a Study by MIT grad-student IEEE’s Soumya Sudhakar has been getting some press recently for claiming that a large fleet of self-driving cars in the distant future could cause as much greenhouse gas emissions as the current global network of computer servers, which is cited as producing 0.3% of modern emissions.
The study makes weak assumptions, and thus its conclusion is incorrect, but these types of studies are often latched onto by opponents of new technologies because of their confirmation bias, and used as propaganda. A bit of digging into the details, so I’ll be here.
They started by saying that the computers in today’s cars consume about 840 watts of power. That’s the equivalent of a typical home, or a few computers in a data center, though it’s a modest fraction of an electric sedan driving 60 mph, which uses about 15,000 watts while driving. What’s more, they expect the amount of energy needed to grow over time rather than shrink—perhaps to a point rivaling the energy used to move a car. The city-street sedan uses a little less than that, so that could start to be an appreciable part of that power overall.
Their calculations indicate that if there are a billion or more of these cars, each driving an hour a day, that adds up to a significant amount of electricity. If that electricity comes from our modern fossil-fuel-fueled grid, that could mean a lot of emissions. They do a lot of math on these incorrect assumptions to generate a bad result. They’re trying to forecast several decades to produce those numbers, and in one projection they imagine 3 billion cars.
Let’s analyze all the questionable assumptions
- It is true that some concept cars may draw power like this for their computing. There are also cars that draw much less. They imagine that the software and other sensors will not become more efficient than the prototypes, when in reality they will likely become vastly more efficient.
- The paper even assumes that computational load will grow exponentially and that the only salvation for power use is that AI processors are also getting exponentially more efficient — but not fast enough, and they’ll need to double in efficiency every year to catch up. The trajectory of computational growth in prototypes is not indicative of future computational needs.
- The predictions are based on the 2022 architecture where many cameras are constantly processing all the images. With other sensors, such as LIDARs, that need less computation, you can also avoid a lot of computation from cameras in areas of the image where nothing is known. This is just one of many tricks, most not yet invented, that will reduce the workload of vision – if vision is in fact the approach decades from now.
- Imagine that all cars are driving independently. It is unlikely that this will happen for a very long time, although all cars will have some features of this kind, even as driver assistance. However, the driver assistance functions are already less power.
- They assume a fossil-heavy grid like today, although massive efforts are underway to improve the emissions of that grid. However, as discussed below, these vehicles will not derive power from the ‘average’ grid and will likely derive almost entirely from surplus solar energy, other forms of surplus energy (nuclear, geothermal, some hydro, wind) and rarely of fossil energy.
- Almost all AV efforts currently are electric vehicles, and this is likely to continue. Autonomous vehicles have accelerated the shift from gasoline-powered cars to electric vehicles, which is a huge emissions gain for decades to come. Utility vehicles facilitate this switch because it eliminates all current issues with electric vehicles in the public’s mind. You don’t care what’s under the hood of the cab, its range or recharge time. There is almost no need to use fossil fuels.
The power source for self-driving electric vehicles
Today, our grid consists of a mixture of fossil fuels, nuclear energy, solar energy, wind energy, hydro energy and some other sources. Renewable energy sources such as solar and wind provide power only when the weather dictates. Nuclear plants provide power all the time whether you need it or not. Hydro and fossil energy can change during the day (sometimes slowly, sometimes quickly) based on expected and real demand. Full grid capacity is only used on hot summer afternoons and evenings to run air conditioning, although this should change in the future through the use of ice-based thermal storage.
Solar power is now the cheapest form of power plants to build in moderately sunny locations, but it only provides power in daylight. A ton of solar panels are expected to be installed to provide power by the afternoon. This same solar power will generate power from 8 am to 3 pm as well, although there may not be enough demand to use it. To avoid wasting it, energy companies will sell the surplus cheaply. Electric cars, especially autonomous ones, will be chasing after that deal — even self-driving vehicles will drive themselves to places to plug in during that time. They won’t want to charge during peak times unless they have to. It’s the peak times when fossil energy is being used – we don’t want to use it when there is more than enough power.
Nuclear plants provide power all day, including all night, but people don’t want a lot of power at night. It doesn’t save much to turn down a nuclear plant at night, so this is another great time to charge up these cars. Cars will do as much charging as they can when energy is cheap—their owners aren’t stupid—and that’s when energy is in excess, when energy doesn’t depend on fossils. They will do some peak charging if they don’t have a choice, but that would be the exception.
Some regions do not have or are shutting down nuclear power. They will need to rely on fossils, wind and water at night, as well as storage. This creates emissions risks because night is definitely the most suitable time to charge cars, since most of them are not running at that time. Charging can be done slower and it’s cheaper and better on the batteries – solar charging from 9:30 to 11:30 and 1:30 pm to 3 pm, when there is excess solar and more cars are idle, it needs to be done faster .
There are also some cars that, due to poor planning or unusual days, will need to charge at 7 p.m., which is the peak dig (and price peak) of the day. They will pay a heavy price for this and it will be their last resort.
I’m not a fan of battery replacement systems for consumer cars, although they can make sense for fleets. With battery swap, you can charge whenever there’s a surplus of power – green and cheap – without worrying about when the cars are on the road. You need to have more battery packs on hand—the ones in cars, the ones at charging stations, the ones waiting to be swapped—but you’re in complete control when charging.
Another modest negative note is that the most efficient combined cycle fossil fuel plants are slow to ramp in response to demand, so the ability of cars to buy surplus energy makes the problem of rampage go away, which may improve the economics of that fuel, but not by much, because again cars will be buying energy Only if it’s cheap. It also rewards greater efficiency, which is better than punishing for it.
Some predict a fully renewable grid before long, either with or even without nuclear power. Obviously, there is no major emissions problem for electric cars in this case. New efficiencies, warehousing technologies and demand-side management point the way to achieving this.
But even when the rest of the grid still has plenty of fossils in it, self-driving electric vehicles will use almost entirely low-emissions energy. The author of the paper could have saved herself all the math – in fact it hardly matters how much energy the calculation takes in cars in relation to emissions. It doesn’t matter much in terms of range, since not all of the MegaJoule that computers use is usable for navigation – so they’ll try to keep that level down if they can, but not because they’re worried about emissions.