In the future, the energy needed to run the powerful computers of a global fleet of autonomous vehicles could produce as many greenhouse gases as all the data centers in the world today.
That is one of the main findings of a new study by MIT researchers, which looked at the potential energy consumption and associated carbon dioxide emissions of the widespread adoption of autonomous vehicles. The Data centers that house the physical computing infrastructure used to run the applications are widely known for their large carbon footprint: they are currently responsible for about 0.3 percent of global greenhouse gas emissions, or about the same amount of carbon as Argentina produces every year. for the International Energy Agency.
Recognizing that less attention has been paid to the potential footprint of autonomous vehicles, MIT researchers created a statistical model to investigate the issue. They found that a billion autonomous vehicles, each driving an 840-watt computer used enough energy per hour per day to produce roughly the same amount of emissions as today’s data centers.
The researchers also found that in more than 90 percent of the scenarios modeled, each vehicle would need to use less than 1.2 kilowatts of computing power to prevent its emissions from reaching those of current data centers, which would require more efficient hardware. In one scenario—where 95 percent of the world’s vehicle fleet is autonomous by 2050, workloads are estimated to double every three years, and global carbon emissions continue at current levels—they found that equipment efficiency would have to double faster than every 1.1 years to keep emissions below these levels.
“If we simply maintain conventional decarbonization trends and the current rate of improvement in equipment efficiency, it does not seem enough to limit emissions from autonomous vehicle calculations.” This could become a big problem. “But if we move forward with it, we could design more efficient autonomous vehicles from the ground up with a smaller carbon footprint,” says first author Soumya Sudhakar, a graduate student in aeronautics and astronautics.
Sudhakar wrote the paper with his advisers, Vivienne Sze, associate professor in the Department of Electrical Engineering and Computing (EECS) and member of the Research Laboratory of Electronics (RLE), and Sertac Karaman, associate professor of aeronautics and astronautics and director of the Laboratory for Information and Decision Systems (LIDS). The study appears today in the January-February issue of IEEE Micro.
Modeling Emissions
Researchers have created a framework to study the operational emissions of computers in a global electric vehicle that is fully autonomous, meaning it does not need a backup driver.
The model is a function of the number of vehicles in the global fleet, the power of the computer in each vehicle, the hours each vehicle is driven, and the carbon intensity of the electricity that operates each computer.
“Simply put, that looks like a deceptively simple equation.” “But all these variables carry a lot of uncertainty because we are considering an emerging application that doesn’t exist yet,” says Sudhakar.
For example, some studies suggest that self-driving vehicles can increase driving time because people drive more often and the young and old drive more. But other research shows that travel times can be reduced because algorithms can find optimal routes to get people to their destinations faster.
In addition to accounting for these uncertainties, the researchers also had to model complex computer hardware and software that did not yet exist. To achieve this, they modeled a popular algorithm for autonomous vehicles known as a deep neural network for multitasking because it can perform multiple tasks simultaneously. They investigated how much power this deep neural network would consume if it processed multiple high-resolution inputs from many high-frame cameras simultaneously.
When they used the probabilistic model to explore different scenarios, Sudhakar was surprised at how quickly the algorithms’ workload grew.
For example, if an autonomous vehicle has 10 deep neural networks that process images from 10 cameras and drives for one hour a day, it will draw 21.6 million conclusions per day. A billion vehicles would make 21.6 trillion conclusions. To put this in perspective, all of Facebook’s data centers around the world make several trillion inferences every day (1 quadrillion is 1,000 trillion).
“It makes a lot of sense after seeing the results, but it’s not visible to many people.” These vehicles can actually use a ton of computing power. “They have a 360-degree view of the world, so if we have two eyes, they can have 20 eyes looking everywhere and trying to make sense of all the things that are happening at the same time,” says Karaman.
He says the Autonomous vehicles would be used to move both goods and people, so vast amounts of computing power could be distributed across global supply chains. And their model only considers data processing; it does not consider the energy consumed by the vehicle’s sensors or the emissions generated during production.
Keeping emissions under control
To keep emissions out of control, researchers have found that each autonomous vehicle must consume less than 1.2 kilowatts of energy for data processing. For this to be possible, computer hardware must become more efficient at a significantly faster rate, doubling in efficiency approximately every 1.1 years.
One way to improve this efficiency could be with more specialized hardware designed to run specific driving algorithms. As researchers become familiar with the navigation and perception tasks required for autonomous driving, it might be easier to design specialized devices for those tasks, Sudhakar says. But vehicles typically have a lifespan of 10 or 20 years, so one challenge in developing specialized hardware would be to “future-proof” it so it can use new algorithms.
Researchers could also make algorithms more efficient in the future, in which case they would need less computing power. However, it is also difficult, as it may be possible to trade off accuracy to increase efficiency.
Source: MIT
Get more news and insights about Global Automobile Industry here
