|Fig. 1: Google self driving car in Mountain View, CA, USA. (Source: Wikimedia Commons)|
In the U.S., transportation is the second largest contributor to greenhouse gas emission (GHGE), accounting for about 26% of total emission.  Notably, personal vehicles and light trucks constitute more than half of this share.  The energy landscape in the U.S. is predicted to closely parallel the GHGE statistics. [2,3] One could suspect that this landscape may change with the introduction and eventual spread of self-driving cars. An example of such a car is shown in Fig. 1.
Earlier this year, the first Uber self-driving car rolled out into the streets of Pittsburg, PA.  Autonomous vehicles are expected to increase in prevalence in the not-so-distant future. In 2012 a group of members of IEEE predicted up to 75% adoption by 2040.  While this panel did not estimate the eventual adoption rates, reaching a high "critical mass" seems plausible. Given the size of transportation's energy-market, analyzing the impact of self-driving cars is of paramount importance.
To the author's knowledge at the time of this writing, only two works have attempted to quantitatively estimate the magnitude and sources of energy impact of the self-driving cars. [6,7] Brown et al. have estimated a range from 95% fuel savings, in the case where only positive externalities occur, to 173% fuel usage increase in the case that only negative externalities occur.  More recently, Wadud et al. have estimated a range spanning from a reduction in fuel usage by almost half to a doubling of fuel use.  As is clear, the exact scenario depends on how the consumers react. In what follows, I will briefly discuss some sources of increase and decrease in efficiency considered by the aforementioned works. Following Brown et al., I will assume the utilization rate is high (plausible in some models suggested in Silberg and Wallace). [6,8]
Under a car-sharing model, individuals could travel with the car size that fits their needs. Research shows that the average occupancy of vehicles in the U.S. is 1.67 individuals per personal vehicle.  As a result, often a vehicle with heavier mass than required will take the trip, which reduces fuel efficiency.
Furthermore, because self-driving cars do not suffer from human errors (often due to, inattention, distraction, driving aggressively, etc.), automation can decrease the rate of crashes.  As the number of crashes is reduced, heavy safety features can be removed from the vehicle and the vehicle can be made from lighter materials without increasing the mortality rate.
Self-driving cars can reduce congestion via multiple mechanisms. The most immediate pathway is through reducing the number of crashes and following good driving practices. However, if the vehicles can communicate, they can further reduce congestion by coordinating their movements (see Hishino for some examples of how clustering/information sharing could reduce travel time and congestion).  Both the decrease in time spent in congestion and reduction in wasteful braking due to coordination of the cars with each other and potentially with the topography can be sources of reduction in energy consumption.
|Fig. 2: Estimated energy impact of various mechanisms.  (Source: A. Pourshafeie)|
Further improvements in transition efficiency can be made through "platooning". Platooning is a grouping where multiple vehicles follow each other, front to back, at close distances. The goal in platooning is to reduce the overall drag force; therefore, the efficiency of platooning increases as the number of cars increases and the distance between cars decreases. Coordinated self-driving cars can follow each other at distances that may be unsafe for human drivers; this not only further reduces congestion, but also reduces the overall drag. [6,7]
Perhaps the biggest wild card is the long-term reaction of the population. The simplest effect would be that new groups that were previously not able to drive can now use self-driving vehicles. Harper et al. estimated an upper bound of 14% in increased traveled mileage while Wadud et al. has estimated a smaller range (see Fig. 2). 
Furthermore, with comfort of automated-vehicles, driving time will be less of a burden and individuals may choose to spend more time in their vehicles. This reduction in the down time of driving and possible reductions in cost of production and maintenance of a vehicle may encourage more driving. This increase in driving will lead to increased fuel consumption. Furthermore, as the time spent in vehicle increases, the vehicles comfort features may increase, which leads to heavier vehicles.
We have included the estimated effect size from the pathways above in Fig 2. These estimates are generated from the estimates provided by of Wadud et al.; however, they have been adjusted for slight differences in the class definitions. It is important to emphasize that many of the benefits mentioned above may appear only if the adoption rate is high; on the other hand, signs of increase in fuel usage may appear even at a small adaptation rate.
In conclusion, while self-driving cars will likely reduce the energy consumption per mile, the overall effect is not clear as they could increase the overall usage of vehicles. Prices and policy could have a great impact on the final outcome.
© Armin Pourshafeie. The author grants permission to copy, distribute and display this work in unaltered form, with attribution to the author, for noncommercial purposes only. All other rights, including commercial rights, are reserved to the author.
 "Fast Facts: U.S. Transportation Sector Greenhouse Gas Emissions, 1990-2014", U.S. Environmental Protection Agency EPA-420-F-16-020, June 2016.
 "Monthly Energy Review: August 2016," U.S. Energy Information Administration, DOE/EIA-0035(2016/8), August 2016.
 "Annual Energy Outlook 2015," U.S. Energy Information Administration, DOE/EIA-0383(2015), April 2015, Table A7.
 H. Somerville, "Uber Debuts Self-Driving Vehicles in Landmark Pittsburgh Trial," Reuters, 14 Sep 16.
 J. Motavilli "Self-Driving Cars Will Take Over By 2040," Forbes, 25 Sep 12.
 A. Brown, J. Gonder, and B. Repac, "An Analysis of Possible Energy Impacts of Automated Vehicle," in Road Vehicle Automation, ed. by G. Meyer and S. Beiker (Springer, 2014), pp. 137-153.
 Z. Wadud, D. MacKenzie, and P. Leiby, "Help or Hindrance? The Travel, Energy and Carbon Impacts of Highly Automated Vehicles," Trans. Res. A-Pol. 86, 1, (2016).
 G. Silberg and R. Wallace, "Self-Driving Cars: The Next Revolution," KPMG, 2012.
 S. C. Davis, S. W. Diegel, and R. G. Boundy, "Transportation Energy Data Book, Edition 31," Oak Ridge National Laboratory, ORNL-6987. July 2012, pp. 8-11
 "Critical Reasons for Crashes Investigated in the National Motor Vehicle Crash Causation Survey," U.S. Department of Transportation, DOT-HS-812-115, February 2015.
 S. Hoshino, "Reactive Clustering Method for Platooning Autonomous Mobile Robots," IFAC Proc. Vol. 46, No. 10 152 (2013.
 C. D. Harper et al., "Estimating Potential Increases in Travel With Autonomous Vehicles For the Non-Driving, Elderly and People With Travel-Restrictive Medical Conditions," Trans. Res. C-Emer. 72, 1 (2016).