Self-driving cars may not be the boon we think they are. The computers that power them have the potential of generating high levels of carbon emissions on a global scale. Experts at the Massachusetts Institute of Technology (MIT) say that wide adoption of fully autonomous cars will result in emissions equal to those produced by all data centers the world over. Preventing this state of affairs means hardware efficiency must improve at a rate far more rapid than we are currently able to accomplish.
Greenhouse gas emissions are one of the most important reasons supplied for moving from gas-powered to electric vehicles. And self-driving cars run on electricity. But the computers that run them leave a Bigfoot-sized carbon footprint. As the world moves closer to going fully-electric on the road, we may want to stop and consider whether autonomous cars cause more harm than benefit to our environment.
Data centers that house the computing infrastructure used to run applications are known to produce large amounts of global greenhouse gas emissions. These data centers, according to the International Energy Agency, are responsible for around 0.3 percent of worldwide emissions, an amount equal to that of all carbon produced by the country of Argentina in a single year. MIT researchers found that little attention has been paid to the gas emissions generated by the computers necessary to power self-driving cars. They decided to build a statistical model to study the issue. What they found is that for every 1 billion self-driving cars driven for one hour a day using a computer consuming 840 watts, the amount of energy consumed is enough to generate emissions equal to that of all data centers
Self-Driving Car Emissions
In order to keep self-driving car emissions from exceeding those of our current rate of worldwide data center emissions, found the researchers, each vehicle would need to keep computing power to under 1.2 kilowatts. The only way to do that is to increase efficiency in the hardware used to power autonomous vehicles. Experts found this to be the case in 90 percent of the modeled scenarios they studied. One of the models showed that in a case where 95 percent of vehicles worldwide are self-driving, by 2050, computational workloads will double every three years. Should the world continue to decarbonize at our current rate, computer hardware efficiency will need to double faster than every 1.1 years to keep emissions below those levels.
“If we just keep the business-as-usual trends in decarbonization and the current rate of hardware efficiency improvements, it doesn’t seem like it is going to be enough to constrain the emissions from computing onboard autonomous vehicles. This has the potential to become an enormous problem,” says lead author Soumya Sudhakar, a graduate student in aeronautics and astronautics. “But if we get ahead of it, we could design more efficient autonomous vehicles that have a smaller carbon footprint from the start.”
Improving Hardware Efficiency
Getting ahead of it, however, means inventing more advanced computing hardware and software than currently exist. Since researchers already know which navigation and perception tasks are required to run a self-driving car, it should make the task of designing the necessary specialized hardware easier, says Sudhakar. The main thing, according to the MIT research team, is to shed light on a problem that no one seems to have considered until now: that self-driving cars may be worse for the environment than gas-powered vehicles, unless we get a lot better at making the computers that run them, more efficient—and fast.
“We are hoping that people will think of emissions and carbon efficiency as important metrics to consider in their designs. The energy consumption of an autonomous vehicle is really critical, not just for extending the battery life, but also for sustainability,” says co-author of the study Vivienne Sze, associate professor in the Department of Electrical Engineering and Computer Science (EECS) and a member of the Research Laboratory of Electronics (RLE).
