ArticlePDF Available

UAV Payload Transportation via RTDP Based Optimized Velocity Profiles


Abstract and Figures

This paper explores the application of a real-time dynamic programming (RTDP) algorithm to transport a payload using a multi-rotor unmanned aerial vehicle (UAV) in order to optimize journey time and energy consumption. The RTDP algorithm is developed by discretizing the journey into distance interval horizons and applying the RTDP sweep to the current horizon to get the optimal velocity decision. RTDP sweep requires the current state of the UAV to generate the next best velocity decision. To the best of the authors knowledge, this is the first time that such real-time optimization algorithm is applied to multi-rotor based transportation. The algorithm was first tested in simulations and then experiments were performed. The results show the effectiveness and applicability of the proposed algorithm
Content may be subject to copyright.
UAV Payload Transportation via RTDP Based
Optimized Velocity Profiles
Abdullah Mohiuddin 1,*, Tarek Taha 2, Yahya Zweiri 1,3 and Dongming Gan 1
Khalifa University Center for Autonomous Robotic Systems, Khalifa University of Science and Technology,
P.O. Box 127788, Abu Dhabi, UAE
2Algorythma’s Autonomous Aerial Lab, P.O. Box 112230, Abu Dhabi, UAE
3Faculty of Science, Engineering and Computing, Kingston University London, London SW15 3DW, UK
*Correspondence:; Tel.: +971-2-501-8558
Received: 9 June 2019; Accepted: 12 July 2019; Published: 8 August 2019
This paper explores the application of a real-time dynamic programming (RTDP) algorithm
to transport a payload using a multi-rotor unmanned aerial vehicle (UAV) in order to optimize
journey time and energy consumption. The RTDP algorithm is developed by discretizing the journey
into distance interval horizons and applying the RTDP sweep to the current horizon to get the optimal
velocity decision. RTDP sweep requires the current state of the UAV to generate the next best velocity
decision. To the best of the authors knowledge, this is the first time that such real-time optimization
algorithm is applied to multi-rotor based transportation. The algorithm was first tested in simulations
and then experiments were performed. The results show the effectiveness and applicability of the
proposed algorithm.
Keywords: UAV; energy optimization; dynamic programming; aerial transportation
1. Introduction
Unmanned aerial vehicles (UAVs) have been widely used by researchers, security and law
enforcement agencies, search and rescue operators, firefighters, farmers, filmmakers, photographers
and delivery companies. UAVs can perform tasks in confined spaces or in hazardous environments.
UAVs are now evolving from just being a sensor to air-borne manipulators. Researchers have started
to focus on attaching manipulators to the UAVs, which enables them to perform tasks such as opening
a valve, and picking and placing objects. Other than manipulation, researchers claim that UAVs,
specifically small UAVs, have the potential to significantly improve the research in remote sensing [
Recently, a Canadian firm has started drone based payload delivery services [2]; another firm started
coffee delivery via drones in Australia [
]. Recently, a drone was tested for its ability to transport
organs for organ transplants [
]. UAVs are used to transport deform-able linear objects such as hose
goods [5,6]
. The Swiss post used UAVs to transport medical samples, which helped them to
reduce a 45 min car journey to a few minutes of flight [
]. A recent study showed that UAVs delivery
might even help with reducing greenhouse emissions caused by the freight industry [7].
Motivated by the potential of UAV based payload transportation, we have investigated the
possibility of applying a real-time multi-criteria optimization approach. Continuous improvement
of the computational capabilities and reduction of the size of computational platforms have allowed
for installing them on-board small UAVs. Therefore, dynamic programming based optimization
techniques can now be applied for controlling UAVs. Specifically, the contribution of this
paper is the development, numerical and hardware testing of a real-time dynamic programming
algorithm for achieving velocity optimized energy efficient aerial transportation using a multi-rotor
platform. The optimization algorithm can be applied in real-time scenarios, and it considers the
aerodynamic influences.
Energies 2019,12, 3049; doi:10.3390/en12163049
Energies 2019,12, 3049 2 of 25
Relevant Work
Energy consumption in multi-rotors is critical since they are powered by batteries that have limited
endurance. The major power consumption results from the motors that are rotating the propellers to
generate thrust that keeps the UAV airborne. The electrical energy consumed by the motors depends
on the thrust requirements, and also includes the electrical losses due to heat and friction and overall
propulsion system efficiency. These include losses of motors and electronic speed controllers. Although
several studies have focused on improving battery charging methods for multi-rotors such as wireless
charging [
], still, on average, the flight time of multi-rotors after every charge is around 20 to
30 min [
], which limits the applications of the multi-rotors. Other causes of energy consumption
include the autopilot or any companion computer attached to the aerial platform, sensors such as
a camera for visual servoing or communication links. The lateral motion of hex-rotors also deals
with parasitic drags, which results in a higher requirement of rotor torque that results in more energy
consumption. The energy constraints in aerial transportation can be addressed via two approaches.
One approach is the design stage of the aerial transportation. The second stage is the energy savings by
the efficient planning of the operation. Energy savings in the design stage can be achieved in various
ways, including reducing the amount of weight carried by the UAV. Batteries account for up to 50%
of total UAV mass for small UAVs [
], the addition of further payload mass drastically affects the
already burdened energy budget. Flying with an optimum mass can maximize the endurance of the
]. In some cases, an aerial platform could utilize a hybrid design such as [
] that travels on
the ground when the flight is not necessary to save energy.
The second energy saving approach is related to efficient motion planning. It is possible in these
cases to choose a minimum energy consumption path from multiple available paths, generated by a
path planner [
] for a single UAV. The method developed by [
] performs an offline energy efficient
path planning based on Dijkstra’s Algorithm.
A paper by [
] showed that energy consumption increases with forward velocity after a first
decrease, when compared to the energy consumption in a hover condition. Another study by [
evaluated the direct relation between energy efficiency and the speed of a multi-rotor system. It was
argued and proved by [
] that a hex-rotors system have to consume energy to continuously support
their weight and minimizing the time in the air could result in overall energy savings. Experiments
showed that there was a 29% difference between the hex-rotor transportation with two different speeds.
A novel path-following controller was presented by [
] in which the speed of the rotor-craft is a
dynamic profile that varies with the geometric requirements of the desired path.
Minimizing time in the air to save energy is significant in case of moderate speeds up to
10 m/s
where parasitic drag can be ignored [
]. However, for higher speeds and for bigger hex-rotors, this
parasitic drag can be significant and would lead to more energy consumption. In some cases, a cage
surrounds the multi-rotor, which also adds to the parasitic drag [
]. It can be inferred from previous
studies that energy efficiency can also be increased by incorporating the aerodynamic factors and
making velocity decisions that result in less overall energy consumption. It is also possible to generate
a minimum energy consumption path for a single UAV as performed by [
]. The proposed method
in [
] takes the brush-less DC motor model into account and finds the optimal path by using a
predefined initial and final configuration of the multi-rotor. However, the method proposed by [
was performed offline and it does not take into account the uncertainties present between actuation
and expected motion. Furthermore, a lack of feedback in the proposed strategy makes it unfeasible for
real-time multi-rotor transportation.
This research study presents an optimization algorithm for a platform’s velocity based on a
real-time dynamic programming algorithm (RTDP). The RTDP algorithm is developed utilizing the
original dynamic programming principle presented by [
]. The developed optimization algorithm is
inspired by a real-time dynamic programming introduced in [
] for the optimization of velocity to
minimize energy consumption in automobiles. To the best of the authors’ knowledge, RTDP based
energy optimization for multi-rotors energy savings has not been attempted before.
Energies 2019,12, 3049 3 of 25
In Section 2, we will describe the fundamentals of the RTDP algorithm used for energy
optimization. After that, Section 3will discuss the experimental methodology used in this study,
and results of the numerical experiments will be presented and discussed in Section 3.7, followed by
software in the loop simulation results in Section 4. Finally, hardware experiments are also discussed
in Section 5followed by conclusions and references.
2. Algorithm Description
The optimization algorithm is based on minimization of a cost function as described in
Equations (1) and (2). The optimization algorithm is applied in two steps as shown in Figure 1.
The first step is applied on a pre-defined horizon length, while not considering the end of the journey.
The cost function for first step with a non-finite stage is described in Equation (1):
VJN=g(xN) +
Jk(Je,Jt,λ), (1)
is the cumulative cost at the last stage,
is the reference velocity of the UAV and decision
is the cost at a particular time step
which is a function of the decision variable
is the energy cost of a particular step
, and is further explained in Section 2.1,
is the journey time cost of the step
is the weighing factor as described further in Section 2.1;
is the terminal cost. The terminal cost is used to approximate the cost incurred after the horizon
up to the end of the journey at an unknown distance, required to minimize the impact on future
horizons of decisions made for the current horizon.
Figure 1. Process flow diagram; first, the multi-rotor takes off and reaches the transportation altitude,
the optimization algorithm then sweeps through search-space in the rolling horizon to find an optimum
policy. The multi-rotor moves with this optimum policy for this distance step. This repeats for every
distance step, until the final horizon is activated. During the final horizon, at each computation, the
number of stages is reduced and the final state is given.
The second step is when the aerial vehicle enters the last horizon; during this step, the end of
the journey is considered. The main difference between the first step and the second step is that,
in the former,
the cost function includes a terminal cost, while the second step defines a final state of
the system. The second step of the optimization is activated when the vehicle enters the last horizon.
Energies 2019,12, 3049 4 of 25
This step is called the Dynamic programming (DP) algorithm with the finite stage. In this step, the
DP algorithm is applied at the start of each stage, but the number of stages is reduced for each DP
computation, based on the proximity of the vehicle to the goal position. The final state is predefined
and, in this case, it is considered to be a velocity state equal to zero. Equation (2) describes the second
step of the optimization algorithm:
Jk(Je,Jt,λ). (2)
As described in [
], apart from the minimization of cumulative cost function, we also need to
describe the system in a discrete time model. The overall model is discretized such that it can be
written as Equation (3):
xk+1=fk(xk,uk,zk), (3)
0, 1, 2, 3.....
1 is the time index showing the stages. Here,
represents the current
state of the system and
is the transition relationship, as a function of the current state
, the
control decisions which are defined further below. xkis the state vector defined as
xk=hVi, (4)
where Vis the velocity of the payload. ukis the decision variable defined as
uk=hV1i, (5)
is the reference velocity of the UAVs in the transport direction. As a simple case, this
optimization is performed for the transportation from point A to B having the same altitude.
The disturbance can be defined as zk
zk=hwind ρi, (6)
which is based on wind speed and ρis the air density at the transportation altitude.
In order to apply dynamic programming in real time, the problem is divided into horizons. If the
DP sweep is performed on the complete journey from point A to point B, then the computational costs
would increase, rendering the method to be incapable of working in real time. The application of DP
sweep on the horizon length reduced the computational time, therefore the DP sweep is performed
frequently. The DP sweep is performed every time the vehicle moves a predefined distance step.
Each horizon includes a pre-defined set of distance intervals. The current horizon is the one that
includes the current state of the agent. The application of the DP algorithm can be seen in Figure 1.
It can
be seen in Figure 1that there are two types of DP sweeps performed: one type of DP sweep with
terminal costs and the other type without terminal costs. The second step of DP sweep is required
because the final state of the UAV is in hovering mode in order to drop the payload at goal position.
During the start of the journey, the DP sweep with terminal costs is performed as follows.
The transition costs are calculated for all stages of current horizon only. Starting with the origin or
taking the current state of the agent, the transition costs of all possible states will be calculated and
stored for the first stage. Now, for the second stage, at each node, the cumulative transition costs of
reaching this node from all states of the first stage will be calculated. This process will be repeated
0 to
1 until the terminal node. After reaching the terminal node of the current
horizon, the transition cost to the terminal node will include the terminal cost as shown in Figure 1
and explained in Equation (1). Starting from the terminal node
with minimum cost, the optimal
solution will be traced back to node
0. After obtaining the optimal policies, control actions will be
immediately utilized until the agent reaches the next stage. When the agent reaches the next stage, we
will take the current state of the agent and consider it as the origin and run the algorithm again until
the end of the rolling horizon. After reaching the terminal node of the current horizon, we will again
Energies 2019,12, 3049 5 of 25
trace back the solution and immediately implement the optimal policies. The process will be repeated
until the vehicle enters the last horizon.
When the vehicle enters the last horizon, the DP sweep is performed following the cost function
Equation (2) and as shown in the end part of the journey, as shown in Figure 1. In the last horizon,
cumulative transition costs will be calculated while considering the final state, which is pre-defined.
Hence, the terminal cost will not be calculated. After each stage, the number of stages will be reduced,
until the vehicle reaches its destination.
2.1. Definition of Cost Function
The cost function is defined based on two parameters: energy consumption and time spent.
A parameter
is defined as the weighting factor to select the preference between the time spent to
reach the goal or the energy consumption to reach the goal. If
is close to one, the energy consumption
is given a priority and, if
is close to zero, then minimization of time spent is of higher priority.
In order to make these factors comparable to each other, a normalization is required. The combined
cost function can be written as follows:
Jk=Jeλ+Jt(1λ), (7)
where Jeand Jtare the normalized energy and journey time cost functions, respectively.
2.1.1. Normalization of Time Spent
The time spent to reach the goal position can be optimized by minimizing the following cost
function at each backup operation, when the algorithm is going through the options within the search
space. In order to make it compatible with the other part of the cost function, the time cost function
is normalized and made dimensionless using
, where
is the time spent by following an
option from search space and
is the normalization factor to make it compatible with the power
consumption factor in the combined cost function. This normalization factor is carefully selected to
make sure that the normalized time spent is less than one. This allows for comparison between the
energy and time. This dimensionless component of the cost function reduces the time spent to reach a
goal position because of the cost function’s minimization.
2.1.2. Normalization of Energy Consumption
Using the weighting factor, the cost function for the energy can be written as
, where
the energy consumed during the transition between two states, whereas
is the normalization factor
used to make it compatible with the time factor. The factor
is selected such that the normalized
energy consumption factor turns out to be less than one.
2.1.3. Terminal Cost
The terminal cost is calculated during first step of the optimization. The terminal cost is calculated
during DP computation; when the cost function of the last node is calculated, the terminal cost is
considered a multiple of the last transition cost to reach the last node. Since a DP sweep is repeatedly
performed at each distance step for the length of the horizon, the terminal cost penalizes the decisions
that can benefit the current horizon at the cost of the future horizons. An example of such is the
reduction of velocity significantly at the end of current horizon, which would result in high acceleration
in the beginning of the next horizon.
2.2. Model of the Evaluated System
The system consists of a hex-rotor as shown in Figure 2. The UAV model was simplified by
assuming that the UAV structure is rigid and CoG (center of gravity) and geometric centre is at origin
of the body frame and coincide. Let
P= [x
be the position vector of the center of mass of
Energies 2019,12, 3049 6 of 25
the hex-rotor relative to the fixed inertial frame
ε= [X0
. The orientation of the hex-rotor is
expressed in Euler angles as
Φ= [φ
, where
is the roll angle about the
is the pitch
angle about the
-axis, and
is the yaw angle about the
-axis of the hex-rotor UAV. Six rotors attached
to identical brush-less DC motors are rotating with a speed
. The following equations best describe
the translational and rotational dynamics model used in this paper for the hex-rotor UAVs [22]:
x=(sin φsin ψ+cos φcos ψsin θ)T
+FD(Vx+Vw x(z),θ,ρ(z)),(8)
y=(cos φsin θsin ψcos ψsin φ)T
z=(cos θcos φ)Tmg +FD(Vz+Vwy(z),φ,ρ(z)), (10)
φ= (IyIz)˙
θu4, (11)
θ= (IzIx)˙
φu4, (12)
ψ= (IxIy)˙
θ+u3, (13)
, which is sum of mass of the platform
and mass of the payload
the gravitational acceleration and
are the translational accelerations of the hex-rotor UAV
- and
is the total thrust produced by all motors calculated as
are the torque and thrust coefficients,
u4= (ω1ω2+ω3ω4+ω5ω6). The rotational inertia of the UAV is expressed as (Ix,Iy,Iz).
Figure 2. Multi-rotor schematic, with body frame and inertial frame.
The model detailed above is suitable for evaluating the optimization algorithm as shown in
Section 3, but it cannot be used for cost function calculation because of the high computational costs.
If we ignore complex maneuvers and only point A to point B transportation, we can then ignore
the energy consumption variation during attitude changes. Therefore, only the Equations (8)–(10)
describing the translational dynamics of the multi-rotor are used to develop a reduced model.
The optimization algorithm developed considers the transportation from point
to point
changing altitude or yaw. Keeping yaw angle fixed, and assuming a constant altitude allows the cost
function calculations to be simple; further studies, however, will be done without these assumptions.
Therefore, in this case of no vertical motion,
0. The total thrust required to stay airborne
would be
(cos θcos φ)
. If the yaw angle is fixed to zero and
from the above equation is substituted
in equations of motion, the multi-rotor equations of motion can be reduced to Equations (14) and (15).
x=m·tan θ·g+FD(Vx+Vwx (z),θ,ρ(z)), (14)
Energies 2019,12, 3049 7 of 25
y=m·tan φ
cos θ·g+FD(Vy+Vwy(z),φ,ρ(z)). (15)
Equations (14) and (15) can be used to find the roll and pitch angles that can then be used to find
the required thrust. The equations must be discretized in order to calculate the thrust required, which
in turn is needed to calculate the power consumption as described in Section 2.2.2. A step by step
description of power consumption calculation is shown in Algorithm 1
Algorithm 1: Calculation of average power consumption for a state transition
Result: Power consumption
Input : Vxi+1,Vxi,Vwx,x,m,
Output : Es
while While (θiθ)<0.01 do
Calculate FD(Vx+Vwx ,θ))
Calculate θifrom Equation (18)
Calculate T=mg
(cos θicos φ)
Calculate Pelec
2.2.1. Discretized Model
Since the stage corresponds to the distance travelled by the agent, it is necessary to convert the
translational dynamics equations from time domain to the distance domain. Chain rule can be used to
perform such conversion. The chain rule
was applied to obtain the translational dynamics
equation as follows. In order to reflect the change from time domain to distance domain, the index k in
the cost function definition is replaced by index s in Equations (1)–(6):
dx =m·tan θs·g+FD(Vx+Vwx,θ), (16)
dy =m·tan φs
cos θs
·g+FD(Vy+Vwy,φ). (17)
The distance based translational dynamics model can be discretized using the forward or
backward Euler method. The equations for the forward Euler approach are described as below:
·(m·tan θs·g+FD(Vx+Vwx ,θ)), (18)
·(m·tan φs
cos θs
·g+FD(Vy+Vwy,φ)). (19)
2.2.2. Motor Energy Consumption
A thrust measurement stand is used to obtain the rotor speed, thrust and energy consumption
plots for the DJI E310 propulsion system, which includes 2312 motors with the DJI 9450 rotors.
All values obtained were compared with the values given by [
] for the same propulsion system.
A power law based empirical equation between the rotor speed and the power consumption is obtained
using curve fitting as shown in Equation (20):
N, (20)
Energies 2019,12, 3049 8 of 25
is the power consumed by rotor
of the UAV, and
is the speed of the rotor in rad/s.
The speed of the rotors is found using the relation
, where
. The energy
consumption during a particular step can be calculated by
Es=ZPelec dt, (21)
which can be written in the discretized form as follows:
Es=Pelec t, (22)
is the time spent in that distance interval,
Pelec =N
is the energy consumed during
that distance interval.
2.2.3. Calculation of Step Travel Time
The time spent for that particular distance step can be calculated using the following equation:
t=Zs f
vds. (23)
Equation (23) is continuous and we need to discretize this equation in order to evaluate the time
consumed in that particular step:
s. (24)
2.3. Parameter Selection
There are several parameters that have to be determined before executing the optimization
algorithm. These parameters include: the horizon length; the distance interval; and the velocity
interval. These parameters can influence the complexity of the algorithm and the computational
time for one DP search space sweep. The complexity of the algorithm
as shown in Equation (25),
and calculated and plotted in Figure 3a, depends on the number of distance intervals
between the
start and goal position, and the number of velocity intervals Nv:
v). (25)
The complexity of the algorithm increases by increasing the number of velocity and altitude
intervals. The selection of distance size and the velocity interval also depends on the hex-rotor‘s ability
to reach that velocity in a particular distance step.
0.5 1 1.5 2
Velocity interval [m]
Computations O [1]
0.5 1 1.5 2
Velocity interval [m]
Computation time [sec]
Figure 3.
The effects of a decrease in velocity interval that results in an exponential increase in
(a) number of computations Oand (b) computation time.
Energies 2019,12, 3049 9 of 25
Sample computation times and number of computations were plotted in Figure 3. The analysis
in Figure 3is based on distance interval of 10 m and velocity interval ranging from 0.25 m/s to
2 m/s. The computation is done using a 2.4 GHz computer with 8 GB RAM. An open-source autopilot
hardware “pixhawk” is used with “PX4” firmware. The position controller in this system works at a
rate of 100 Hz, which means it takes 0.01 s for one position control cycle. Currently, the computation
time for one DP sweep for only the
-axis takes between 0.002 s to 0.03 s depending on the number of
velocity intervals. If we consider matching the frequency of position controller of PX4 with this DP
sweep, we can consider a 0.5 m/s velocity interval. Ideally, if the DP sweep is to replace the position
controller, we should try to optimize the computation time so that it is equal to or below 100 Hz.
This could be achieved with very small horizon lengths.
2.3.1. Distance Interval Selection
An ideal distance interval would have constant acceleration for the application of discretized
translational dynamics equation. The change in velocity states is performed via PID velocity controller.
Hence, the ideal case of constant acceleration can only be approximated to the response of velocity
controller, if the distance interval is such that the time spent in this distance interval is equal to or
greater than the rise time (90% of steady state) of the velocity response. In cases where the velocity
change in the two states is such that, for that distance interval, it will take a time equal to the rise time to
reach the desired velocity, then the assumption of constant acceleration can be applied. If the required
velocity change in the two states will take less time than the time vehicle travels a particular distance
step, it would result in the vehicle accelerating and then coasting in one distance step.
This simply
means that, in such a case, a constant acceleration assumption cannot be applied. In such a case, the
transition cost must be calculated in two steps. The first step is to use the rise time to calculate the
transition cost and the second step is to calculate the transition cost from rise time to the end of the
distance interval. In other words, in this case, the distance interval is divided into two parts: the
first part with an acceleration and the second part without any acceleration. In cases where the time
required to travel the distance interval and reach the desired velocity is greater than the rise time,
this desired velocity is then not feasible and should be discarded from the search-space. The distance
interval should have a value high enough that for the minimum velocity intervals it has possible states
to explore.
2.3.2. DP Sweep Trigger
Each time the UAV wants to travel a distance equal to the distance interval, the DP algorithm is
activated. In order to achieve this, the following equation is used to convert the position vector into a
periodic function
ς=sin (P+$)
, where
is the position(currently considering only x), and
is used
to select the distance interval value to trigger the DP algorithm. The values of
range from
1 to 1,
and whenever the value ςcrosses zero, the DP algorithm is triggered.
When the DP sweep is not triggered, the last set-point velocity is continuously being sent to the
drone. This is important since, apart from sending the velocity commands for the transport direction,
the algorithm is also continuously sending the velocity commands for vertical and lateral position
control. The lowest allowable publish rate for position controller is 2 Hz since the firmware used in
autopilot of software in loop simulations and hardware experiments triggers the fail-safe mode if the
velocity commands are publishing below the threshold frequency.
2.3.3. DP Sweep Sample Plots
Two sample DP sweeps are plotted in Figure 4. These plots are developed using the DP code
with the following specifications: the DP sweep consists of only one horizon length with 10 distance
intervals. The velocity states ranged from 0 m/s to 14 m/s. In Figure 4, the weighted cost function
is represented by colored lines with varying thickness. The cost function values were normalized
and, based on the cost function value, the thickness of the line was given. Figure 4shows the cost
Energies 2019,12, 3049 10 of 25
function values distribution for all possible states. We can see that the cost function has a cumulative
increasing effect since the first stages show thin blue lines and the last stages show thick and red lines.
After doing this search space sweep, the DP algorithm traces back the the optimal path by starting
from the last transition costs and selects the minimum cost all the way back to the first node.
(a) (b)
Figure 4.
A Dynamic programming (DP) sweep with velocity interval 2 m/s and horizon length 10 m,
initial velocity 2 m/s and final velocity 0 m/s. Width of the lines and intensity in redness shows a high
value of combined cost function (a)λ=0.5; (b)λ=0.9.
3. Numerical Simulation Experiments
The algorithm was tested on a complete hex-rotor model developed in MATLAB Simulink.
The purpose of developing a simulation tool is two-fold: first to asses the proposed algorithm,
secondly, the simulation tool can also be used to asses the mission completion feasibility prior to the
actual flight or software in loop simulations. It also assists in deciding the parameters of the algorithm,
which otherwise would be very difficult to do with the real flight. The Simulink model consisted of a
hex-rotor model, control block and optimization algorithm block as shown in Figure 5. The hex-rotor
model uses Equations (8)–(13) as described in Section 2.2 previously in detail. The optimization block
consists of the real-time dynamic programming optimization algorithm.
Figure 5.
Description of the complete system for simulation tests of the algorithm, including the
controller block, the hex-rotor block, and the optimization algorithm block.
Several factors affect the energy consumption during an aerial transport using a UAV.
Some of
these factors include the drag caused by wind speed and the density of air. Wind speed and air density
are known to vary with elevation. Air density decreases at higher altitudes and wind speed increases
Energies 2019,12, 3049 11 of 25
with elevation. Several blocks were added to the hex-rotor model, which include a power consumption
estimation block, atmospheric density calculation block, wind speed estimation block, and the payload
model block. The interactions of all these blocks and information inflow and outflows from these blocks
are explained in Figure 5. The controller block, atmospheric density model, wind speed model, and
payload model are described further below in Sections 3.13.4, respectively.
The inertial
parameters of
the DJI-F550 UAVs were found by using the developed CAD model and a CAD software was used to
find the inertial parameters. The mass of the UAV was calculated after taking measurements from a
weight scale. The altitude, attitude and position controllers were added to the Simulink model and
tuned. All parameters used in the simulation are presented in Table 1.
The controller
gains mentioned
in Table 1were obtained by tuning the system manually.
Table 1. List of all the parameters for DJI F-550 drone model.
Parameter Component Value Parameter Component Value
Kpvx Controller 0.15 lmi n UAV 0.88m
Kivx Controller 0.001 Kppx Controller 0.4
Kdvx Controller 0.04 Kp py Controller 0.4
Kpvy Controller 0.15 Max Velocity UAV 12 m/s
Kivy Controller 0.001 Max Altitude UAV 2.2 m
Kdvy Controller 0.04 Min Altitude UAV 1 m
θController 20 UAV mass UAV 3.4 Kg
θController 0 Payload mass Payload 0.586 Kg
θController 8 Arm length UAV 0.27 m
φController 20 Iz UAV 0.05
φController 0 Ix UAV 0.037
φController 8 Iy UAV 0.037
3.1. Controller Block
The controller block consists of a position controller, which generates velocity set-points and
a velocity controller, which generates the attitude set-points, the attitude controller, generates the
required motor speeds.
3.1.1. Position Controller
The position controller is simply a proportional controller, which generates the velocity set-points
to maintain the desired position:
Vs p =epKpp,
where Kpp is the proportional gain for position controller and
ep=Pdes Pcurrent.
3.1.2. Velocity Controller
The velocity controller is a PID controller, which generates the attitude set-points:
θsp =evKpv +Kiv Zev+KD
is the proportional gain for velocity controller, and
is the integral gain for velocity
controller, Kdv is the derivative gain for velocity controller, and
ev=Vdes Vcurrent.
Energies 2019,12, 3049 12 of 25
3.2. Atmospheric Density Model
The atmospheric density model is based on a US standard atmosphere [
]. It is implemented in
this study using an international standard atmosphere block, which takes the altitude as an input and
provides the density as an output.
3.3. Wind Speed Model
Wind speed is an important factor which can affect the power consumption of a multi-rotor during
the flight. Off-line energy assessment would require the measurements of wind speed. Wind speed
measurements are usually available online from the meteorological department or can be obtained via
an on-board wind speed sensor, and it is measured at a certain altitude. However, by using a power
law, wind speed can be estimated at the required altitude, provided that we have measurements of
wind speed at the near ground altitude. The power law is explained as Equation (26) and is obtained
from [25]:
, (26)
is the estimated wind velocity at the elevation
, and
is the wind velocity at near ground
elevation zg.αis called the Hellman exponent, whose values can be found using Table 2.
Table 2. Hellman exponents for different locations
Location Hellman Exponent
Unstable air above open water surface: 0.08
Neutral air above open water surface: 0.10
Unstable air above flat open coast: 0.11
Neutral air above flat open coast: 0.18
Stable air above open water surface: 0.27
Unstable air above human inhabited areas: 0.27
Neutral air alcove human inhabited areas: 0.30
Stable air above flat open mast: 0.40
Stable air above human inhabited areas: 0.60
3.4. Payload Model
Payload is currently considered as a block with a surface area
, and hence a drag co-efficient
CD, and a mass mp. The drag force acting on the payload is calculated by using Equation (27):
FDp =1
2·CDp ·ρa·Ap(Vv+Vw)2, (27)
is the vector representing the drag forces acting on the payload,
is the co-efficient of
drag for payload,
is the payload surface area exposed,
is the velocity of the hex-rotor,
is the
wind speed and ρais the density of air at an altitude a.
3.5. Thrust Irregularity in Forward Flight
The thrust is generally modeled as
, which is called the thrust model at hover. This model,
however, does not accurately predict the thrust and motor speed relation during high speed forward
flights as it is observed in [
]. It was also concluded in [
] that, for the same rotor speed, the thrust
decreases greatly at higher angles of attack and higher speeds. Another experimental study by [
showed that the total lift produced by various multi-rotors decreases by decreasing pitch angles from
when subject to 6 m/s wind speed. In another study [
], a controller is proposed to be
able to compensate for thrust irregularity during forward flights. This thrust deficit in high speed
forward flight means that the propellers have to rotate faster than usual to produce the thrust required.
Energies 2019,12, 3049 13 of 25
This simply means that more energy is consumed during forward flight to maintain the same thrust.
In [30], this thrust irregularity is modeled using Equations (28)–(30).
The induced air velocity due to rotation of rotors is shown in Equation (28)
2ρA, (28)
is the air velocity caused by the rotation of propellers in hovering conditions,
is the thrust
in hovering condition, ρis air density and Ais the propeller sweep area:
vivαsinβ, (29)
is the actual thrust produced,
is the induced air velocity from the propellers caused by
forward flight.
is the upstream air velocity and
is the attitude angle. The induced velocity
the air from propellers can be calculated using the following equation as [31]
q(vαcos β)2+vivβsin β2. (30)
The above formulation was applied on results from [
] and was found to be matching with
experimental results.
The effects of forward velocity and the pitch angle can be further seen via plotting the ratio
from Equation (29) in Figure 6. Forward velocities
ranging from 0–12 m/s and pitch angles
ranging from 0–40
were plugged in Equation (30) along with the hover air velocity
as calculated
from Equation (28) to calculate induced velocity
. The induced velocity
, pitch angle
, forward
and air velocity in hover
were plugged in Equation (29) to obtain the ratio
for each
pitch angle value and forward velocity and plotted in Figure 6. At very low pitch angles and higher
velocities, the ratio
remains above
. However, at higher pitch angles and higher velocities, the
ratio stays below
. This simply shows that, for the same power, the thrust produced in the hovering
condition will be higher than the thrust produced in forward velocity. The energy consumption
calculation in the numerical model accounts for this thrust deficit in high speed forward flight using
the method described in Figure 7.
Figure 6.
The variation of
with respect to the pitch angle and the forward velocity is shown using
the color-bar.
Energies 2019,12, 3049 14 of 25
Figure 7. Corrected power consumption calculation based on thrust irregularity.
3.6. Assumptions
The attitude dynamics of the UAV is decoupled with the payload, therefore any drag force acting
on the payload will not effect the attitude angle of the UAV. This assumption is supported by using
a multi degree of freedom based free joints arm in software in the loop simulations. In hardware
experiments, the payload attachment point can be a ball joint or a small cable; this will decouple
the attitude dynamics of the multi-rotor with payload. We do not model the thrust irregularity
happening at higher pitch angles and higher forward velocities in the dynamic model. Therefore,
a thrust irregularity compensator is not required as it is suggested by [
]. The energy consumption
calculation is, however, performed while considering the thrust irregularity in forward flight as
discussed in Section 3.5. We also assume that the attitude, altitude and velocity controller of the UAV
is robust enough to deal with extreme conditions (high velocities, high attitude angles). The velocity
controller is not a model based controller as it is shown in [
], therefore drag force compensation
during high speed forward flight is not required. The RTDP algorithm receives the current state of the
UAV and provides a velocity decision. This velocity decision can be implemented using the controller
presented in this paper or more calibrated ones such as [
]. We assume that the UAV is capable of
lifting a cube with a frontal area of 0.12 m
. Energy loses during fast maneuvers are not considered.
These losses include the braking of the motors to change the motor speeds to achieve differential thrust
required for attitude change. This assumption can be supported by the fact that frequent attitude angle
changes are not happening between point A to point B during the transport. Side wind is neglected in
this analysis. Based on the lateral frame drag, it is necessary for the multi-rotor to balance the drag
force to be able to maintain the lateral position. This drag can also influence the energy consumption
in addition to the drag caused due to forward flight. A sudden change in air density is out of the scope
of the current paper. The sudden change in air density can occur if there is a dust devil vortex or a
thermal plume in the path. The horizontal length of such thermal plumes is considerably less than the
journey length of our simulations. In real applications, the air density should be updated based on the
elevation of the UAV.
3.7. Numerical Simulation Results
A numerical simulation test was conducted where the UAV was carrying a cubic payload of
0.12 m
surface area and negligible weight from [0, 0, 10 m] to [250 m, 0, 10]. The initial velocity
at point of origin [0, 0, 10 m] and at goal position [250 m, 0, 10] were 0 m/s. The velocity interval
0.1 m/s
was considered. Distance intervals of 1, 2, 3, 4 m were considered, whereas the horizon
considered consisted of 10 distance interval steps. The DP sweep was triggered at every distance
interval step. Three test cases with values of
λ= (
0.2, 0.5, 0.7
were performed. The results of the
test cases are shown in Tables 36. These results show the percentage of energy savings when the
is increased from 0.2 to 0.7. Table 5shows that the journey time decreases when we decrease the
Energies 2019,12, 3049 15 of 25
from 0.7 to 0.2. Another test was performed with an increment of velocity interval of 0.5 m/s and the
difference in energy saving was found to be less than 1%. Extended simulations were performed with
distance interval ranging from 1 m–4 m for a velocity interval of 0.1 m/s. The results of the extended
simulations are tabulated in Tables 3,4and 6.
Table 3. Numerical simulation results with distance interval 1 m.
Weight-age (λ)Time (S) Difference (%) Energy (kJ) Difference (%) Velocity
Interval (m/s)
0.2 29.73 15.64% 28.2 +18.01% 0.1
0.5 34.39 0 23.12 0 0.1
0.7 40.446 +17.6% 22.39 3.15% 0.1
Table 4. Numerical simulation results with distance interval 2 m.
Weight-Age (λ)Time (S) Difference (%) Energy (kJ) Difference (%) Velocity
Interval (m/s)
0.2 31.19 6.7% 29.3 +11.12% 0.1
0.5 33.46 0 26.04 0 0.1
0.7 38.58 +15.3% 23.16 11.05% 0.1
Table 5. Numerical simulation results with distance interval 3.12 m.
Weight-age (λ)Time (S) Difference (%) Energy (kJ) Difference (%) Velocity
Interval (m/s)
0.2 29 14.48% 27.6 +11.44% 0.1
0.5 33.2 0 24.44 0 0.1
0.7 38.5 +15.96% 22.89 6.34% 0.1
0.2 29 15.37% 27.6 +10.9% 0.2
0.5 33.46 0 24.59 0 0.2
0.7 38.5 +15.06% 22.83 7.15% 0.2
0.2 29.93 9.75% 28.01 +12.85% 0.3
0.5 32.85 0 24.41 0 0.3
0.7 38.5 +17.19% 22.83 6.47% 0.3
0.2 29.93 9.62% 28.01 +12.7% 0.4
0.5 32.81 0 24.45 0 0.4
0.7 38.5 +17.34% 22.83 6.62% 0.4
0.2 29.96 11.64% 28.03 +11.48% 0.5
0.5 33.45 0 24.81 0 0.5
0.7 38.08 +13.84% 22.76 8.26% 0.5
Table 6. Numerical simulation results with distance interval 4 m.
Weight-age (λ)Time (S) Difference (%) Energy (kJ) Difference (%) Velocity
Interval (m/s)
0.2 30.89 2.39% 28.14 +5.044% 0.1
0.5 31.63 0 26.72 0 0.1
0.7 36.5 +15.39% 23.63 11.56% 0.1
The results of the numerical simulation are also presented in Figure 8. These results include
the transport trajectory, the velocity profiles, the pitch angles, the opposing drag forces, and the
energy consumption profiles. The plots include the numerical simulation results for a value of
λ= (
0.2, 0.5, 0.7
. The velocity profile for
0.2 shows the highest value, which results in more drag
forces and thus requires higher pitch angles to maintain the velocity. The higher pitch angles and
higher forward velocities supplement the increase of thrust deficit, which means that more power is
consumed. Therefore, although
0.2 results in the fastest transportation, it results in more power
Energies 2019,12, 3049 16 of 25
consumption. However, for
0.7, the velocity is lower, which results in lower drag forces, which
in turns results in a lower pitch angle. The lower pitch angle and smaller forward velocities result
in a reduction of thrust deficit, hence the total energy consumption during this transportation is low
although the mission took more time to complete.
(a) (b) (c)
(d) (e)
Figure 8.
Flight plots for numerical simulations performed for
λ= (
0.2, 0.5, 0.7
. (
) trajectory of the
UAV; (
) velocity of the UAV; (
) pitch angle of the UAV; (
) opposing drag force of the UAV; (
) energy
consumption of the UAV.
4. Software in the Loop Simulations
Examples of Software in the loop (SITL) simulations in literature are [
]. More accurate
simulation frameworks are also being developed [
] for multi-UAV simulations. Software in the
loop simulations (SITL) are performed to verify the functionality of the code before performing the
hardware experiments. In general, our SITL setup can be explained by the following Figure 9.
Figure 9. Software in the loop simulation architecture.
Energies 2019,12, 3049 17 of 25
Using the Table 1, the DJI F-550 drone was modeled in Gazebo as shown in Figure 10, and the
same PX4 flight stack was simulated as it is used in the real autopilot. The ROS, MAVROS package,
PX4 flight stack simulator, Gazebo simulator and the RTDP algorithm were running on the same
Linux computer. The computer has a 2.4 GHz processor and 32 GB RAM. All communication between
MAVROS, Gazebo and gripper node were performed via the local ROS MASTER.
The RTDP algorithm was communicating with the flight stack through ROS Master using the
MAVROS package. The flight stack was communicating with the DJI F550 drone in the Gazebo
simulator. It is a widely accepted practice to test new algorithms in SITL simulations before testing
them on real hardware. The autopilot firmware is developed by hundreds of researchers and it is
available online with the source code. The RTDP algorithm developed can use MAVROS via the
robotics toolbox of MATLAB to be able to communicate with the autopilot. A multi-link arm was fixed
at the base of the DJI-F550 model as shown in Figure 10. All the joints in the multi-link arm were free
and were mimicking the behavior of the cable. A gripper ROS node was running which can create and
delete links between the payload and the gripper attached at the end of the multi-link arm.
(a) (b)
Figure 10.
Software in the loop Gazebo DJI F550 model along with the description of the manipulator;
(a) DJI-F550 model in Gazebo simulator with PX4 flight stack simulation; (b) the multi-link free joints
based arm.
SITL Experiments
Similar parameters were used in the SITL simulation experiments as discussed in Section 3.7.
RTDP algorithm was developed in Simulink, which was running on the same laptop where the Gazebo
simulator was running. A velocity interval of 0.1 m/s was used. Similar to the autopilot, the simulated
flight stack also records the flight data including the actuator signals, attitude, velocity and position
estimates. The autopilot logs were used to plot the performance plots. The publish rates of the topics
during the aerial transport in SITL simulations were observed using an ROSTOPIC tool and were
plotted in Figure 11. A delay in publishing of the topic can trigger the fail-safe activation for off-board
failure. The frequency of publishing is very high when the DP sweep is not triggered. When the DP
sweep is triggered, we expect the publishing frequency to go down. The lowest publishing frequency
is 4.8 Hz. The results of the simulations are plotted in Figure 12. The numerical simulation results and
software in the loop simulation results are similar. In both cases, at
0.2, the DP sweeps resulted in
high velocities, which created higher parasitic drag. This resulted in higher pitch angles, which lead
to high power consumption. The drone behavior was similar for numerical simulations and SITL
simulations for values of
0.5, 0.7. This shows that the algorithm developed is compatible with
the autopilots’ firmware and it satisfies the constraints imposed by the parameters of the firmware.
These parameters include the multi-rotor constraints such as max velocity, max altitude, fail-safe mode
activation conditions, etc. It also proves the applicability and usefulness of the algorithm.
Energies 2019,12, 3049 18 of 25
(a) (b)
Figure 11.
Publish rates and time as obtained by the ROSTOPIC tool with a sample of two consecutive
messages. (
) publish rate of the velocity command topic; (
) minimum and maximum time between
two commands.
(a) (b) (c)
(d) (e)
Figure 12.
Flight plots for Gazebo SITL simulations performed for
λ= (
0.2, 0.5, 0.7
. (
) trajectory of
the UAV; (
) velocity of the UAV; (
) pitch angle of the UAV; (
) power consumption profile; (
) energy
consumption of the UAV.
5. Real-Time Flight Experiments
The system hardware as shown in Figure 13 consists of an external computer, and a DJI-F550 with
Pixhawk autopilot with PX4 firmware for low level control. DJI-F550 also had an on-board companion
computer, which was connected to the autopilot via MAVLINK. A cable-based gripper inspired by
Gough–Stewart platform was developed and attached to the Hex-rotors. The gripper is a payload
attachment mechanism that allows for automation of attachment and release of the payload.
The electro-permanent magnet was activated and deactivated by using an Arduino Nano
connected to the companion computer via USB link. The gripper can be activated and deactivated
during autonomous operation, using an ROS node. The electro-permanent magnet required power for
activation and deactivation only; therefore, it was powered by the same battery that was powering the
Energies 2019,12, 3049 19 of 25
drone. A current and voltage sensor called Elogger V4 is connected in series between the batteries
and the ESCs of the UAVs while bypassing the power supply of the onboard computer as shown in
Figure 13. Elogger was also used by [
] for logging the current and voltages of a UAV during
the flight. The cable material was chosen to withstand the weight of the payload requirements.
The software for the experiment is based on ROS interface MAVROS. An Optitrack system is used
for position estimation. The algorithm is written in MATLAB Simulink and it communicates to the
drone via MAVROS through ROS MASTER. Further details of experimental procedure are provided in
“Supplementary Materials”. The videos of all experiments are available at [41].
(a) (b)
Figure 13.
A description of the hardware used in the testing of the RTDP algorithm. (
) software
architecture of experimental setup; (
) DJI-F550 drone used in the experiments; (
) power source and
route for various components of the drone.
5.1. Baseline Experiment
An initial baseline experiment was performed, where the Drone01 was kept in hover at fixed
altitude set-point. There was no payload attached to the Drone01. The drone01 is the same drone that
was modeled in SITL and Simulink simulations. The power consumption was recorded and compared
with the SITL and Simulink model hover power consumption. The results are plotted in Figure 14.
It can
be seen in Figure 14 that power consumption in SITL and Simulink models at hovering condition
matches with the hardware experiments. However, there are minor discrepancies during the takeoff,
which can be ignored for this study since the optimization is performed for transportation from point
A to point B.
Energies 2019,12, 3049 20 of 25
Figure 14. Hover power consumption in real hardware and the SITL and Simulink experiments.
5.2. Lab Scale RTDP Transportation
An experiment was performed to test the feasibility of transportation from point
to point
using RTDP experiments. A distance interval of 0.1 m was considered, with 10 distance intervals in one
horizon. The starting point was
0, 0, 0.8
, whereas the goal position was
0, 6, 0.8
. A velocity interval
of 0.1 m/s was considered with a velocity range of 0–12 m/s. The time-energy weight-age factor of
0.7 was chosen. The drone01 took off manually while it was transported to point
. At point
the RTDP algorithm was triggered and, when the drone reached point
, the RTDP algorithm was
switched off and drone was manually landed. The trajectory of the transportation and the velocity
profiles of the numerical simulation and the experiments are plotted in Figure 15.
(a) (b) (c)
Figure 15.
Lab scale RTDP based aerial transportation. (
) drone01 trajectory in simulation and
experiment; (
) drone01 velocity profiles in experiments and simulations; (
) drone01 energy profiles
in experiments and simulations; (
) drone01 trajectory in simulation and experiments also showing DP
sweep with terminal costs in first horizons and DP sweep without terminal cost in the last horizon.
Energies 2019,12, 3049 21 of 25
The experiment showed that the drone01 moved from point
to point
The numerical simulation results and experimental results are in close agreement. This proves the
compatibility and experimental applicability of the algorithm.
After comparing the simulation and experiments for
0.7, two more experiments were
performed to compare
0.7 with
0.5, 0.2. The horizon length, distance interval, start point
and goal point were kept the same and fixed. The results of the experiments are plotted in Figure 16
showing the comparison of velocity profiles and trajectories between
0.7, 0.5, 0.2. As expected,
the drone01 reaches the destination in the order
0.2, 0.5, 0.7, which is also reflected in the velocity
profiles. This test was performed in the lab environment where, due to space constraints and safety
limitations, high speed flights were not achieved. However, these tests confirmed that the algorithm
is capable of transporting the drone01 from point A to point B successfully, while manipulating the
journey time. Minimization of journey time is useful in energy savings when the opposing drag forces
are not significant, which is the case in these lab experiments.
(a) (b)
Figure 16.
Lab scale RTDP based aerial transportation. (
) drone01 trajectory in experiments for
0.2, 0.5, 0.7; (
) drone01 velocity profiles in experiments for
0.2, 0.5, 0.7; (
) drone01 trajectory
in experiments for λ=0.2, 0.5, 0.7.
5.3. Variable Goal Position Experiment
The real-time DP algorithm’s effectiveness in changing the velocity profile based on goal position
change is tested in the lab environment. The results of the simulation and experiments for the variable
goal position are plotted in Figure 17. In this simulation and experimental test, the goal position is
initially provided as 2.5 m, whereas, after 3.5 s, the goal location is changed to 5 m. This is reflected in
the velocity profile and the position-time profile. The
-trajectory of the drone shows a change at 3.5 s.
The velocity profile also shows a sudden change between 3 and 4 s. It shows that the velocity profile
Energies 2019,12, 3049 22 of 25
was first selected for a goal position at 3 m, but then, because of the goal position change, the velocity
profile was also changed. The experiments and the numerical simulations are in good agreement.
(a) (b)
Figure 17.
Lab scale RTDP based aerial transportation while the goal position changed at 3.5 s.
) drone01 trajectory in simulation and experiment; (
) drone01 velocity profiles in experiment and
simulations; (
) drone01 trajectory during experiment and numerical simulation, initial and final drop
location are shown.
6. Conclusions
In this research study, a real-time dynamic programming algorithm is proposed for achieving
energy efficient motion for aerial UAV transportation. The proposed algorithm is based on
minimization of cost function. The cost function is the weighted sum of the energy cost and journey
time cost. A factor
is used to assign weight-age to favor either the time of the journey or energy
consumption. After the description of the algorithm, a numerical model was developed and explained,
which is used for the testing of the algorithm. After testing in the numerical model, the algorithm was
also tested in Software-in-the-loop Gazebo simulation environment. The lab scale experiments were
also conducted to validate the simulation results. After that, low speed flight tests were conducted for
0.2, 0.5, 0.7, which showed that, by changing
to a lower value, we can reduce the journey time
for low speed flights. The real-time decision-making of the algorithm was tested by changing the goal
position mid-flight in another experiment. The algorithm is capable of altering the journey time or
energy spent based on the provided value of
. Due to the definition of
, it was found that increasing
the value of
decreases the energy consumption, while the journey time increases. The future works
include expansion of cost function to include the 3D motion. The cost function will be modified to
remove current assumptions such as assumption of constant altitude. After that, we plan to expand the
current work to dual UAV payload transportation to test the scalability of the developed algorithms.
Supplementary Materials:
Further description of experimental procedure is available online at http://www.
Energies 2019,12, 3049 23 of 25
Author Contributions:
T.T., D.G. and Y.Z. supervised the design of experiments and algorithm development;
A.M. developed the algorithm, developed the numerical and SITL model and performed the experiments; T.T.
provided assistance with lab experiments and SITL simulations; Y.Z. supervised the Simulink based numerical
model; A.M. wrote the paper, while T.T., D.G. and Y.Z. revised the manuscript.
This publication is based upon work supported by the Khalifa University of Science and Technology
under Award No. RC1-2018-KUCARS.
Acknowledgments: The authors would like to give special thanks to the lab staff for the support.
Conflicts of Interest: The authors declare no conflict of interest.
The following abbreviations are used in this manuscript:
CoG Center of gravity
UAV Unmanned aerial vehicles
SITL Software in the loop
RTDP Real-time dynamic programming
EPM Electro permanent magnet
ESC Electronic speed controller
Lippitt, C.D.; Zhang, S. The impact of small unmanned airborne platforms on passive optical remote sensing:
A conceptual perspective. Int. J. Remote Sens. 2018,39, 4852–4868. [CrossRef]
DDC. Drone Delivery Canada. Available online: (accessed on
6 April 2019).
Wall Street Journal. Google Drones Can Already Deliver You Coffee In Australia. Available online: (accessed on 6 April 2019).
Zraick, K.. Like ‘Uber for Organs’: Drone Delivers Kidney to Maryland Woman. The New York Times.
Available online: (accessed on
30 April 2019).
Estévez, J.; Lopez-Guede, J.M.; Graña, M. Quasi-stationary state transportation of a hose with quadrotors.
Robot. Auton. Syst. 2015,63, 187–194. [CrossRef]
Swiss Post. Drones: A Vision Has Become Reality. Available online:
(accessed on 1 July 2019).
Stolaroff, J.K.; Samaras, C.; O’Neill, E.R.; Lubers, A.; Mitchell, A.S.; Ceperley, D. Energy use and life cycle
greenhouse gas emissions of drones for commercial package delivery. Nat. Commun.
,9, 409. [CrossRef]
Junaid, A.; Konoiko, A.; Zweiri, Y.; Sahinkaya, M.; Seneviratne, L. Autonomous wireless self-charging for
multi-rotor unmanned aerial vehicles. Energies 2017,10, 803. [CrossRef]
Campi, T.; Cruciani, S.; Feliziani, M. Wireless power transfer technology applied to an autonomous electric
uav with a small secondary coil. Energies 2018,11, 352. [CrossRef]
Morbidi, F.; Cano, R.; Lara, D.; Morbidi, F.; Cano, R.; Lara, D.; Generation, M.E.P.; Morbidi, F.; Cano, R.;
Lara, D. Minimum-Energy Path Generation for a Quadrotor UAV. In Proceedings of the IEEE International
Conference on Robotics and Automation, Stockholm, Sweden, 16–21 May 2016; pp. 2–8.
Kumar, V.; Michael, N. Opportunities and challenges with autonomous micro aerial vehicles. Int. J. Robot. Res.
2012,31, 1279–1291. [CrossRef]
Abdilla, A.; Richards, A.; Burrow, S. Power and endurance modelling of battery-powered rotorcraft.
In Proceedings of the IEEE International Conference on Intelligent Robots and Systems, Hamburg, Germany,
28 September–2 October 2015; pp. 675–680.
Morton, S.; Papanikolopoulos, N. A small hybrid ground-air vehicle concept. In Proceedings of the
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Vancouver, BC, Canada,
24–28 September 2017; pp. 5149–5154.
Economou, J.; Kladis, G.; Tsourdos, A.; White, B. UAV optimum energy assignment using dijkstra’s algorithm.
In Proceedings of the European Control Conference (ECC), Kos, Greece, 2–5 July 2007; pp. 287–292.
Energies 2019,12, 3049 24 of 25
Karydis, K.; Kumar, V. Energetics in robotic flight at small scales. Interface Focus
,7, 20160088. [CrossRef]
Gandolfo, D.C.; Salinas, L.R.; Brandão, A.; Toibero, J.M. Stable path-following control for a quadrotor
helicopter considering energy consumption. IEEE Trans. Control. Syst. Technol.
,25, 1423–1430.
Bangura, M.; Mahony, R. Nonlinear dynamic modeling for high performance control of a quadrotor.
In Proceedings of the Australasian Conference on Robotics and Automation, Wellington, New Zealand,
3–5 December 2012; pp. 1–10.
Salaan, C.J.; Okada, Y.; Hozumi, K.; Ohno, K.; Tadokoro, S. Improvement of UAV’s flight performance by
reducing the drag force of spherical shell. In Proceedings of the 2016 IEEE/RSJ International Conference on
Intelligent Robots and Systems (IROS), Daejeon, Korea, 9–14 October 2016; pp. 1708–1714.
19. Bellman, R.E.; Dreyfus, S.E. Applied Dynamic Programming; Princeton University Press: Princeton, NJ, USA,
2015; Volume 2050.
Levermore, T.; Sahinkaya, M.N.; Zweiri, Y.; Neaves, B. Real-time velocity optimization to minimize energy
use in passenger vehicles. Energies 2016,10, 30. [CrossRef]
Bertsekas, D.P.; Bertsekas, D.P.; Bertsekas, D.P.; Bertsekas, D.P. Dynamic Programming and Optimal Control;
Athena Scientific Belmont: Belmont, MA, USA, 2005; Volume 1.
Yacef, F.; Rizoug, N.; Bouhali, O.; Hamerlain, M. Optimization of Energy Consumption for Quadrotor
UAV. In Proceedings of the International Micro Air Vehicle Conference and Flight Competition (IMAV),
Toulouse, France, 18–21 September 2017; pp. 215–222.
Filatov, D.M.; Friedrich, A.I.; Devyatkin, A.V. Parameters identification of thrust generation subsystem for
small unmanned aerial vehicles. In Proceedings of the 2nd International Conference on Control in Technical
Systems (CTS), St. Petersburg, Russia, 52–27 October 2017; pp. 381–383.
24. U.S. Standard Atmosphere; NASA: Greenbelt, MD, USA, 1976.
Bañuelos-Ruedas, F.; Angeles-Camacho, C.; Rios-Marcuello, S. Analysis and validation of the methodology
used in the extrapolation of wind speed data at different heights. Renew. Sustain. Energy Rev.
,14, 2383–2391.
Gill, R.; D’Andrea, R. Propeller thrust and drag in forward flight. In Proceedings of the IEEE Conference on
Control Technology and Applications (CCTA), Mauna Lani, HI, USA, 27–30 August 2017; pp. 73–79.
Svacha, J.; Mohta, K.; Watterson, M.; Loianno, G.; Kumar, V. Inertial Velocity and Attitude Estimation for
Quadrotors. In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems
(IROS), Madrid, Spain, 1–5 October 2018; pp. 1–9.
Molter, C.; Cheng, P. Propeller performance calculation for multicopter aircraft at forward flight conditions
and validation with wind tunnel measurements. In Proceedings of the International Micro Air Vehicle
Conference and Flight Competition (IMAV), Stuttgart, Germany, 12 December 2017; pp. 307–315.
Russell, C.R.; Jung, J.; Willink, G.; Glasner, B. Wind Tunnel and Hover Performance Test Results for Multicopter
UAS Vehicles. In Proceedings of the 72nd American Helicopter Society (AHS) International Annual Forum
and Technology Display, West Palm Beach, FL, USA, 17–20 May 2016; NASA: Greenbelt, MD, USA, 2016;
pp. 1–20.
Huang, H.; Hoffmann, G.M.; Waslander, S.L.; Tomlin, C.J. Aerodynamics and control of autonomous
quadrotor helicopters in aggressive maneuvering. In Proceedings of the IEEE International Conference on
Robotics and Automation, Kobe, Japan, 12–17 May 2009; pp. 3277–3282.
Leishman, G.J. Principles of Helicopter Aerodynamics with CD Extra; Cambridge University Press:
Cambridge, UK, 2006.
Svacha, J.; Mohta, K.; Kumar, V. Improving quadrotor trajectory tracking by compensating for aerodynamic
effects. In Proceedings of the 2017 International Conference on Unmanned Aircraft Systems (ICUAS),
Miami, FL, USA, 13–16 June 2017; pp. 860–866.
y, V.; Báˇca, T.; Saska, M.; Pˇeniˇcka, R.; Krajník, T.; Thomas, J.; Thakur, D.; Loianno, G.; Kumar, V.
Cooperative autonomous search, grasping, and delivering in a treasure hunt scenario by a team of unmanned
aerial vehicles. J. Field Robot. 2019,36, 125–148. [CrossRef]
Aminzadeh, A.; Atashgah, M.A.; Roudbari, A. Software in the loop framework for the performance assessment
of a navigation and control system of an unmanned aerial vehicle. IEEE Aerosp. Electron. Syst. Mag.
2018,33, 50–57. [CrossRef]
Energies 2019,12, 3049 25 of 25
Jaleel, H.; Abdelkader, M.; Shamma, J.S. Real-time distributed motion planning with submodular
minimization. In Proceedings of the IEEE Conference on Control Technology and Applications (CCTA),
Copenhagen, Denmark, 21–24 August 2018; pp. 885–890.
Schofield, S.D.; Edwards, M.J.; Green, R.D. Sensitivity analysis of multirotor position control.
In Proceedings of the International Conference on Image and Vision Computing New Zealand (IVCNZ),
Christchurch, New Zealand, 4–6 December 2017; pp. 1–6.
Hinas, A.; Ragel, R.; Roberts, J.; Gonzalez, F. A Framework for Vision-Based Multiple Target Finding and
Action Using Multirotor UAVs. In Proceedings of the International Conference on Unmanned Aircraft
Systems (ICUAS), Dallas, TX, USA, 12–15 June 2018; pp. 1320–1327.
Heintz, C.; Bailey, S.C.; Hoagg, J. Formation Control of Fixed-Wing Unmanned Aircraft: Theory and
Experiments. In AIAA Scitech 2019 Forum; AIAA: Reston, VA, USA, 2019; p. 1168.
Gadalla, M.; Zafar, S. Analysis of a hydrogen fuel cell-PV power system for small UAV. Int. J. Hydrog. Energy
2016,41, 6422–6432. [CrossRef]
Chekiri, R.; Lavoie, P.; Richarz, W.G. Experimental aeroacoustics of a two-strut, two-wheel landing gear
in a propeller wake. In Proceedings of the 52nd Aerospace Sciences Meeting, National Harbor, MD, USA,
13–17 January 2014; p. 0020.
Abdullah Mohiuddin. UAV Velocity Optimized Aerial Transportation via RTDP—Youtube. Available online: (accessed on 11 April 2019).
2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (
... The work [25] was the first demonstration of real time energy optimization for single UAV aerial transportation, followed by other papers such as [26]. However, to the best of our knowledge, the work presented in this paper is the first to propose dual UAV optimized payload transportation, while considering the load distribution between two UAVs. ...
... The lack of studies in this area is due to additional difficulties related to the highly coupled dynamics where errors in localization in one UAV can significantly destabilize the other UAV. The RTDP based optimization performed on a single UAV in [25] can be scaled to multi-UAVs. This is achieved by finding optimal policy decisions for the velocity set points in the transport direction while regulating the payload distribution. ...
... This complication is due to the fact that when thrust is regulated between the two UAVs, the attitude angles are different and the individual energy consumption is also different. This required amendment in the previous approach in [25] for the calculation of the energy part of the cost function. This procedural correctness comes at the cost of increased computational time and costs. ...
Full-text available
in this paper, a real-time dynamic programming (RTDP) approach was developed for the first time to jointly carry a slung load using two unmanned aerial vehicles (UAVs) with a trajectory optimized for time and energy consumption. The novel strategy applies RTDP algorithm, where the journey was discretized into horizons consisting of distance intervals, and for every distance interval, an optimal policy was obtained using a dynamic programming sweep. The RTDP-based strategy is applied for dual-UAV collaborative payload transportation using coordinated motion where UAVs act as actuators on the payload. The RTDP algorithm provides the optimal velocity decisions for the slung load transportation to either minimize the journey time or the energy consumption. The RTDP approach involves minimizing a cost function which is derived after simplifying the combined model of the dual-UAV-payload system. The cost function derivation was also accommodated to dynamically distribute the load/energy between two multi-rotor platforms during a transportation mission. The cost function is used to calculate transition costs for all stages and velocity decisions. A terminal cost is used at the last distance interval during the first phase of the journey when the velocity at the end of the current horizon is not known. In the second phase, the last stage or edge of the horizon includes the destination, hence final velocity is known which is used to calculate the transition cost of the final stage. Once all transition costs are calculated, the minimum cost is traced back from the final stage to the current stage to find the optimal velocity decision. The developed approach was validated in MATLAB simulation, software in the loop Gazebo simulation, and real experiments. The numerical and Gazebo simulations showed the successful optimization of journey time or energy consumption based on the selection of the factor λ . Both simulation and real experiments results show the effectiveness and the applicability of the proposed approach.
... Regarding the control of the path that has the least energy consumption, the authors in [5] evaluate the relationship between navigation speed and energy consumption in a miniature quadrotor helicopter, which travels a desired path through experimental tests. Other studies have been conducted on in real time dynamic programming (RTDP) algorithms [6]. In [8], a trajectory generation approach using quadratic programming is investigated for the control of an aerial vehicle equipped with a robot arm. ...
... . As a terminal step, we need to compute the concrete inputs α j (t i ), j = 1, 2, 3, 4, i.e. the four motor accelerations, to plug into Eq. (6). To this end, from the knowledge of the angle velocities ϕ(t i ), θ(t i ), ψ(t i ) and the virtual inputs u z (t i ), u ψ (t i ), u ϕ (t i ), u θ (t i ), we proceed with the partial inversion of (2) to compute the required motor velocities ω j (t i ), j = 1, 2, 3, 4, so that the digital actuations are obtained, by first-order discretization of the last equations in (6), as ...
... One of the main challenges is the limited flight time due to small battery capacities. Despite all the advances in battery technologies, flight time for small and medium quadrotors is about half an hour [4]. Therefore, researchers started to suggest different methods to save energy and increase flight times. ...
... (2) To the best of our knowledge, there is no investigation of the effect of polynomial trajectories on the energy consumption of quadrotors with a cable-suspended payload. Additionally, investigating quadrotor's energy consumption for quadrotor's with cable-suspended payload for any trajectories other than polynomial trajectories is rare in the literature, e.g., [4]. Most of the literature studied only quadrotors without payloads. ...
Conference Paper
The paper presents the methodology of pitching moment prediction in drone rotors equipped with a variable pitch propeller. The proposed study describes extension of the available software like QPROP to calculate the blade pitching moment. The simulation results are validated with experimental data from the wind tunnel test and shows an example use of the proposed method. The research highlights potential applications where this analysis is crucial and where challenges of variable pitch propeller design might be solved with the proposed method.
Flying robots popularly known as drones or UAVs are emerging technologies of the current era. A significant amount of research work has been undertaken in this area in the last few years. Considering the current scenario where aerial vehicles are taking a major part of the market it is important to have an effective and robust design of flying robots. This paper aims to examine the categories of flying robots based on the features that include a range from petite to large and its body structure, wing designs, tail design, propulsion system, and gripper mechanisms along with the associated materials and manufacturing techniques. Again the work is intended to review the respective challenges faced by each category. Mostly the challenges faced by flying robots are design challenges, material selection, and fabrication challenges which are discussed in the paper. In this paper, we have summarized various designs of flying robots developed to date as well as we have focused on major features to be taken care of while designing flying robots. This paper has tried to focus on different design aspects and challenges faced by flying robots so that further research can be carried out to develop effective flying robots in the future.
This article focuses on investigating the effect of quadrotor's trajectory, especially polynomial trajectories, on its energy consumption. First, model-free expressions for power and energy quotients are introduced to relate quadrotor's power and energy directly to its acceleration. This allows to qualitatively estimate quadrotor's energy consumption and compare the effect of different trajectories on energy consumption of identical or different quadrotors independent of quadrotor's manufacturing specifications. Then, polynomial trajectories are analytically investigated for rest-to-rest 1-D scenarios. Scenarios in 3-D with arbitrary kinematic boundary conditions are analyzed via Monte Carlo Simulations with a sample of 10 000 sets of arbitrary boundary conditions. Polynomial trajectories are compared to energy-minimized trajectories in the literature. The results show that increasing the degree of the polynomial increases quadrotor's energy consumption. Moreover, this article suggests using minimum acceleration trajectories as energy-efficient polynomial trajectories. Finally, the results are validated experimentally.
This paper introduces a method to navigate the UAV flight and to suppress the payload swing by a nonlinear coupling control augmented with time-varying cable length. Special error signals are introduced that bring further coupling of dynamics among various states of the UAV and enable the flight control to effectively suppress the payload swing. The control is designed with the help of the dynamic model of the UAV in lift and transport operation. The stability of the control is proven in the Lyapunov sense. Extensive simulations and experiments have been carried out to demonstrate the effectiveness of the proposed nonlinear coupling control. Two popular control approaches in the literature are chosen to compare with the current work. It is found that the proposed control is quite effective in suppressing the payload swing while tracking the pre-determined path at the same time.
In this work, the flow behavior of a dragonfly-inspired corrugated wing was undertaken using a subsonic wind tunnel to assess the aerodynamic performance and flow characteristics. The test was performed at Reynolds numbers (Re) 46,000 and 67,000, which is the flying regime of micro aerial vehicles. A wing having the same geometrical dimensions as the midsection of the dragonfly forewing known as corrugated wing and another wing having the same geometry as the first wing without corrugations known as hybrid wing were fabricated using 3-D printing machine. The tufts of three different colors were glued at three locations, i.e., 0, 30, and 60% of the semispan of both wings at the trailing edge to visualize the flow separation and reversal phenomenon. The boundary layer rack was used at these three locations to obtain velocity gradient and boundary layer thickness. The result of the tuft flow visualization showed that the flow pattern at all three locations is not the same for the same Re and angles of attack. At high AOA, the corrugated wing shows lesser velocity gradients than the hybrid wing for both tested Reynolds number. The results clearly demonstrate that the bio-inspired corrugated wing surpassed the hybrid wing by delaying the stall up to 28%.KeywordsBio-inspired wingBoundary layerTuft flow visualizationStall angleFlow reversal
In addition to ground robots, UAVs have also been regarded as a promising means to reshape future wireless communication systems. In the rest of the book, we consider the usage of UAVs in wireless communication. This chapter discusses how to plan the trajectory of a solar-powered UAV under a cloudy condition to secure the communication between the UAV and a target ground node against multiple eavesdroppers. We propose a new 3D UAV trajectory optimization model by taking into account the UAV energy consumption, solar power harvesting, eavesdropping, and no-fly zone avoidance. An RRT method is developed to construct the UAV trajectory. Computer simulations and comparisons with a baseline method demonstrate that the proposed method is able to produce trajectories to ensure the valid wireless communication link with the ground node and prevent eavesdropping.
Full-text available
This paper addresses the problem of autonomous cooperative localization, grasping and delivering of colored ferrous objects by a team of unmanned aerial vehicles (UAVs). In the proposed scenario, a team of UAVs is required to maximize the reward by collecting colored objects and delivering them to a predefined location. This task consists of several subtasks such as cooperative coverage path planning, object detection and state estimation, UAV self‐localization, precise motion control, trajectory tracking, aerial grasping and dropping, and decentralized team coordination. The failure recovery and synchronization job manager is used to integrate all the presented subtasks together and also to decrease the vulnerability to individual subtask failures in real‐world conditions. The whole system was developed for the Mohamed Bin Zayed International Robotics Challenge (MBZIRC) 2017, where it achieved the highest score and won Challenge No. 3—Treasure Hunt. This paper does not only contain results from the MBZIRC 2017 competition but it also evaluates the system performance in simulations and field tests that were conducted throughout the year‐long development and preparations for the competition.
Unmanned airborne systems (UAS), particularly the class of UAS referred to as small-unmanned airborne systems (S-UAS), have the potential to revolutionize the science, practice, and role of remote sensing. S-UAS-collected remote sensing data differ from that acquired from larger airborne and space-borne platforms in myriad ways. To provide an indication of the novel remote sensing capabilities that S-UAS are poised to enable and identify research priorities for realizing the full potential of remote sensing from these novel platforms, characteristics of S-UAS platforms and their impact on data and information products are analysed in the context of remote sensing model and the remote sensing communication model. Results indicate that S-UAS will not only enable a range of novel remote sensing capabilities but also present clear challenges to the remote sensing community. These challenges, including increased data volume, a paucity of appropriate analysis approaches, and restrictions on autonomous operation (both regulatory and technological), point towards several near-term research priorities.
The use of autonomous vehicles is currently a growing trend, particularly in applications such as manufacturing, hazardous material handling, and surveillance [1]. A quadcopter is an emerging rotorcraft concept for an unmanned aerial vehicle (UAV) that is equipped with four rotors, two pairs of counter-rotating, and fixedpitch blades located at the four corners of the vehicle [2]. In a quadcopter system, yaw is controlled by increasing the angular velocity of rotors that spin in the same direction to a level higher than the velocity of other rotor pairs, and pitch or roll is controlled by increasing the angular velocity of a rotor to a level greater than the velocity of a diametrically opposite rotor (Figure 1) [3].