A new study conducted by MIT researchers has found that in the future, the amount of energy required to run computers on board an international fleet of AVs could generate the same amount of greenhouse gas emissions as all the world’s current data centers. The study explored the potential energy consumption and related carbon emissions if autonomous vehicles were widely adopted.
At present, the data centers that house the physical computing infrastructure used for running applications account for approximately 0.3% of global greenhouse gas emissions (GHG). As there has been little focus on the potential footprint of AVs, MIT researchers developed a statistical model to study the potential issue.
The research team calculated that one billion AVs driving for an hour each day, with each vehicle’s computer using 840W, would consume enough energy to generate roughly the same number of emissions as global data centers do currently. Researchers also found that in 90% of modeled scenarios, to keep AV emissions from surpassing present day data center emissions, the vehicle would have to use under 1.2kW of computing power. To achieve this target, more efficient hardware for AVs would be required.
In one test, the team modeled a scenario in 2050 where 95% of the global fleet is made up of AVs. During this scenario, computational workloads doubled every three years and the Earth continued to decarbonize at the current rate. Upon completion of this simulation, researchers found that hardware efficiency would need to double faster than every 1.1 years to keep emissions under 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 on board autonomous vehicles,” said first author Soumya Sudhakar, a graduate student in aeronautics and astronautics. “This has the potential to become an enormous problem. But if we get ahead of it, we could design more efficient autonomous vehicles that have a smaller carbon footprint from the start.”
The paper was written by Sudhakar and her co-advisors Vivienne Sze, associate professor in the Department of Electrical Engineering and Computer Science (EECS) and a 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 research can be found in the January/February issue of IEEE Micro.
A framework was built by the researchers to investigate the operational emissions from computers on board a global fleet of fully autonomous EVs.
The model consists of the number of vehicles in the global fleet, the power of each computer on each vehicle, the hours driven by each vehicle and the carbon intensity of the electricity powering each computer.
“On its own, that looks like a deceptively simple equation,” said Sudhakar. “But each of those variables contains a lot of uncertainty because we are considering an emerging application that is not here yet.”
Some research suggests that the amount of time driven in AVs may increase because the vehicles can facilitate people being able to multi-task while driving in addition to young or old demographics being able to drive more. Alternative research, however, suggests that time spent driving may get lower because faster routes could be found to get occupants to their destinations faster.
Additionally, researchers also needed to consider and model advanced computing hardware and software that doesn’t exist yet. To achieve this, the team modeled the workload of a popular algorithm for AVs called a multi-task deep neural network. The team investigated how much energy the deep neural network used if it was processing multiple high-resolution inputs from several cameras with high frame rates at the same time.
When the team used the probabilistic model to explore different scenarios, Sudhakar was said surprised by how quickly the algorithms’ workload added up.
For example, if an AV has 10 deep neural networks processing images from 10 cameras, and the vehicle drives for one hour a day, it will make 21.6 million inferences each day. One billion vehicles would make 21.6 quadrillion inferences.
“After seeing the results, this makes a lot of sense, but it is not something that is on a lot of people’s radar,” said Karaman. “These vehicles could actually be using a ton of computer power. They have a 360° view of the world, so while we have two eyes, they may have 20 eyes, looking all over the place and trying to understand all the things that are happening at the same time.”
In addition to AVs being used to transport people, they will also be used for transporting goods, meaning a large amount of computing power could be distributed along global supply chains, explained Karaman. The team’s model only considers computing, and not the energy consumed by vehicle sensors, or the emissions generated during manufacturing, he added.
To ensure that AVs consume less than 1.2KW of energy for computing power, manufacturers must create increasingly efficient hardware at a much faster pace with efficiency doubling every 1.1 years.
In order to boost that efficiency, more specialized hardware could be used that is designed to run specific driving algorithms. Because researchers know the navigation and perception tasks required for autonomous driving, it could be easier to design specialized hardware for those tasks, explained Sudhakar. With vehicles estimated to have a 10- or 20-year lifespan, a challenge arises in future-proofing specialized hardware to enable them run new algorithms.
Moving forward, researchers could make the algorithms more efficient, meaning they would require less computing power. This is viewed as challenging because the trade-off of accuracy for more efficiency could affect vehicle safety.
Having demonstrated the framework, the MIT researchers will continue to explore hardware efficiency and algorithm improvements. Furthermore, the team states that the model can be enhanced by characterizing embodied carbon from AVs— the carbon emissions generated when a car is manufactured — and emissions from a vehicle’s sensors.
“We are hoping that people will think of emissions and carbon efficiency as important metrics to consider in their designs,” said Sze. “The energy consumption of an autonomous vehicle is really critical, not just for extending the battery life, but also for sustainability.”
