Lawmakers in Nevada made a pretty forward-thinking move a couple weeks ago when they passed a measure ordering new regulations for driverless cars. Many vehicles already participate in once-human-driven activities like parking and skid control, and it’s not long until they’ll be able to navigate, make decisions and drive totally by themselves.
But in some ways, the world of self-governing cars is already upon us. Using relatively simple software and adjustments to existing hardware, major automakers in the U.S. and Europe are making cars work smarter and greener in a way that has nothing to do with hybrid engines or alternative fuels.
Connected to each other and to the cloud, cars will be able to make their own decisions — so the future of driving, put simply, will be largely out of human hands.
Algorithms and analytics will predict driver behavior and forecast future commutes, studying the future from a few seconds to several hours down the road. Radar-equipped sedans will sense their surroundings, and road trains and car-to-car networks will be reducing congestion, preventing fatalities and improving fuel economy.
“A lot of the interactions we have today are to help the driver do their work, but what we’re tying to do is help the car, to make the car smart,” says Ryan McGee, an engineer at Ford Research and Innovation in Dearborn, Mich. “When you take a car and connect it to the cloud, there are so many possibilities.” Here are just a few of them.
Your Car Will Predict Where You’ll Drive
If you don’t use mass transit, odds are your drives to and from work or school follow a typical routine — in the morning, you go from home to the coffee shop to the office; in the evening, you leave the office, maybe make a grocery store pit stop, and go back home. Ford would like its cars to take advantage of your predictability, and guess where you’re going when you turn the ignition.
Researchers are feeding driving history into a Google software service called the Prediction API, which uses a machine-learning algorithm to generate a model of predicted behavior — in this case, a particular driver’s habits.
“The question we ask the model is, ‘Where is this person going to go next?'” said McGee, who is working on the model. “The model would say, it’s Wednesday at 5:00, so you’re pretty likely going to go home, and it sends that data back to the car.”
The current system connects to the Internet and records where the car is at what time and on what day. The algorithm computes a list of likely trips for that place and time. Based on the trip possibilities, the car can shift its power consumption to run on a battery instead of gasoline — which will be useful for plug-in hybrid cars, McGee explained. Ford has been testing it in an Escape SUV plug-in hybrid.
In the future, governments might enforce low-emissions zones near a school or a park, or build battery-only lanes. McGee usually picks up his son from soccer practice on his way home from work, he said. If his car knows his habits, and knows there’s a low-emissions zone in the area, the car will save some battery juice for that portion of his day.
“It will make sure I save enough energy so when I get to that area, I can drive electric,” he said. “If we know more about how people are going to use the car, we can optimize their performance.”
But all routines can be broken — what if you want a burrito on Wednesday but crave a cheeseburger on Thursday? The car will have no idea where you’ll go come lunchtime. Because of randomness and drivers’ fickle nature, motorists will be able to make additions and corrections to the system if it’s ever installed in mass-market vehicles, McGee said.
Your Car Will Predict What Traffic Will Do
Once a car knows its driver’s own habits, it can incorporate other data from the cloud to make more informed suggestions. If traffic forecasters have enough data, and they have good enough models to interpret it, they should be able to tell you at lunchtime what your 5 p.m. commute home is going to look like. IBM made a smartphone app to do this.
IBM says the app can learn a driver’s patterns, predict what the traffic will look like 30 minutes or more in the future, and send the forecast to the driver before he or she gets behind the wheel. Big Blue has been testing the system in the San Francisco Bay area, according to John Day, Program Manager for IBM Smarter Traffic.
The app uses a phone’s GPS capability to track a driver’s movements during various times of day, learning his or her routines. It also connects to a system of 700 road sensors previously installed by the California Department of Transportation. The sensors track how fast cars are going and the flow of traffic, counting cars per minute. The sensors collect data every 30 seconds, but IBM aggregates that into a 5-minute set and feeds it into an algorithm. The algorithm makes correlations among the data to pull out patterns, Day said.
“The tool is good at discovering and learning the signature of slowdowns,” he said. “It will notice at a particular area or especially an interchange, when traffic slows down, then 83 percent of the time, or whatever it is, you get a much bigger problem.”
Using those signatures, the system analyzes real-time data to build a constantly updating model of the traffic situation, changing every 5 minutes. While existing prediction systems base their forecasts on current conditions, predicting what traffic will look like if nothing changes, the algorithm can recognize and account for the ripple effects of single actions.
Users can log in to the app from a computer to see “journey history,” and can add or delete routes as they choose to control how much information is stored, Day said. He even envisions the app offering coupons for businesses the driver frequents.
“If you have the capability to recognize that somebody drives by a coffee shop so often, that would be valuable information. It has to be used carefully, and managed such that the user has full control over whether they want to have that data shared, but yes, there are certainly capabilities there,” he said.
But the app’s real strength will be in helping a driver avoid traffic, he said. It can suggest alternate routes and even check web-based transit timetables. So if a driver normally leaves for work at 8:30 a.m., he might get a text message at 8 a.m. with the day’s forecast — “your normal commute looks bad today, but this train leaves at 8:30, and there are 48 parking spaces available at the park-and-ride.”
Day said the system could conceivably work anywhere there are road sensors and drivers with GPS-enabled smartphones.
“Go to any city and watch the news, and you usually see a heat map when somebody is talking about traffic, just like the weather person,” he said. “[Cities] almost all have some sort of road sensor network in place.”
Your Car Will Predict Pedestrians’ and Other Drivers’ Next Moves
Now that your car knows what you’re going to do, and what the masses are going to do, it needs to know what the car 20 feet ahead is about to do. Car-to-car networks and advanced control algorithms can ensure there are no surprises, and hopefully someday no accidents. They can also improve engine efficiency and reduce emissions by preventing stop-and-go traffic.
Just this week, the U.S. Department of Transportation announced a pilot program that will let drivers test future connected car capabilities. Systems will enable cars to communicate with each other and with road infrastructure like traffic lights and railroad crossings, but drivers have to get used to it first. Clinics in a half-dozen cities will let humans test wirelessly connected car technology to see how well we can adapt.
If we don’t adapt well, cars will soon be equipped to deal with it. Researchers at MIT are developing new algorithms that incorporate models of human behavior to warn drivers of potential collisions, and assume control of the car to prevent a crash.
For the purposes of the system, driving is boiled down into two actions: braking and accelerating. Depending on which action the driver is taking at a given point in time, there are only so many possible outcomes for where the car will be next, according to MIT News, which featured the new algorithm last month. Domitilla Del Vecchio, assistant professor of mechanical engineering at MIT, also incorporated models that predict human behavior, such as when drivers slow down or speed up at an intersection.
The resulting system can determine the points in an intersection where vehicles are in danger of colliding, according to MIT. A car equipped with this algorithm will try to predict what the other car will do — also consulting traffic lights and its own onboard sensors — and act accordingly to avoid a crash.
For a future involving no human drivers at all, algorithms will have to account for variable numbers of cars acting in concert, accounting for the ripple effects of one action. For instance, if one car slows to avoid a crash, other cars must alter their behavior, too. A distributed control system would control acceleration, braking, lane changes and highway exits for all the cars in a given group.
Researchers at Carnegie Mellon University built a simulation that can prove such a system’s safety, even with multiple cars performing multiple complicated tasks. A team led by Andre Platzer, an assistant professor of computer science, started with just two cars in one lane. Then they added more cars to show it can work with an arbitrary number of vehicles, and added more lanes to show that number can vary, too. Ultimately, the system remains crash-proof regardless of the vehicles or lanes involved — on a straight highway, that is. Future simulations will have to account for variables like curved roads, Platzer said in a CMU release.
Control systems for autonomous car fleets are slightly farther in the future, however. Cars don’t need advanced algorithms to work in concert to cut congestion and prevent collisions — European researchers are already doing this, simply by allowing cars to communicate with each other. Systems like this can reduce emissions and improve engine efficiency by letting cars travel in a slipstream.
Swedish automaker Volvo has been testing “road trains,” which involve convoys of cars led by a professional driver. Cars can hook up to the train and connect via a wireless link, and an adaptive cruise control system will match the leader’s speed. Sensors in the cars will ensure everyone keeps a safe distance from each other. Earlier this year, Volvo tested a road train with a single car and a semitrailer.
The goal is to cut fuel use, cut congestion and make driving safer. The automaker hopes to deploy the technology by 2020.
And last month, German researchers found that just five cars communicating out of every 1,000 is enough to reduce congestion. Sensors in the cars can collect data and exchange it with other cars via local wireless networks, as well as relay it to a central traffic command center. Preliminary data from a one-year test showed that a minimum of five cars was sufficient to make an impact on congestion. The test was part of the European Union’s Dynamic Information and Application for Mobility with Adaptive Networks and Telematics Infrastructure (DIAMANT) project.
Your Car Won’t Change That Much — Its Computers Will
For automakers and for consumers, the good news is that most of these technologies can be implemented with a few simple codes and algorithmic changes — cars don’t have to reinvent the wheel to enable major improvements. With updates like these, driving could become much safer and much less frustrating within the next few years. Now if humans would just drive smarter, too…