, the team looked at an autonomous vehicle with 10 cameras whose data is being processed by 10 deep neural networks. After one hour of driving, the vehicle would make 21.6 million inferences.
If the technology really took off, and there were a billion autonomous vehicles on the world’s roads , the global fleet would make 21,600 trillion inferences per hour.In a scenario in which there are a billion AVs on the roads, driving for an hour a day, the power required to process all of that data would emit the equivalent of approximately 0.14 gigatons per year, or 0.3 percent of all global emissions, the study found.
To be clear, that is still significantly less than internal combustion vehicles currently produce – a recent studyfound that passenger vehicles currently account for 15 percent of global greenhouse gas emissions – but an hour of driving per day is quite a conservative estimate. Research suggests that autonomous vehicles are likely to increase the number of miles passenger vehicles travel every day.
The study’s authors believe, though, that finding efficiencies in algorithmic development and more specialized hardware could help reduce“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,” said Soumya Sudhakar, who conducted the study. “This has the potential to become an enormous problem.