PreprintPDF Available

Drone flight data reveal energy and greenhouse gas emissions savings for small package delivery

Preprints and early-stage research may not have been peer reviewed yet.


The adoption of Uncrewed Aerial Vehicles (UAVs) for last-mile deliveries will affect the energy productivity of package delivery and require new methods to understand the associated energy consumption and greenhouse gas (GHG) emissions. Here we combine empirical testing of 187 quadcopter flights with first principles analysis to develop a usable energy model for drone package delivery. We develop a machine-learning algorithm to assess energy use across three different flight regimes: takeoff, cruise, and landing. Our model shows that, in the US, a small electric quadcopter drone with a payload of 1 kg would consume approximately 0.05 MJ/km and result in 41 g of CO$_{2}$e per package. The energy per package delivered by drones (0.19 MJ/package) can be up to 96\% lower than conventional transportation modes. Our open model and generalizable coefficients can assist stakeholders in understanding and improving the energy use of drone package delivery.
Drone flight data reveal energy and greenhouse gas emissions savings
for small package delivery
Thiago A. Rodrigues1,2
, Jay Patrikar3
, Natalia L. Oliveira4,5
H. Scott Matthews1
, Sebastian Scherer3
, Constantine Samaras 1, 2
The adoption of Uncrewed Aerial Vehicles (UAVs) for last-mile deliveries will affect the energy produc-
tivity of package delivery and require new methods to understand the associated energy consumption and
greenhouse gas (GHG) emissions. Here we combine empirical testing of 187 quadcopter flights with first
principles analysis to develop a usable energy model for drone package delivery. We develop a machine
learning algorithm to assess energy use across three different flight regimes: takeoff, cruise, and landing.
Our model shows that, in the US, a small electric quadcopter drone with a payload of 1 kg would consume
approximately 0.05 MJ/km and result in 41 g of CO2e per package. The energy per package delivered
by drones (0.19 MJ/package) can be up to 96% lower than conventional transportation modes. Our open
model and generalizable coefficients can assist stakeholders in understanding and improving the energy use
of drone package delivery.
Key words: quad-copter drone, last-mile delivery, energy consumption, greenhouse gas emissions, robot
delivery, autonomous delivery.
1 Introduction
Achieving large improvements in the energy productivity of the freight transportation sector is challenging,
especially in the overwhelmingly petroleum-powered transport sector where medium and heavy trucks in the
US comprises 24% of transportation energy use. This sector is responsible for 37% of transportation-related
greenhouse gas (GHG) emissions while light-duty vehicles comprise 57% of transportation GHG emissions and
64% of transportation energy use. In addition, transportation remains a large source of nitrogen oxides (NOx)
and other air pollutants [1]. However, the way that consumers are obtaining goods in the U.S. is changing
rapidly [2].
Even before COVID-19, the growing demand for fast, contactless deliveries has been driving firms to experi-
ment with automated package delivery vehicles, such as Uncrewed Aerial Vehicles (UAVs) that can avoid traffic
in urban centers and rapidly reach rural areas that would not be served otherwise [3, 4]. Initial survey data of
483 customers in Portland, Oregon by Pani et al. [5] show that COVID-19 is contributing to an environment
where more than 60% of online customers are willing to pay extra to receive their packages using autonomous
delivery robots. Nevertheless, along with technology and policy challenges, increasing shipping costs is one of
the main limitation for the adoption of autonomous delivery vehicles [6].
The appeal of delivery robots also reflects new physical distancing demands to avoid the spread of coronavirus
in product deliveries [7]. As autonomous delivery technologies advance, new companies emerge to compete for
this market niche [8, 9, 10, 11, 12]. At the same time, alternative transport modes, such as electric cargo
bicycles, are becoming cost-effective alternatives to delivery trucks for short-distance deliveries [13], drastically
reducing the CO2emissions of last-mile delivery in highly dense metropolitan areas [14]. With the increased
electrification of delivery vehicles, the energy consumption and environmental impacts of the transportation
sector are expected to change drastically over the coming years [15, 16], and both technology and demand
are primary drivers. Widespread adoption of UAVs to replace a portion of first/last-mile truck pickups and
deliveries could reshape this sector by changing demand patterns and shifting fuel demands from fossil fuels to
electricity. Autonomous Delivery Robots are coming to the transportation sector, but how these vehicles and
systems could be designed to maximize energy productivity is less clear.
1Department of Civil and Environmental Engineering. Carnegie Mellon University. 5000 Forbes Avenue, Pittsburgh, 15213,
2Corresponding authors: and
3Robotics Institute, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, 15213, USA
4Department of Statistics and Data Science, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, 15213, USA
5Machine Learning Department, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, 15213, USA
arXiv:2111.11463v1 [eess.SY] 22 Nov 2021
So far, a few studies have estimated the energy consumption of quadcopter vehicles and the energy esti-
mations vary considerably among the different methods used [17]. Some studies have created models based on
theoretical principles [18, 19, 20, 21, 22, 23, 24, 25, 26], while others have developed models based on regression
models built on small flight samples [20, 27, 28, 24]. Finally, a comparison of the energy consumption and
greenhouse gas (GHG) emissions between package delivery UAVs and different transportation modes have been
estimated by a few studies [23, 29, 30, 31], but alternative emerging delivery modes, such as electric cargo
bicycles are not included.
Here we help stakeholders and researchers understand the energy use of uncrewed aerial package delivery
drones. We provide an energy model based on extensive empirical data from 187 flights of a quadcopter
drone DJI Matrice 100, from which we developed a novel and high-resolution dataset [32] of package delivery
drone energy use. We also develop an algorithm that automatically identifies the flight regime across takeoff,
cruise and landing. We show the impact of the cruise speed and payload mass on the drone’s range, provide
generalizable energy use coefficients, and we compare the energy consumption and GHG emissions of a small
quadcotper drone to other last-mile transportation modes on a energy per package basis.
2 Results and Discussion
We collected data on 187 flights to assess the power profile of a package delivery drone given a set of operational
parameters (payload, altitude, and speed during cruise). The data, available at [32], and a data descriptor [33],
provide the details of the experiment. In addition, we have developed an algorithm that separates the data into
three different flight regimes: takeoff, cruise, and landing, in order to better understand the energy consumption
profile during flight, see Supplementary Figures S1 to S4.
Then, we conducted a first principles analysis and developed a model to estimate the energy required to
power a quadcopter. Each of the flight regimes were modeled separately, that is, each energy model was treated
as a model class and three different optimal models from that class were selected, one per regime. In order to
fairly compare the model classes’ performance and avoid overfitting, we split the data into train and test folds
following a stratification strategy by flight ID number. With 120 flights, the training fold was used to estimate
the parameters of each model, which were then applied to the remaining 67 flights from the test fold in order
to evaluate the performance of the energy models on unseen data.
2.1 Energy Model
Our energy (E) model uses the Induced Power (Pi), which is the power required to overcome gravity in a
hover-no-wind situation, as a parameter estimator of the average power observed throughout the flight.
E= (b1Pi+b0)t(1)
where, tis the flight duration, and b1and b0are coefficients that linearly correlate Piand the average power
throughout the flight. The induced power, used in eq. 1, is calculated as
2ρA (2)
where, mis the total mass of the drone (including the payload), gis the acceleration of gravity, ρis the air
density, and Ais the total area under the propellers.
The estimated coefficients and their standard errors are shown in Table 1.
Table 1: Estimated coefficient ±bootstrap standard error
Coef. Take off Cruise Landing
b11.97 ±0.08 1.69 ±0.06 1.62 ±0.14
b013.8 ±0.01 16.8 ±0.01 -4.7 ±0.01
The coefficients shown in Table 1 were obtained by performing a linear regression between Piand the average
power observed throughout each of the 120 flights. The results were then applied to the remaining flights and
the absolute relative error was 3% on average, proving the accuracy of the energy model in terms of estimation
of energy consumption.
With the energy model validated, we estimated the energy consumption of a package delivered by a small
quadcopter drone. Figure 1a shows the total Energy Consumption for a two-way delivery trip (delivery and
return) according to the delivery distance and the total weight of the drone (with payload) operating at a cruise
speed of 4 and 12 m/s and altitude of 100 m. Our analysis also shows that variations in the cruise speed have
great impact on the total energy consumption per trip and consequently range of the drone (Figure 1a). The
total time of flight is reduced as the speed increases, which results in longer distances for the same amount of
energy. In addition, we calculated the GHG emissions per package delivered based on the US electricity grid,
upstream electricity generation, and battery life-cycle emissions according to the delivery distance (Figure 1b)
Figure 1: (a) Total energy consumption by distance of delivery varying payload mass and cruise speed. (b)
CO2e emissions according to the delivery distance varying according grid emission factor and battery life cycle
emissions. The total energy and CO2e correspond to takeoff, cruise from the origin to destination and landing
loaded, and takeoff, cruise from destination to origin and landing empty. As an energy limitation, the nominal
capacity of LiPO TB48D battery is 130 Wh [34]. Altitude during cruise of 100 m, takeoff speed 2.5 m/s and
landing speed 2 m/s.
2.2 Comparison between different transportation modes
We compared the energy consumption of quadcopter drones against diesel and electric medium-duty trucks and
small vans, and electric cargo bicycles.
Figure 2: Energy Consumption for different transportation modes. Error bars represent variations in (a) driving
styles and vehicle characteristics, and (b) number of packages delivered per distance.
Total energy consumption per distance of small quadcopter drones is among the lowest across transportation
modes, as the vehicle is small, light, and has lower payload capacity (Figure 2a). However Figure 2b shows
the energy consumption per package of drone-equivalent deliveries, i.e., assuming that all packages delivered
by the other modes are within the payload and space capacity of a small quadcopter drone [35]. On an energy
consumption per package basis, small quadcopter drones are also among the most efficient methods of delivery.
The number of stops per kilometer and the number of packages delivered per stop varies according to the
transportation mode and delivery density (highly dense areas are more likely to have more stops and packages
delivered per kilometer).
Similarly, an analysis of the greenhouse gas (GHG) emissions of the fuel of each transportation mode
shows that quadcopter drones are among the most efficient vehicles in grams of CO2e per km (Figure 3a)
and a competitive alternative in terms of GHG emissions per package (Figure 3b). On the other hand, it is
important to note that small drones are considerably limited in terms of weight and volume of the packages
transported. Therefore, an analysis of the energy consumption and GHG emissions on a per metric ton-km
basis (Supplementary Figure S5) shows that small drones are the most energy-intensive vehicles. Also, local
airspace regulations that require longer delivery routes, not considered in this study, can potentially increase
the energy consumption and GHG emissions of drones [36]. Finally, alternative methods, such as the concept
of mobile warehouses [37] can be an effective alternative to incorporate drones and mitigate current limitations
by combining drones and delivery trucks.
Figure 3: Greenhouse gas (GHG) emissions for different transportation modes. Error bars represent uncertain-
ties due to variations on fuel carbon intensity, battery life cycle emissions, and number of packages delivered
per distance.
We compared our results to values provided by the United Parcel Service, Inc. (UPS). In 2019, UPS
reported the energy intensity for U.S. Domestic Package operations was 28 MJ/package, from which ground
vehicles represented approximately 9.5 MJ/Package or 34% (airline fuel, facility heating fuel and indirect energy
correspond to 60%, 3% and 3%, respectively), with GHG emissions (CO2e) intensity of 1 kg/package [38]. It is
important to note that these values encompass the entire ground fleet, rather than only last mile delivery.
We estimate a small quadcopter drone, with a payload of 1 kg operating at a cruise speed of 12 m/s and
cruise altitude of 100 m, consumes approximately 0.05 MJ/km and is generates 41 g of CO2e per package when
charged on average U.S. electricity. Our energy model has simple, generalizable, and accurate coefficients that
can provide stakeholders and researchers an energy consumption estimation for speed ranges below 12 m/s.
However, at greater speeds or using drones with more surface area, a more comprehensive energy profile method
could provide more accurate predictions.
The energy consumption of small quadcopter drones is comparable to the most energy efficient modes of
last-delivery when the total mass of delivery is not the main feature considered. For example, in delivery
situations where small and light items with high added value, such as small electronics and medicines, drones
might became a competitive tool to reduce transportation emissions in large urban centers [39]. In these
scenarios, we found that drones can reduce the energy consumption by 96% and 60% and GHG emissions by
91% and 59% per package delivered by replacing diesel trucks and electric vans, respectively. We also found
that the delivery intensity, i.e. the number of packages delivered per km, and the fuel carbon intensity are the
main factors contributing to the drone’s energy and environmental performances. It is also important to note
that the drone used to collect the data was not optimized to minimize energy consumption, which could further
improve the its efficiency.
3 Methods
3.1 Experiment
We performed a series of flights to empirically measure the energy consumption of a quadcopter UAV. An
experimental protocol was created and followed to ensure a reliable approach for data acquisition [33].
A DJI®Matrice 100 (M100) quadcopter was equipped with an anemometer, current and voltage monitor,
GPS, and accelerometer collecting data on wind speed and direction, battery current and voltage demand,
and position, orientation, velocity and acceleration. The flights were performed in a pre-established route
with varying altitude (25 m, 50 m, 75 m and 100 m), speed (4 m/s, 6 m/s, 8 m/s, 10 m/s and 12 m/s) and
payload mass (no payload, 250g and 500g). Each combination was repeated at least three times, totaling 187
flights. The data provided by each sensor were synchronized to a frequency of approximately 5Hz using the
ApproximateTime [40] message filter policy of Robot Operating System (ROS).
For a better understanding of the energy consumption profile of each flight we created an algorithm to
automatically divide the data into three different flight regimes: takeoff, cruise, and landing (see Supplementary
Figures S1 to S4).
3.2 First Principles Analysis
The energy required to power a UAV can be estimated using a first principle analysis based on helicopter
aerodynamics [41]. First, we defined the working coordinate frames for a quadcopter drone (see Supplementary
Figures S6 to S8). Then, the we assessed the power required to maintain the drone at a steady hover condition.
Finally, we expanded the power analysis to include other power demands.
The main power demand of a drone is in the form of induced power (Pi). The induced power represents the
power required to overcome the force of gravity in order to keep the aircraft in the air, and it can vary according
to the flight maneuver [41]. The most basic way to estimate Piis considering a hover condition without wind
(Figure 4).
Figure 4: Hover condition without wind.
In that case, the thrust (T) equals the only force acting on the drone, its weight (W=mg) [41], and Pican
be estimated as
Pi=T vi(3)
where viis the induced velocity.
During hover, vican be simplified as
2ρA (4)
where ρis the air density and Ais the total area covered by the all four propellers.
Combining Eq. 4 and 3
2ρA =(mg)1.5
2ρA (5)
where mis the total mass of the drone, gis the gravitational acceleration.
More details of the first principles analysis and an expanded first-principles energy model is available in the
supplementary information (see supplementary Figures S9 to S11).
3.3 Energy Model
Our energy model inquires how effectively Pican be used as an estimator for the energy consumed during a
package delivery flight. In such a case, the average power ( ¯
P) throughout the flight is modeled as a linear
regression of the induced power ¯
where b1and b0are the slope and intercept of the linear regression, respectively.
Eq. 6 is expanded to account for the sum of the three flight regimes (Figure 5) and the total energy
consumption (E) is estimated as
r∈R, l∈L
(b(r, l)
i+b(r, l)
0)t(r, l)(7)
for R={takeof f, cruise, landing }and L={loaded, unloaded}.
Figure 5: Linear regression of flights separated by flight regime.
3.4 Machine Learning Approach
However, evaluating if the model’s performance is good given the available measurements cannot be inferred
from its performance alone. Therefore, we compare the predictive power of the energy model to a flexible
nonlinear algorithm [42], XGBoost, available in the programming environment R. This boosted tree algorithm
prioritizes predictive power against interpretability, and it is appropriate for predictive performance given the
available features. If our energy model presents similar accuracy to XGBoost, it indicates that the parametric
and functional restrictions we have made for the energy model development are suitable.
We fitted a gradient boosted tree algorithm, XBGoost [42]. The algorithm was separately trained for each
flight regime with a quadratic loss function and for all regimes, we used 75% as a subsample ratio of both
features and observations for each tree. For hyperparameter tuning, we varied learning rate, maximum tree
depth, and regularization parameter γin a grid search approach. 5-fold CV was used for error estimation; for
tuning only, we compared Absolute Relative Error (ARE, Equation 8) instead of quadratic error. After tuning,
the model was trained with the optimal hyperparameters on the entire training set and AREs were computed
for the flights on the test set.
Standard errors To obtain standard errors of the estimated coefficients, we used a nonparametric bootstrap
approach [43]. 1000 bootstrap replications were used to resample with replacement the 120 training flights,
and the two energy models for the three flight regimes were refitted for each bootstrap sample. At the end,
standard error of the coefficients were obtained from their sampling distribution.
Model predictive power To evaluate the model’s predictive power, the regime-specific fitted models were
then applied to the testing flights of the test set and were compared by Absolute Relative Error (ARE), computed
at flight resolution. That is, for each flight from the test set, we computed their Emeasured integrating power
over time, and the Em
estimated as the sum of the integral of the estimated power over time for the three flight
regimes via method m:
ARE(m) =
Emeasured Em
for m∈ {Energy Model,XBGoost}.
3.5 Drone’s range
Our analysis also shows the impact of varying operational parameters (speed, altitude and payload) on the
range of the drone. The two-way drone range (d) can be calculated considering the cruise speed
E=X(b1Pi+b0)t=XEtakeof f +XElanding +X(b1Pi+b0)d
Expanding for a two-way trip and solving for d
d=hEmax El
takeof f +Eu
takeof f +El
landing +Eu
i+ 2b0
where, Emax is the energy available in the battery; Etakeof f and Elanding are the energy consumed during
takeoff and landing for delivery (loaded = l) and returning (unloaded = u), respectively; Piis the induced
power, calculated using Eq. 5, for delivery (loaded = l) and return (unloaded = u); b1and b0the coefficients
from Table 1 for cruise and Vcr is the average inertial cruise speed.
The energy during takeoff and landing can be calculated as
E= (b1Pi+b0)h
where h is the cruise altitude, V is the average speed during takeoff and landing.
For instance, a small quadcopter operating at Vcr = 12 m/s, payload = 1000 g, h = 100 m, takeoff average
speed (Vtk ) = 2.5 m/s, landing average speed Vld = 2.0 m/s has a range of approximately 11 km (5.7 km of
delivery range), consuming approximately 120 Wh (per round trip delivery). Therefore, a quadcopter drone
flying under these conditions would consume approximately 0.039 MJ/km, not considering charging and trans-
mission losses. The energy consumption during takeoff and landing for this trip corresponds to approximately
17% of the total energy consumption (19.4 Wh per trip). This share of energy could be reduced by 95% in a
5-m takeoff (from 19.4 to Y 0.97 Wh per trip), which could be achieved, for instance, if the drone would depart
from the top of a building. This would reduce total trip energy by 15% (120 to 102 Wh), or increase the range
from 5.7 to 6.7 km.
3.6 Transport mode comparison
We compare the small quadcopter drone to different transportation modes in terms of energy consumption and
CO2e emissions and validate it against top-down sustainability reports from the United Parcel Service, Inc.
(UPS)’s 2019 . The energy consumption of a medium-duty diesel truck is considered as 11 MJ/km [44]. Whereas
a Medium-duty electric truck has energy consumption of 1.4 kWh/mile[44], or 3.13 MJ/km. Diesel Vans operate
at 18.4 MPG on average[45], or 4.9 MJ/km, (conversion factors: 1 gallon of diesel = 137,381 Btu[46], 1 MJ =
947.817 Btu, 1 mile = 1.60934 km). On the other hand, electric vans operate with energy consumption of 0.38
kWh/km[47], or 1.36 MJ/km. Finally, an electric cargo bicycle operates at 0.023 kWh/km[48], or 0.08 MJ/km
(conversion factors: 1 kWh = 3.6 MJ). In addition, variations in driving style can vary energy consumption by
40%[49]. Based on our energy model, a small quadcopter drone consumes approximately 120 Wh in a 5.5 km
delivery distance (11 km total distance), or 0.039 MJ/km, when delivering at maximum capacity (1 kg payload
with unloaded return) and cruise speed of 12 m/s. Transmission losses of 6.5% and a charging efficiency of
88% [50, 51, 52, 53] were included to the energy consumption of the electric vehicles (Table 2). Supplementary
Table S1 summarizes the nominal energy consumption and also provides the payload capacity of each mode.
The electricity CO2e emissions were considered to be the 2019 American average of 182 g/MJ (656 g/kWh),
with the lower of 107 g/MJ (384 g/kWh) from New England and the upper limit of 249 g/MJ (896 g/kWh)
reflecting non-baseload emissions from the central Mid-West[52]. CO2e emissions for Diesel Fuel combustion
was considered as 1.61x104lb/Btu[54], or 69.35 g/MJ. Upstream GHG emissions for diesel and electricity
generation are 15 g/MJ and 22 g/MJ[55], respectively. The drone’s LiPO Battery life cycle emissions were
assumed to be similar to Li-iron phosphate 2 g/MJ (base case), 0.6 g/MJ (low case) and 4 g/MJ (high case)[55].
Similarly, the electric cargo bicycle has battery life cycle emissions of 5.1 g/MJ (base case), 1.1 g/MJ (low case)
and 16.9 g/MJ (high case), for the electric van 18.7 g/MJ (base case), 5.6 g/MJ (low case) and 37.4 g/MJ (high
case), and for the medium duty electric truck 32.5 g/MJ (base case), 9.7 g/MJ (low case) and 65 g/MJ (high
The battery life cycle emissions for the drone (assumed to be similar to Li-iron phosphate) was calculated
as 0.76 g/km (base case), 0.23 g/MJ (low case) and 1.52 g/MJ (high case). Similarly, the electric cargo bicycle
has battery life cycle emissions of 1.3 g/km, considering a Li-ion NMC811 battery. For the electric van and
electric medium duty truck we assumed a battery of Li-ion NMC811, resulting in 14.1 g/km for the van and
24.5 g/km for the truck.[55]
The energy consumption per package (Epack) was calculated as
Epack =Edist
Sfreq ·Pfreq
where Edist is the energy consumption per distance unit, Sf req is the number of stops to delivery packages per
distance unit, and Pfreq is number of packages delivered per stop on average.
Similary, the greenhouse gas emissions per package (GHGpackage) is calculated as
GHGpack =Edist ·GHGenergy
Sfreq ·Pfreq
where GHGenergy is the mass of CO2e per energy unit.
Table 2 summarizes the values calculated per vehicle.
Table 2: Base-case energy consumption and GHG emissions for different vehicles.
Vehicle Class Energy Consumption
Fuel GHG
emissions [g/km]
Upstream GHG
emissions [g/km]
Battery GHG
emissions [g/km]
Energy consumption
GHG emission
Medium duty truck 11.00 764.5 168.7 5.24 444.4
Small diesel van 4.90 340.6 75.2 1.41 119.5
Medium duty electric truck 3.74 681.4 82.4 24.5 1.78 375.4
Small electric van 1.63 296.1 35.8 14.1 0.47 99.4
Electric cargo bicycle 0.10 18.1 2.2 1.3 0.10 21.6
Quad-copter drone 0.05 8.5 1.0 0.8 0.19 41.1
This work was supported by the U.S. Department of Energy’s Vehicle Technologies Office, Award Number DE-
EE0008463. This article was prepared while C.S. was employed at Carnegie Mellon University, and is currently
on public service leave. The opinions expressed in this article are the authors’ own and do not reflect the view
of the United States government or any other organization.
Data Availability
All drone data is available is at and the all the modeling code is available
at consumption
Author Contributions Statement
The research project was conceived by C.S, H.M. and S.S. The drone data collection experiment was designed
by J.P. The energy model was developed by T.R. The results were analyzed by T.R. and N.O. All authors
contributed to the writing of the manuscript.
Declaration of Interests
The authors declare no competing interests.
Supplementary Information
Supplementary Information is available with this manuscript.
[1] Stacy Davis and Robert Boundy. Transportation Energy Data Book: Edition 39. Tech. rep. Oak Ridge,
TN (United States): Oak Ridge National Laboratory (ORNL), Feb. 2021. doi:10.2172/1767864.
[2] Florian Dost and Erik Maier. “E-commerce effects on energy consumption: A multi-year ecosystem-level
assessment”. In: Journal of Industrial Ecology (2018). issn: 15309290. doi:10.1111/jiec.12639.
[3] Christopher Mims. The Scramble for Delivery Robots Is On and Startups Can Barely Keep Up. 2020.
[4] Timothy B. Lee. The pandemic is bringing us closer to our robot takeout future. Apr. 2020.
[5] Agnivesh Pani et al. “Evaluating public acceptance of autonomous delivery robots during COVID-19
pandemic”. In: Transportation Research Part D: Transport and Environment 89 (Dec. 2020), p. 102600.
issn: 13619209. doi:10.1016/j.trd.2020.102600.
[6] Sebastian Kapser and Mahmoud Abdelrahman. “Acceptance of autonomous delivery vehicles for last-mile
delivery in Germany – Extending UTAUT2 with risk perceptions”. In: Transportation Research Part C:
Emerging Technologies 111 (Feb. 2020), pp. 210–225. issn: 0968090X. doi:10.1016/j.trc.2019.12.016.
[7] Murad Mehrab Abrar, Raian Islam, and Md Abid Hasan Shanto. “An Autonomous Delivery Robot to
Prevent the Spread of Coronavirus in Product Delivery System”. In: 2020 11th IEEE Annual Ubiquitous
Computing, Electronics and Mobile Communication Conference, UEMCON 2020. Institute of Electri-
cal and Electronics Engineers Inc., Oct. 2020, pp. 0461–0466. isbn: 9781728196565. doi:10 . 1109 /
[8] Martin Joerss, Florian Neuhaus, and J¨urgen Schr¨oder. How customer demands are reshaping last-mile
delivery — McKinsey. 2016.
[9] Greg Bensinger and Laura Stevens. Amazon’s Newest Ambition: Competing Directly With UPS and FedEx.
[10] Doug Gross. Amazon’s drone delivery: How would it work? - CNN. 2013.
[11] Mike Murphy. This is how Google wants its drones to deliver stuff to you. 2016.
[12] UPS. UPS Tests Residential Delivery Via Drone Launched From atop Package Car. 2017.
[13] Manali Sheth et al. “Measuring delivery route cost trade-offs between electric-assist cargo bicycles and
delivery trucks in dense urban areas”. In: European Transport Research Review 11.1 (Dec. 2019), pp. 1–12.
issn: 18668887. doi:10.1186/s12544-019-0349-5.
[14] F. N. McLeod et al. “Quantifying environmental and financial benefits of using porters and cycle couriers
for last-mile parcel delivery”. In: Transportation Research Part D: Transport and Environment 82 (May
2020), p. 102311. issn: 13619209. doi:10.1016/j.trd.2020.102311.
[15] Robert Gebeloff and Karl Russel. How the Growth of E-Commerce Is Shifting Retail Jobs. 2017.
[16] D. F. Dominkovi´c et al. The future of transportation in sustainable energy systems: Opportunities and
barriers in a clean energy transition. Feb. 2018. doi:10.1016/j.rser.2017.06.117.
[17] Juan Zhang et al. “Energy consumption models for delivery drones: A comparison and assessment”. In:
Transportation Research Part D: Transport and Environment 90 (Jan. 2021), p. 102668. issn: 13619209.
[18] Raffaello D’Andrea. “Guest editorial can drones deliver?” In: IEEE Transactions on Automation Science
and Engineering 11.3 (2014), pp. 647–648. issn: 15455955. doi:10.1109/TASE.2014.2326952.
[19] Miguel A. Figliozzi. “Lifecycle modeling and assessment of unmanned aerial vehicles (Drones) CO2e
emissions”. In: Transportation Research Part D: Transport and Environment 57 (Dec. 2017), pp. 251–
261. issn: 13619209. doi:10.1016/j.trd.2017.09.011.
[20] Kevin Dorling et al. “Vehicle Routing Problems for Drone Delivery”. In: IEEE Transactions on Systems,
Man, and Cybernetics: Systems 47.1 (Jan. 2016), pp. 70–85. issn: 10834427. doi:10.1109/TSMC.2016.
[21] Zhilong Liu, Raja Sengupta, and Alex Kurzhanskiy. “A power consumption model for multi-rotor small
unmanned aircraft systems”. In: 2017 International Conference on Unmanned Aircraft Systems (ICUAS).
IEEE, June 2017, pp. 310–315. isbn: 978-1-5090-4495-5. doi:10.1109/ICUAS.2017.7991310.
[22] Andrew J Lohn et al. The Drones Are Here What’s the Buzz? Tech. rep. 2018.
[23] Joshuah K. Stolaroff et al. “Energy use and life cycle greenhouse gas emissions of drones for commercial
package delivery”. In: Nature Communications (2018). issn: 20411723. doi:10 . 1038 / s41467 - 017 -
[24] Ho Young Jeong, Byung Duk Song, and Seokcheon Lee. “Truck-drone hybrid delivery routing: Payload-
energy dependency and No-Fly zones”. In: International Journal of Production Economics 214 (Aug.
2019), pp. 220–233. issn: 09255273. doi:10.1016/j.ijpe.2019.01.010.
[25] Maryam Torabbeigi, Gino J. Lim, and Seon Jin Kim. “Drone Delivery Scheduling Optimization Consider-
ing Payload-induced Battery Consumption Rates”. In: Journal of Intelligent and Robotic Systems: Theory
and Applications 97.3-4 (Mar. 2019), pp. 471–487. issn: 15730409. doi:10.1007/s10846-019-01034-w.
[26] Thomas Kirschstein. “Comparison of energy demands of drone-based and ground-based parcel delivery
services”. In: Transportation Research Part D: Transport and Environment 78 (Jan. 2020), p. 102209.
issn: 13619209. doi:10.1016/j.trd.2019.102209.
[27] Chien-Ming Tseng et al. “Autonomous Recharging and Flight Mission Planning for Battery-operated
Autonomous Drones”. In: 25 (Mar. 2017).
[28] Chien-Ming Tseng et al. Flight Tour Planning with Recharging Optimization for Battery-operated Au-
tonomous Drones. Tech. rep. 2017.
[29] Miguel A. Figliozzi. “Carbon emissions reductions in last mile and grocery deliveries utilizing air and
ground autonomous vehicles”. In: Transportation Research Part D: Transport and Environment 85 (Aug.
2020), p. 102443. issn: 13619209. doi:10.1016/j.trd.2020.102443.
[30] Anne Goodchild and Jordan Toy. “Delivery by drone: An evaluation of unmanned aerial vehicle technology
in reducing CO2 emissions in the delivery service industry”. In: Transportation Research Part D: Transport
and Environment 61 (June 2018), pp. 58–67. issn: 13619209. doi:10.1016/j.trd.2017.02.017.
[31] Brennan Borlaug et al. “Heavy-duty truck electrification and the impacts of depot charging on electricity
distribution systems”. In: Nature Energy 6.6 (June 2021), pp. 673–682. issn: 2058-7546. doi:10.1038/
[32] Thiago A. Rodrigues et al. Data Collected with Package Delivery Quadcopter Drone. figshare. July 2020.
[33] Thiago A. Rodrigues et al. “In-flight positional and energy use data set of a DJI Matrice 100 quadcopter
for small package delivery”. In: Scientific Data 8.1 (Mar. 2021), pp. 1–8. doi:10 .1038 /s41597 - 021 -
[34] DJI. MATRICE 100 SPECS. Available at
[35] Connie Guglielmo. Turns Out Amazon, Touting Drone Delivery, Does Sell Lots of Products That Weigh
Less Than 5 Pounds. Dec. 2013.
[36] Mo Elsayed and Moataz Mohamed. “The impact of airspace regulations on unmanned aerial vehicles in
last-mile operation”. In: Transportation Research Part D: Transport and Environment 87 (Oct. 2020),
p. 102480. issn: 13619209. doi:10.1016/j.trd.2020.102480.
[37] S. Srivatsa Srinivas and Rahul R Marathe. “Moving towards “mobile warehouse”: Last-mile logistics
during COVID-19 and beyond”. In: Transportation Research Interdisciplinary Perspectives (Mar. 2021),
p. 100339. issn: 25901982. doi:10.1016/j.trip.2021.100339.
[38] UPS. 2019 GRI Content Index. Tech. rep. Atlanta, 2019.
[39] Mine Isik, Rebecca Dodder, and P. Ozge Kaplan. “Transportation emissions scenarios for New York City
under different carbon intensities of electricity and electric vehicle adoption rates”. In: Nature Energy 6.1
(Jan. 2021), pp. 92–104. issn: 2058-7546. doi:10.1038/s41560-020-00740-2.
[40] Open Source Robotics Foundation ApproximateTime. 2010.
[41] Constantin Rotaru and Michael Todorov. “Helicopter Flight Physics”. In: Flight Physics - Models, Tech-
niques and Technologies. InTech, Feb. 2017. doi:10.5772/intechopen.71516.
[42] Tianqi Chen et al. xgboost: Extreme Gradient Boosting. 2020.
[43] Bradley Efron and Robert J Tibshirani. An Introduction to the Bootstrap. Monographs on Statistics and
Applied Probability 57. Boca Raton, Florida, USA: Chapman & Hall/CRC, 1993.
[44] Robert Prohaska et al. Field Evaluation of Medium-Duty Plug-in Electric Delivery Trucks. Tech. rep.
Golden, CO (United States): National Renewable Energy Laboratory (NREL), Dec. 2016. doi:10.2172/
[45] Advantage Outfitters. Best Cargo Vans for Small Businesses: Gas Mileage, Cargo Space and More. July
[46] U.S. Energy Information Administration (EIA). Energy units and calculators explained. 2020.
[47] Chiara Fiori and Vittorio Marzano. “Modelling energy consumption of electric freight vehicles in urban
pickup/delivery operations: analysis and estimation on a real-world dataset”. In: Transportation Research
Part D: Transport and Environment 65 (Dec. 2018), pp. 658–673. issn: 13619209. doi:10.1016/j.trd.
[48] Johan Erlandsson. The Armadillo electric cargo bike use 6 % of the electricity of a small electric van.
Available at https: // www. velove. se /news /armadillo - cargo - bike- use - 6- electricity - small-
electric-van. 2017.
[49] U.S. Department of Energy. Many Factors Affect MPG. Available at
[50] Argonne GREET Publication. Cradle-to-Grave Lifecycle Analysis of U.S. Light-Duty Vehicle-Fuel Path-
ways: A Greenhouse Gas Emissions and Economic Assessment of Current (2015) and Future (2025-2030)
Technologies. Tech. rep. 2016.
[51] Ching Shin Norman Shiau et al. “Impact of battery weight and charging patterns on the economic and
environmental benefits of plug-in hybrid vehicles”. In: Energy Policy 37.7 (July 2009), pp. 2653–2663.
issn: 03014215. doi:10.1016/j.enpol.2009.02.040.
[52] US EPA. Emissions & Generation Resource Integrated Database (eGRID). 2019.
[53] U.S. Energy Information Administration (EIA). How much electricity is lost in electricity transmission
and distribution in the United States? May 2021.
[54] U.S. Energy Information Administration (EIA). How much carbon dioxide is produced when different fuels
are burned? 2020.
[55] Argonne National Laboratory. GREET. Lemont, Illinois, USA, 2020.
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Major technological advancements and recent policy support are improving the outlook for heavy-duty truck electrification in the United States. In particular, short-haul operations (≤200 miles (≤322 km)) are prevalent and early candidates for plug-in electric vehicles (EVs) given their short, predictable routes and return-to-base applications, which allows vehicles to recharge when off shift at their depots. Although previous studies investigated the impacts of added electrical loads on distribution systems, which included light-duty EVs, the implications for heavy-duty EV charging are underexplored. Here we summarize the causes, costs and lead times of distribution system upgrades anticipated for depot charging. We also developed synthetic depot charging load profiles for heavy-duty trucks from real-world operating schedules, and found that charging requirements are met at common light-duty EV charging rates (≤100 kW per vehicle). Finally, we applied depot charging load profiles to 36 distribution real-world substations, which showed that most can accommodate high levels of heavy-duty EV charging without upgrades. Increasing attention is being paid to the electrification of trucks, in particular for short-haul operations. Borlaug et al. simulate depot charging load profiles based on real-world operating schedules to explore future charging requirements and assess the likely distribution substation upgrades needed to support them.
Full-text available
Like many cities around the world, New York City is establishing policies to reduce CO2 emissions from all energy sectors by 2050. Understanding the impact of varying degrees of electric vehicle adoption and CO2 intensities on emissions reduction in the city is critical. Here, using a technology-rich, bottom-up, energy system optimization model, we analyse the cost and air emissions impacts of New York City’s proposed CO2 reduction policies for the transportation sector through a scenario framework. Our analysis reveals that the electrification of light-duty vehicles at earlier periods is essential for deeper reductions in air emissions. When further combined with energy efficiency improvements, these actions contribute to CO2 reductions under the scenarios of more CO2-intense electricity. Substantial reliance on fossil fuels and a need for structural change pose challenges to cost-effective CO2 reductions in the transportation sector. Here we find that uncertainties associated with decarbonization of the electric grid have a minimum influence on the cost-effectiveness of CO2 reduction pathways for the transportation sector.
Full-text available
Utilizing autonomous unmanned aerial vehicles (drones) in the last-mile delivery of parcels is regarded as the ultimate disruptive technology that might significantly reduce the GHG emissions in the freight sector. This study estimates the CO2e emissions for UAVs under different policies compared to diesel and electric ground delivery modes. First, the international UAV flight regulations are synthesized and classified into three groups representing varying degrees of policy strictness. Second, utilizing real-word delivery demand data, full-day parcel-delivery operations of a three-digit postal code area in both urban and rural contexts are simulated for UAVs and ground delivery modes. The results show that in general, UAVs produce significantly lower emissions compared to ground delivery per parcel-km and up to 35% compared to electric vehicles. However, UAV emissions are highly dependent on the fuel mix used in electricity generation. In urban contexts, UAV policy strictness can increase GHG emissions by up to 400%.
Supply chains in general and last-mile logistics in particular, have been disrupted due to COVID-19. Though several innovative last-mile logistics solutions have been proposed in the past, they possess certain limitations, especially during COVID-19 motivating the need for an alternative last-mile logistics solution. We present a review of literature related to last-mile logistics and supply chain disruptions to identify the limitations of existing last-mile delivery practices during COVID-19. Using a stylized analytical model, we then propose that “mobile warehouse” can be an effective solution to last-mile logistics issues faced during COVID-19 and beyond under certain conditions. A mobile warehouse is a truck dedicated to a particular geographical location and carries the inventory of various products based on the estimated demand requirements for these products in that geographical location. We provide the condition under which the B2C e-commerce providers find it profitable to adopt a truck as a mobile warehouse to sell high demand items quickly.
Energy consumption is a critical constraint for drone delivery operations to achieve their full potential of providing fast delivery, reducing cost, and cutting emissions. This paper provides a uniform framework to facilitate understanding different drone energy consumption models and the inter-relationships between key factors and performance measures to facilitate decision making for drone delivery operations. We review, classify and assess drone energy consumption models. We then document the very wide variations in the modeled energy consumption rates resulting from differences in: (1) the scopes and features of the models; (2) the specific designs of the drones; and (3) the details of their assumed operations and uses. The results show that great care must be taken in adopting a particular drone energy consumption model and that more research is needed, especially empirical research, to ensure the selected model accurately reflects delivery drone designs and uses.
Autonomous delivery robot (ADR) technology for last-mile freight deliveries is a valuable step towards low-carbon logistics. The ongoing COVID-19 pandemic has put a global spotlight on ADRs for contactless package deliveries, and tremendous market interest has been pushing ADR developers to provide large-scale operation in several US cities. The deployment and penetration of ADR technology in this emerging marketplace calls for collection and analysis of consumer preference data on ADRs. This study addresses the need for research on public acceptance of ADRs and offers a detailed analysis of consumer preferences, trust, attitudes, and willingness to pay (WTP) using a representative sample of 483 consumers in Portland. The results reveal six underlying consumer segments: Direct Shoppers, E-Shopping Lovers, COVID Converts, Omnichannel Consumers, E-Shopping Skeptics, and Indifferent Consumers. By identifying the WTP determinants of these latent classes, this study provides actionable guidance for fostering mass adoption of low-carbon deliveries in the last-mile.
The rapid growth of e-commerce and package deliveries across the globe is demanding new solutions to meet customers’ desire for more and faster deliveries. New driverless air and ground vehicles are being launched and tested to deliver products or services in the areas of retail, groceries, and healthcare. This research focuses on the efficiency of autonomous (driverless) air and ground delivery vehicles in terms of vehicle-miles, energy consumption, and CO2 emissions. Three types of autonomous vehicle are analyzed: drones or unmanned aerial vehicles (UAVs), sidewalk autonomous delivery robots (SADRs), and road autonomous delivery robots (RADRs). The CO2 emissions of these autonomous vehicles are compared against emissions from an electric van (e-van), a conventional internal combustion engine van, and driving to a store utilizing electric and conventional vehicles. The impacts of vehicle capacity, range, and time constraints are analyzed as well as the impacts of number of deliveries, service time, area of service, and depot-service area distance. Novel results are found regarding the efficiency of each vehicle type and tradeoffs between driving to a store and store delivery as a function of order size and type of vehicle driven by consumers.
Parcel carriers face increasingly difficult operating conditions in busy metropolitan areas due to growing consumer demand for ever faster delivery services and having to cope with traffic congestion and city authority measures that may restrict or penalise access for certain types of vehicle. This paper evaluates the potential environmental and financial benefits of switching from traditional van-based deliveries to an alternative operating model, where porters or cycle couriers undertake deliveries supported by a substantially reduced van fleet. Results using a specially-developed algorithm to model operations of a real carrier in an area of central London, UK, suggested that the carrier could reduce CO2 emissions by 45%, NOx emissions by 33%, driving distance by 78% and curbside parking time by 45%. Overall cost savings to the carrier were estimated to be in the range 34–39%. Scaling up the modelled emissions savings to London’s Central Activities Zone, an area of approximately 30 km² and with current total annual parcel delivery distance of around 15 million km, could see annual emissions savings in the region of 2 million kg CO2 and 1633 kg NOx if all carriers utilised porters or cycle couriers. The key operating challenges identified were related to sorting and consolidating items by weight and volume, parcel handover arrangements and how to deal with express items and failed deliveries.
The inevitable need to develop new delivery practices in last-mile logistics arises from the enormously growing business to consumer (B2C) e-commerce and the associated challenges for logistics service providers. Autonomous delivery vehicles (ADVs) are believed to have the potential to revolutionise last-mile delivery in a way that is more sustainable and customer focused. However, if not widely accepted, the introduction of ADVs as a delivery option can be a substantial waste of resources. At present, the research on consumers’ receptivity of innovations in last-mile delivery, such as ADVs, is limited. This study is the first that investigates the users’ acceptance of ADVs in Germany by utilising an extended Unified Theory of Acceptance and Use of Technology (UTAUT2) and adapted it to the context of ADVs in last-mile delivery. Quantitative data was collected through an online survey approach (n = 501) and structural equation modelling was undertaken. The results indicate that price sensitivity is the strongest predictor of behavioural intention (i.e., user acceptance), followed by performance expectancy, hedonic motivation, perceived risk, social influence and facilitating conditions, whereas no effect could be found for effort expectancy. These findings have important theoretical and practical contributions in the areas of technology acceptance and last-mile delivery.