A preview of this full-text is provided by Springer Nature.
Content available from Nature Energy
This content is subject to copyright. Terms and conditions apply.
Articles
https://doi.org/10.1038/s41560-020-0644-3
1Department of Engineering and Public Policy, Carnegie Mellon University, Pittsburgh, PA, USA. 2Department of Mechanical Engineering,
Carnegie Mellon University, Pittsburgh, PA, USA. 3Wilton E. Scott Institute for Energy Innovation, Carnegie Mellon University, Pittsburgh, PA, USA.
✉e-mail: venkvis@cmu.edu
Electric vehicles (EVs) form an increasing share of new
vehicle sales around the world. Several countries are seek-
ing to phase out vehicles with internal combustion engines.
Simultaneously, automated vehicles (AVs) are being tested on pub-
lic roads. Automation could reduce vehicle energy use through
smoother driving, platooning, shared mobility and optimal rout-
ing1–3. There is limited literature on the trade-offs between automa-
tion and electrification4. Early automated electric vehicles (AEVs)
may be heavier, need extra computing and sensor power, and (due
to the possible need for protruding sensors) be less aerodynamic
than EVs. It has been suggested that if these qualities substantially
reduce driving range, the first AVs will be gas–electric hybrids5,
which could slow the electrification of the light vehicle fleet.
In this paper we compare the vehicle-level energy use, range and
battery life of a vehicle equipped to attain Society of Automotive
Engineers (SAE) levels 4–5 automation to human-driven EVs, by
undertaking a careful consideration of the effect on vehicle-level
energy use of the different components needed for automated driv-
ing, as well as the potential increase in drag from LiDAR (light
detection and ranging). We find that automation is likely to reduce
EV range, and that the aerodynamic impact of LiDAR is a notable
driver of this reduction with suburban driving, while computing
loads have a greater effect for city driving. Across different EVs,
we find a median reduction in the range of 3–5% for a suburban
drive profile and no drag impacts from LiDAR. Including LiDAR
increases the median range reduction to between 8–12%. A reduc-
tion in range will lead to more frequent charging, speeding up bat-
tery degradation, but we find that this impact on battery longevity
is negligible.
AEV energy consumption
Sripad and Viswanathan6 developed a physics-based vehicle dynam-
ics model to estimate the energy demands of an EV given a realistic
driving profile. They determined the battery size needed for a given
vehicle range or equivalently, the range of an EV given the battery
size. Using a realistic velocity profile with a 1 s temporal resolution,
the model calculates the instantaneous power needed each second
to overcome vehicle inertia, aerodynamic drag and road friction.
We extend this model for AEVs by adding the weight of the differ-
ent components to the mass of the vehicle and battery pack, increas-
ing the drag coefficient for automated solutions with a roof-based
spinning LiDAR. If no LiDAR is used, or if solid-state LiDAR that
is incorporated into the aerodynamic profile of the vehicle is used,
the increase in drag is zero. We also modify the velocity profile to
account for potentially smoother driving and add the computing
and sensor loads at each second. Keeping track of the total energy
used, we repeat the driving profile until the battery is fully depleted.
This gives us an estimate of the AEV range for a given battery capac-
ity. We then compare this AEV range to that of the EV to understand
how automation affects vehicle range. We run this simulation for
two types of velocity profile: the California Unified Cycle Driving
Schedule, which is a composite profile that is a city–highway mix,
and the Urban Dynamometer Driving Schedule, which is a city-only
profile (Supplementary Fig. 1). Full details of the physics model are
provided in the Methods.
There are several different combinations of sensor hardware
that are currently being tested on vehicles that aim to achieve full
automation. For example, some developers are using solutions that
include LiDAR, while others are relying solely on cameras and
radar. There are also differences in the choice of vendor for LiDAR
or radar, and in the number of sensors. Given the numerous pos-
sible combinations, we assume a uniform distribution for the sensor
and connectivity load. We bound this between 30 W and 150 W. The
lower case represents a low-powered LiDAR solution such as the
15 W Ouster OS1 system7 along with two Bosch mid-range radars
(MRRs) of 4.5 W each8, three Point Grey Dragonfly cameras of
1.5 W each9 and 1.5 W for connectivity. Some industry developers
have also suggested that a sensor package without LiDAR is suf-
ficient for high-level automation. As such, the lower case could
also represent an optical-only system with no LiDAR and nine
cameras (1.5 W each), two MRRs of 4.5 W each and the remain-
ing 7.5 W for communication and connectivity. The upper estimate
Trade-offs between automation and light vehicle
electrification
Aniruddh Mohan 1, Shashank Sripad 2,3, Parth Vaishnav 1,3 and
Venkatasubramanian Viswanathan 2,3 ✉
Weight, computing load, sensor load and possibly higher drag may increase the energy use of automated electric vehicles
relative to human-driven electric vehicles, although this increase may be offset by smoother driving. Here, we use a vehicle
dynamics model to evaluate the trade-off between automation and electric vehicle range and battery longevity. We find that
automation will likely reduce electric vehicle range by 5–10% for suburban driving and by 10–15% for city driving. The effect on
range is strongly influenced by sensor drag for suburban driving and computing loads for city driving. The impact of automation
on battery longevity is negligible. While some commentators have suggested that the power and energy requirements of auto-
mation mean that the first automated vehicles will be gas–electric hybrids, our results suggest that this need not be the case if
automakers can implement energy-efficient computing and aerodynamic sensor stacks.
NATURE ENERGY | VOL 5 | JULY 2020 | 543–549 | www.nature.com/natureenergy 543
Content courtesy of Springer Nature, terms of use apply. Rights reserved