Model of Collaborative UAV Swarm Towards Coordination and ...zcliu/file/uav-agent.pdf · as a drone...
Transcript of Model of Collaborative UAV Swarm Towards Coordination and ...zcliu/file/uav-agent.pdf · as a drone...
1 Northwestern Polytechnical University (NPU) 2 Universitat Autònoma de Barcelona
Model of Collaborative UAV Swarm Towards Coordination and Control Mechanisms Study
Xueping Zhu1 Zhengchun Liu2 Jun Yang1
An unmanned aerial vehicle (UAV), known in the mainstream as a drone and also referred to as an unpiloted aerial vehicle and a remotely piloted aircraft (RPA) by the International Civil Aviation Organization (ICAO), is an aircraft without a human pilot aboard.
IntroductionUAV term
Delivery
IntroductionUAV applications
Modern Technology Make it:1.Cheaper; 2.Easier & Faster to Deploy; 3.smaller and flexible.
UAV Civilian Applications: 1.Disaster management; 2.Remote area surveillance; and hazardous environment monitoring;
3.Fast package delivery; 4.Aerial photo & video.
Disaster
OutlineUAV challenges and proposed solution
sophisticated tasks, such as searching survival points, multiple target monitoring and tracking UAV swarms is foreseen
more complex control, communication and coordination mechanisms
mechanisms are difficult to test and analyze under flight dynamic conditions, especially robustness!
model UAV swarm as multi-agent system (for testing in virtual environment)
Flying UAV have many uncertainties and constraints that are difficult to model through a “If...Then...Else” statement.
Mathematic (6DoF) +
Computational (MAS principle)invisible message switch
layer
Interaction
agent model
Test-Bed,especially for robustness test +
Simulation!
Fixed-wing Rotary-wing
๏ Hand/catapult launched ๏ Long flight endurance, can cover a
lot of area๏ Difficult to fly if not fully automated ๏ Requires fully automated flight features
for full usability ๏ Minimal maintenance, modest
expenses
๏ Automated launch ๏ Limited flight endurance๏ Hover capability -> Precise 3D
inspection of stationary objects/delivery of chemicals
๏ Difficult to fly if not fully automated ๏ Requires fully automated flight features
for full usability ๏ Minimal maintenance, modest
expenses
Study ObjectFixed VS. Rotary
Fixed-wing as an example(Long flight endurance, can cover a lot of area)
heterogeneous (fixed and rotary) in the future.
the agent model
F b
aero
=
2
4C
x
Cy
Cz
3
5 qS, M b
aero
=
2
4m
x
my
mz
3
5 qSL, q =
1
2
⇢V 2(1)
2
4Fxb
Fyb
Fzb
3
5= mg0
2
4� sin ✓
cos ✓ sin�cos ✓ cos�
3
5+
2
4XYZ
3
5+
2
4P0
0
3
5(2)
Cz
= f (V,↵, �y
) = Cz0 + C↵
z
· ↵+ C�yz
· �y
(3)
.
Cz0 is the lift coe�cient when both ↵ and �z are equal to zero (the zero lift
coe�cient).
C↵z and C
�yz are related with velocity, write as: C↵
z = f (V ), C↵z = f
0(V ).
Where:
6DoF - agent modelFlight Dynamic
e.g.,
Table 1: List of aerodynamic data needed to model an UAV
Notation
Definition
m, IThe mass and moment of inertia of the rigid body.
SSum of wing area.
LThe chord length.
C�zx,y,z
-V The drag/side/lift force coe�cient derivative due to correspond-
ing horizontal tailplane to velocity.
C↵
x,y,z
-VThe drag/side/lift force coe�cient derivative due to angle of
attack to velocity.
C(x,y,z)0 The zero drag/side/lift force coe�cient.
m�zx,y,z
-VThe roll/pitch/yaw moment coe�cient derivative due to corre-
sponding horizontal tailplane to velocity.
m↵
x,y,z
-VThe roll/pitch/yaw coe�cient derivative due to angle of attack
to velocity.
m(x,y,z)0 The zero roll/pitch/yaw moment coe�cient.
6DoF - agent modelParameters for Flight Dynamic Model
Physical Parameters
Aerodynamic Force Coefficient
Moment Coefficient
˙Vb
=
1
m
2
4Fxb
Fyb
Fzb
3
5+ V
b
⇥ !, ! =
2
4pqr
3
5(1)
! = I�1
0
@
2
4M
xb
Myb
Mzb
3
5+ (I!)⇥ !
1
A , I =
0
@Ixx
�Ixy
�Ixz
�Iyx
Iyy
�Iyz
�Izx
�Izy
Izz
1
A(2)
2
4pqr
3
5=
2
4'0
0
3
5+
2
41 0 0
0 cos' sin'0 � sin' cos'
3
5
2
40
˙✓0
3
5+
2
41 0 0
0 cos' sin'0 � sin' cos'
3
5
2
4cos ✓ 0 � sin ✓0 1 0
sin ✓ 0 cos ✓
3
5
2
40
0
˙
3
5= J�1
2
4'˙✓˙
3
5
(3)
2
4'˙✓˙
3
5= J
2
4pqr
3
5=
2
41 (sin' tan ✓) (cos' tan ✓)0 cos' � sin'0 (sin' sec ✓) (cos' sec ✓)
3
5
2
4pqr
3
5(4)
↵ = arctan
Vzb
Vxb
, � = arcsin
Vbyq
Vxb
2+ V
yb
2+ V
zb
2(5)
O
O O
I
I
6DoF - agent modelKinematic - Agent State Transition
O
Model an unmanned aerial vehicle as autonomous agent. The sensors and communication devices give it possible to interact with companions and environment. The 6 degree-of-freedom (6DoF) flight dynamic model reflects constraints and uncertainties as well as state transition.
Model Concept of UAV as an Agent
Communication amongst UAVs(base of interaction)
Direct In-direct
๏ WiFi, ZigBee, Bluetooth etc; ๏ low power consumption, less latency,
free of charge, etc; ๏ without support base station; ๏ distance restriction.
๏ GSM, Satellite etc; ๏ high power consumption, long latency,
pay by bytes; ๏ with ground support base station; ๏ almost no distance restriction.
Communication amongst UAVs(crucial for agents to coordinate properly)
Proper combination could be better; communication ways should be carefully considered;
Communication Mode / Media
message types descriptionbroadcast an agent share its current state to a set of agents
actively; query an agent send a message to a set of agents to request
their states;sync an agent (e.g. group leader) send a sync request
message to a group of agents, then all the agents who receive the sync request will broadcast their current state to all the other group members.
Interaction message amongst UAVs
As for implementation, we implemented a message switch layer (MSL) work in an observer role (not exist in real situation, only for implementation, all information about simulation world is accessible by observer), all the communication amongst UAVs will pass through this layer and MSL will record all these messages for post-simulation analysis. This provides a flexible way for control mechanism designer to optimize the communication policies in micro-level behavior.
A proof-of-concept Case Study
Demo. Application
z(x, y) =nX
i=1
h
i
⇤ exp �✓x� c
xi
g
xi
◆2
�✓y � c
yi
g
yi
◆2!
Environment Model:
.
h
i
controls height of peak i;
c
xi and c
yi mark central position of peak i;
g
xi and g
yi control peak gradient in x and y orientation respectively.
Where:
is the altitude of the given coordinate in the simulated terrain.z (x, y)
n is the number of peaks (mountains).
1. a mountain area of size 10 × 20km2; 2. Five UAVs “scan” the area on preset height (100
meters over ground); 3. distance between two neighboring UAVs should be
kept as 500 meters (due to the restriction of sensor, e.g., camera);
4. vehicles are set to fly with their curing speed (35 m/s) to save power consumption.
Demo. ApplicationSimulating Conditions
1 2 3 4 5
µ
Demo. ApplicationSwarm Contro Model
x = �Lx+ bµ
y = c
Tx
2
66664
x1
x2
x3
x4
x5
3
77775=
2
66664
�1 1 0 0 01 �2 1 0 00 1 �2 1 00 0 1 �2 10 0 0 1 �1
3
77775
2
66664
x1
x2
x3
x4
x5
3
77775+
2
66664
10000
3
77775µ
y =⇥1 0 0 0 0
⇤
2
66664
x1
x2
x3
x4
x5
3
77775
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
0−1000
01000
20003000
40005000
60007000
80009000
10000
0
100
200
300
400
500
z e(meter)
uav5uav4uav3uav2uav1
xe(meter)
ye(meter)
Demo. Application
3D View of UAV swarm trajectory
0 100 200 300 400 500 600−0.5
−0.4
−0.3
−0.2
−0.1
0
0.1
0.2
0.3
0.4
0.5
time(second)
a z(m/s2 )
uav5uav4uav3uav2uav1
constraint
0 100 200 300 400 500 600−40
−30
−20
−10
0
10
20
30
40
time(second)
heig
ht e
rror
(m)
uav5uav4uav3uav2uav1
0 100 200 300 400 500 60033
33.5
34
34.5
35
35.5
36
36.5
37
time(second)
sear
ch sp
eed
(m/s
)
Conclusions and Future work(Conclusions)
1. This work is a part of a longterm project about intelligent UAV swarm application.
2. Model of UAV swarm; 3. Implemented a prototype in NetLogo3D; 4. one proof-of-concept case study was presented to show
its potential usage;
Conclusions and Future work(future work)
1. in a Netlogo3D world the number of patches grows very quickly that slow the model down or even cause NetLogo to run out of memory. => implement by using Google’s Go programming language without concept of patch.
2. rotary-wing, such as quadcopter is also widely used in some application. So, we will construct the model of quadcopter in order to provide the ability of building heterogeneous UAV swarm.
Thank You Very Much for Your Attention!
Q + A = Progress
1 Northwestern Polytechnical University (NPU) 2 Universitat Autònoma de Barcelona (UAB)
Model of Collaborative UAV Swarm Towards Coordination and Control Mechanisms Study
Xueping Zhu1 Zhengchun Liu2 Jun Yang1