How Self-Driving Cars Work: The Nuts and Bolts Behind Google’s Autonomous Car Program
Being able to commute back and forth to work while sleeping, eating, playing Trivia Crack or catching up on your favorite blogs in Feedly is a concept that is equally appealing and seemingly far-off and too futuristic to actually happen.
When Google announced their autonomous car project in 2008, visions of Minority Report began to swirl in our heads while we wondered about the possibilities of a car that really had no need for us to do anything other than turn it on. This same car wouldn’t have to worry about accidents, distraction, or driving under the influence while it made thousands – or even millions – of split-second calculations in order to keep your safe.
You see, as it turns out, humans are remarkably bad at driving. According to Joshua Schank, of the Eno Center for Transportation:
“People are not great at driving — 30,000 people die in car accidents each year (in the United States). Machines can be much better than humans when it comes to driving; they don’t drink or text and can think faster.”
While not a single driver would argue that there aren’t some real mind-numbingly bad drivers on the road, just how safe is the Google Car ?
Well, in a total of 700,000 road miles logged, the Google Car has been in exactly two accidents, of which neither could be blamed on the car. The first, happened when a human driver rear-ended the autonomous car, and the second occurred while a human was driving the Google Car on a test run. In contrast, there is one fatality for every 1.7 million miles driven in the UK and one death per 1.13 million miles in the United States (2009 data). While Google Cars have yet to log as many miles, the rate of accidents per mile driven is decidedly lower than cars operated by humans.
The Technology Behind Google’s Autonomous Car
The promising thing about Google’s driverless cars is that most of the technology is currently being used both on the road , and in other applications. This means that the technology that keeps passengers safe isn’t new, or untested, and outside of their proprietary software and algorithms, Google cars feature a lot of tested – and safe – hardware.
Let’s take a look at some of it.
Laser Illuminating Detection and Ranging – or LIDAR – is used to build a 3D map and allow the car to “see” potential hazards by bouncing a laser beam off of surfaces surrounding the car in order to accurately determine the distance and the profile of that object. The Google Car uses a Velodyne 64-beam laser in order to give the on-board processor a 360-degree view by mounting the LIDAR unit to the top of the car (for unobstructed viewing) and allowing it to rotate on a custom-built base.
While LIDAR is great for accurately mapping surroundings, its one fatal flaw is in its ability to accurately monitor speed of surrounding vehicles in real time. This is where the four bumper-mounted radar units pick up the slack. With two sensors in the front bumper, and two in the rear, the radar units allow the car to avoid impact by sending a signal to the on-board processor to apply the brakes, or move out of the way when applicable. This technology works in conjunction with other features on the car such as inertial measurement units, gyroscopes, and a wheel encoder in order to send accurate signals to the processing unit (the brain) of the vehicle in order to better make decisions on how to avoid potential accidents.
The actual camera technology and setup on each driverless car varies, but one prototype uses cameras mounted to the exterior with slight separation in order to give an overlapping view of the car’s surroundings. This technology is not unlike the human eye which provides overlapping images to the brain before determining things like depth of field, peripheral movement, and dimensionality of objects.
Each camera has a 50-degree field of view and is accurate to about 30 meters. The cameras themselves are quite useful, but much like everything else in the car they are redundant technology that would allow the car to work even if they were to malfunction.
Again, each prototype car built by Google is slightly different, but some of those tested have featured advanced sonar technology. The limitations of sonar are its narrow field of view and its relatively short effective range (about 6 meters). However, the inclusion provides yet another redundant system that allows the car to effectively cross-reference data from other systems in real time to apply the brakes, pre-tension seat belts for impact, or swerve to avoid obstacles.
A car with no steering wheel, no brakes and no accelerator would be essentially useless without advanced positioning systems to track its course and plot an appropriate route to its destination. For this challenge, Google uses its own map system , as well as GPS satellites, inertial measurement units, and a wheel encoder to determine actual speed. The system works alongside the on-board cameras to process real-world information as well as GPS data, and driving speed to accurately determine the precise position of each vehicle, down to a few centimeters all while making smart corrections for things like traffic, road construction, and accidents.
The software processes all of the data in real-time as well as modeling behavioral dynamics of other drivers , pedestrians, and objects around you. While some data is hard-coded into the car, such as stopping at red lights, other responses are learned based on previous driving experiences. Every mile driven on each car is logged, and this data is processed in an attempt to find solutions to every applicable situation.
The learning algorithm processes the data of not just the car you’re riding in, but that of others in order to find an appropriate response to each possible problem. Behavioral dynamics are also mapped and this data is used to help recognize situations before they happen, much like a human driver. For example, the cars are smart enough to recognize – and adapt to – situations such as:
- A slow-moving vehicle in the right line suggests a higher probability that the car following it will attempt to pass.
- A pot hole or foreign item in the street shows a higher probability of a driver swerving to avoid it.
- Congestion in the left lane means that drivers are more likely to attempt to enter the right lane.
Major Hurdles the Project Faces Before Widespread Adoption
While the technology in Google’s autonomous car program is nothing short of amazing, that doesn’t mean that there aren’t significant problems that still need to be addressed before we arrive at a future dominated by self-driving cars .
Before addressing any other problems the autonomous car faces, it’s important to note that technology is still the biggest barrier separating driverless cars from the consumer market.
One of these problems Google faces is the adaptability of the mapping system. The maps used by these cars aren’t like those you see in your GPS unit or on Google Maps. Each map is highly detailed down to the height of the curbs, and the dimensions of the lane the car is currently traveling in.
The problem with this level of detail is the enormity of mapping the entire country – or the world. Currently, Google has mapped approximately 2,000 miles of road for the driverless car to operate on. To give you an idea of scale, there are more than 170,000 miles of road in California alone, and over 4-million miles of public road in the United States.
The reason the cars have performed so well in their initial 700,000 mile test is largely due to the fact that the cars get to “cheat” in the way in which they respond to their environment. That is to say, each car isn’t making decisions in real-time on how to respond to external stimuli, and Google hasn’t tested the car’s ability to respond to situations outside of these mapped environments. Of course, this is a problem that could – at some point – correct itself to an extent, as each Google Car on the road isn’t just driving, it’s also helping to create 3D maps for other autonomous cars by charting data.
Additional technology-related problems are:
- So far, the car has issues that would prevent it from driving in snow, ice or heavy rain.
- It’s unable to tell the color of traffic lights when sensors are blinded by sun or glare.
- Sensors detect objects as pixelated shapes, so hypothetically, the car would respond the same way – by swerving – to miss a child in the road, or a newspaper that was floating past.
By the letter of the law, at least in the home state of Google HQ (California), self-driving cars are currently legal. Chris Urmson, head of the self-driving car project at Google, says:
“The law that was passed almost a year and a half ago made it quite clear that effectively driverless operation of vehicles was permitted in California and in general we believe that’s true across much of the US.”
That said, I’m not sure many governments imagined a future with driverless cars when enacting their current traffic laws and regulations. It’s improbable to assume that the government won’t step in at some point and do their own investigation as the the practicality and safety of autonomous vehicles .
Do Consumers Even Want It?
While there has been wide-spread excitement about the technology from some, others have dismissed the idea entirely due to safety concerns, or the unwillingness to give up control of the wheel. While some may never embrace the technology , it’s safe to say that some segment of the consumer population would be interested assuming two main factors are addressed: safety and pricing.
While the safety concerns are still being worked out, the car does appear to have a remarkable safety record to this point. One philosophical quandary has been brought up that’s quite interesting though; how will the car respond to so-called “trolley problems?” Trolley problems are the philosophical term used to describe issues that arise where there is no correct answer. For example, a Google car has to swerve to avoid a child, but in doing so it would impact the mother, or a group of children on the other side of the road; what does it do?
So What’s Next?
Overall, it’s an exciting idea and the adoption of the technology has other practical applications. For example, a car that doesn’t need a driver, also doesn’t need a passenger. This would allow your car to drop off packages without you, give friends a ride home (and then return to your garage), or find its own parking spot after dropping you off somewhere.
In fact, the most exciting idea of all might be the fact that this could make the notion of “owning” a car obsolete. These cars could hypothetically be left anywhere, and serve as a sort of taxi for entire communities without ever having to purchase and maintain the car yourself.
Driverless cars are an exciting idea for the future of transportation, but until Google address the technological challenges and prove that these cars are indeed safer than your average driver, then they’ll remain just that, an idea. I, however, am confident that Google will work out the kinks, and that this technology will have a huge impact on the future of transportation .
Would you buy a self-driving car?