//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
I’m a big believer in the “law of unintended consequences,” and I have found that there are at least three of variations of this law:
- Unintended, unforeseen consequences that are inconsequential or even benign: Sometimes, you just get lucky.
- Unintended, unforeseen consequences that are not at all benign but instead have adverse consequences. These may even be counterproductive, with the opposite of what you intended to accomplish in the first place.
- Finally, those unintended, adverse consequences you know of or found out about early on but chose to ignore because they got in the way of your primary objective. Let’s call this the “not my problem” or “head in the sand” scenario.
A research study I recently saw reminded me yet again of this “law” and its second and third variations in the above bulleted list, even though I am normally very skeptical of almost all studies that use various projections, extrapolations and estimates. Many of these studies and their research papers ignore or minimize constraints like nonlinearities or inflection points that occur in the real world. After all, a simplistic discussion that merely extrapolates the thermal properties of water would not acknowledge what we know happens at 0˚C and 100˚C.
Still, these studies can serve a purpose: They can at least make you stop and think, and that’s what one recent one did for me. A 12-page paper from a team at MIT and published in IEEE Micro looked at the energy consumption required to support the processing power of autonomous vehicles (Figure 1). (There’s also a corresponding 10-minute TEDx presentation.)
The authors conclude that as these self-driving cars at various levels become more autonomous and many reach Level 5 (fully autonomous), and the number of such cars also grows worldwide, the aggregate electric-power requirements for their “smarts” will be a very large number. In fact, their summary gets right to the point: “1 billion autonomous vehicles, each driving for one hour per day with a computer consuming 840 W, would consume enough energy to generate about the same amount of emissions as data centers currently do.”
Well… maybe yes, maybe not. Despite the academic veneer and a detailed table explaining their model and its embedded assumptions, the inherent nature of the model has many rough estimates, sophisticated guesses and outright assumptions. I’d compare it more to a fancy “back of the envelope” assessment (Figure 2) rather than a rigorous academic analysis. Step back, and there’s the reality that any assessment of the future for a situation such as this is simply not possible.
Perhaps the final number has a band of error of ±50%, and I don’t think that doing an analysis of variance (ANOVA) would make much difference. I still see reports offered for sale by market-research organizations that assess the market for some widget or service five and even 10 years out, to three significant figures, which is totally meaningless. It seems that excess and meaningless precision is not limited only to academic studies.
I suspect that for designers of autonomous vehicles, their greater immediate concern is the heat load that the processing system adds to the vehicle. In addition, drawing nearly 1 kW continuously from the vehicle batteries reduces all-important range.
There are other aspects of these vehicles that I haven’t seen mentioned much when reading the usually optimistic stories in the general media describing how they will reduce accidents, save lives and so on (and remember, the pundits said these vehicles would be on sale to the public by 2020). The first is how much the full Level 5 capability would add to the cost of the car. I’m guessing (sorry) that it is likely between $5k and $20k, but those are just guesses.
In addition, the cost of ownership will likely be much higher with more complex regular “tune-ups” (imaging-system alignments, for example), and certainly insurance and repair costs will be higher (maybe fewer accidents but much higher per-incident repair costs). These facts are conveniently overlooked by promoters of these vehicles, who assume that almost everyone will run out to buy one as soon as they are on the market or need to replace their current vehicles. I don’t think that’s going to happen.
Despite my issues with the study and its model, I think the issue the authors raise is an important one. The authors conclude that there would have to be a sharp drop in processing-system power consumption by the vehicle “smarts” to reduce aggregate power use significantly and give some suggested numbers. Better, less processing-intensive algorithms will also help, they added.
Call me cynical, but it seems that the design reality is that system demands and thus power use often expand to the maximum the system can provide or tolerate. Remember, when transistors and ICs were devised and then went to ever-lower voltage/power, the per-function dissipation dropped by several orders of magnitude—you can’t argue with that fact.
However, coincidentally with this good news, user requirements increased at a greater rate, so now you have server racks pushing almost 10 kW each—an almost unthinkable figure just a decade ago, up from about 2.4 kW each in 2011 according to one survey (I have seen others, and those numbers are similar) (Figure 3).
What’s your view on the power-consumption scenario of autonomous vehicles? Will it be a significant additional burden on the grid regardless of ultimate power source? Are such estimates even meaningful? Will dissipation be reduced by lower-power electronics, or will algorithm and processing requirements to get to from Level 4 (high automation) to Level 5 (full automation) vehicles negate any savings?
Will we look back at this in 10, 20 or more years and wonder why someone didn’t connect all the dots about this energy use (note that “connect the dots” is a favorite cliché of the hindsight experts)? Or will people say, “What was all the fuss about?” Only time will tell.
Related content
Autonomous Vehicles, Autonomous Bugs
Can you use supercaps to power electric vehicles?
