-driving car needs to emulate a human driver, so it needs senses (sensors), muscles (motors) and a brain (computer).
The “motors” part just means that the car needs to be able to steer the wheel and operate the pedals by itself. Considering how much electronic components a regular car already has, this is considered “the easy part”.
For the “sensors” part, a self-driving car will need a combination of the following ones, depending on the autonomy level desired.
- LiDAR (Light Detection And Ranging): This sensor emits a pulsed laser light, and has a sensor to measure the time it takes for the reflected pulse to return. The problem is, that only gives the distance of the nearest obstacle directly ahead. For cars, the LiDAR has to be mounted in a rotating platform so it can do a “sweep” and get the distance of a grid of points around the vehicle. Those rotating platforms need to be very fast and precise, so this kind of sensor is very expensive. You’ll recognize it because it’s always mounted on top of the car’s roof. As of this moment, it has only been used in prototypes.
- Radar: It emits radio waves and measures the time it takes for them to bounce back to the sensor to estimate the distance to the nearest obstacle directly ahead. This sensor is already used on most of the cars that have “adaptative cruise control” or “colision warnings“. One crucial advantage is that it isn’t affected by meteorological adversities like fog or heavy rain.
- Ultrasonic sensors: It emits an ultrasound and measures the time it takes for it to bounce back to the sensor to estimate the distance to the nearest obstacle directly ahead. It has much shorter range than Radar (around 1 meter, usually), so it’s primarly used for parking assistance.
- Cameras: Since an autonomous car needs to interpret road signs and lanes, it has to have at least one camera. Many cars (Opel Insignia comes to mind) already have some sort of camera and computer that “reads” some road signs and shows that information to the driver. But their usefulness doesn’t end there. If you think about it, humans are able to drive a car in every possible situation using just the 2 cameras we have in our faces. Artificial cameras can have some extra features too, like improved night vision (infrarred cameras) or 360º vision (just putting a camera in every corner of the car).
Even for the same level of autonomy, different companies will use a different combination of sensors, depending on the judgement of their engineers. For example, Google/Waymo relies heavily on LiDAR for their cars, while Tesla considers that it’s not needed and instead opted for having more cameras covering every possible angle, and a single Radar.
The “computer” is a special kind of hardware and software capable of processing the sensors’ data, interpreting it, and emitting commands to the motors. The human brain is very good at quickly identifying and classifying visual patterns (you’re doing it right now without even noticing). Computers are still not as good as us, but they’re getting better, thanks to advances in “Machine Learning“, “Neural Networks” and “Deep Learning“. I don’t really know anything about those particular areas, but I just need to know that self-driving cars need to have powerful computers with specialized software that allows them to recognize shapes based on the sensors’ input. Shapes such as signs, traffic lights, road lanes, other cars (with direction and speed estimations), pedestrians, etc.
To be continued…
That’s it for today, I hope I’ve cleared some doubts. If you think I’ve missed something, or you didn’t understand some of the explanations (which would probably be entirely my fault, not yours), just tell me in the comments.
In my next article, I go over which are the companies that are closer to bring self-driving cars to market.