Nonlinear Control of UAVs Using Dynamic Inversion Alejandro Osorio Department of Aerospace Engineering Cal Poly Pomona AIAA Aerospace Systems and Technology (ASAT) Conference May 3, 2014
Nonlinear Control of UAVs Using
Dynamic Inversion
Alejandro Osorio
Department of Aerospace Engineering
Cal Poly Pomona
AIAA Aerospace Systems and Technology (ASAT) Conference
May 3, 2014
Overview
• Unmanned Aerial Vehicles
• Motivations
• Research Objectives
• Twin-Engine Airplane
• Nonlinear Flight Dynamics Model
• Flight Test for Data Acquisition
• Nonlinear Dynamic Inversion
• Future Work
2
Advantages of Unmanned Aerial Vehicles
• Do not contain or need a qualified pilot on board
• Can enter environments that are dangerous to human life
• UAVs are indispensable for military and civilian
applications
• Military
• Reconnaissance, battlefield damage assessment, strike
capabilities, etc.
• Civilian
• Infrastructure maintenance, agriculture management, disaster
relief, etc.
• Significantly lower operating costs
3
Motivations
• Existing UAVs have a high acquisition cost and are
limited to restricted airspace
• Cost effective operations require
• Increased autonomy, reliability, and availability
• Most existing autopilots are designed using linearized
flight dynamics model and lack robustness
• Nonlinear controllers can work for entire flight envelope,
thereby helping increase UAV autonomy
4
Research Objectives
• Develop and validate nonlinear flight dynamics models for
Cal Poly Pomona UAVs
• Use Dynamic Inversion Technique for the design of
nonlinear controllers for the UAVs
• Verify the controllers in software and hardware-in-the-loop
simulations
• Validate the design in flight tests
• Use H∞ (H-Infinity) control system design technique along
for the design and implementation of robust nonlinear
controllers
5
Cal Poly Pomona UAV Lab
• Dedicated to research on advanced topics in
flight dynamics and control
• The lab consists of airplane and helicopter UAVs
of various sizes and payload capacity
• Sensors and associate equipment
• Internal measurement units
• Differential GPS
• Air data probes
• Commercial-off-the-shelf autopilots
• Laser altimeter
6
CPP Research UAVs
Sig Kadet Airplane
SR-100 Helicopter
12’ Telemaster Airplane
Raptor-90 Helicopter
Twin-Engine Airplane
7
Twin-Engine Airplane
• DA 50 Gasoline engine powered
• Length- 95 inches, wing span- 134 inches
• Empty weight- 42 lbs, payload- up to 25 lbs
• Equipped with Piccolo II autopilot for autonomous flight
and data acquisition
8
Nonlinear Flight Dynamics Model
9
Force Equations:
𝑈 = 𝑅𝑉 +𝑊𝑄 − 𝑔 sin 𝜃 +𝐹𝑋
𝑚
𝑉 = −𝑈𝑅 +𝑊𝑃 + 𝑔 sin ϕ cos 𝜃 +𝐹𝑌
𝑚
𝑊 = 𝑈𝑄 − 𝑉𝑃 + 𝑔 cos ϕ cos 𝜃 +𝐹𝑧
𝑚
Kinematic Equations:
ϕ = 𝑃 + 𝑡𝑎𝑛𝜃(𝑄𝑠𝑖𝑛ϕ + Rcosϕ)
𝜃 = 𝑄𝑐𝑜𝑠ϕ + Rsinϕ
𝛹 =𝑄𝑠𝑖𝑛ϕ + 𝑅𝑐𝑜𝑠ϕ
𝑐𝑜𝑠𝜃
Moment Equations:
P = 𝑐1𝑅 + 𝑐2𝑃 𝑄 + 𝑐3𝐿 + 𝑐4𝑁
𝑄 = 𝑐5𝑃𝑅 + 𝑐6 𝑃2 − 𝑅2 + 𝑐7𝑀 𝑅 = 𝑐8𝑃 − 𝑐2𝑅 𝑄 + 𝑐4𝐿 + 𝑐9𝑁
Navigation Equations:
𝑥 = 𝑈𝑐𝑜𝑠𝜃𝑐𝑜𝑠𝛹 + 𝑉 −𝑐𝑜𝑠ϕsin𝛹 + 𝑠𝑖𝑛ϕsinθ𝑠𝑖𝑛𝛹+𝑊 sinϕsin𝛹 + cosϕsinθ𝑐𝑜𝑠𝛹
𝑦 = 𝑈𝑐𝑜𝑠𝜃𝑠𝑖𝑛𝛹 + 𝑉 𝑐𝑜𝑠ϕcos𝛹 + 𝑠𝑖𝑛ϕsinθ𝑠𝑖𝑛𝛹+𝑊 −sinϕcos𝛹 + cosϕsinθ𝑠𝑖𝑛𝛹
ℎ = 𝑈𝑠𝑖𝑛𝜃 − 𝑉𝑠𝑖𝑛ϕcosθ − 𝑐𝑜𝑠ϕcosθ
Aerodynamic Model
10
Aerodynamic Forces and Moments
𝐿𝐴 = 𝑞 𝑆𝐶𝑙𝑏
𝑀 = 𝑞 𝑆𝐶𝑚𝐶 𝑁 = 𝑞 𝑆𝐶𝑛𝑏
Aerodynamic Coefficients
𝐶𝐷 = 𝐶𝐷𝑜 + 𝐶𝐷𝛼𝛼 + 𝐶𝐷𝑞𝑄𝐶
2𝑉𝑜+ 𝐶𝐷𝛼 𝛼
𝐶
2𝑉𝑜+ 𝐶𝐷𝑢
𝑢
𝑉𝑜+ 𝐶𝐷𝛿𝑒𝛿𝑒
𝐶𝑌 = 𝐶𝑌𝛽𝛽 + 𝐶𝑌𝑝𝑃𝑏
2𝑉𝑜+ 𝐶𝑌𝑟𝑅
𝑏
2𝑉𝑜+ 𝐶𝑌𝛿𝑎𝛿𝑎 + 𝐶𝑌𝛿𝑟𝛿𝑟
𝐶𝐿 = 𝐶𝐿𝑜 + 𝐶𝐿𝛼𝛼 + 𝐶𝐿𝑞𝑄𝐶
2𝑉𝑜+ 𝐶𝐿𝛼 𝛼
𝐶
2𝑉𝑜+ 𝐶𝐿𝑢
𝑢
𝑉𝑜+ 𝐶𝐿𝛿𝑒𝛿𝑒
𝐶𝑙 = 𝐶𝑙𝛽𝛽 + 𝐶𝑙𝑝𝑃𝑏
2𝑉𝑜+ 𝐶𝑙𝑟𝑅
𝑏
2𝑉𝑜+ 𝐶𝑙𝛿𝑎𝛿𝑎 + 𝐶𝑙𝛿𝑟𝛿𝑟
𝐶𝑚 = 𝐶𝑚𝑜+ 𝐶𝑚𝛼
𝛼 + 𝐶𝑚𝑞𝑄
𝐶
2𝑉𝑜+ 𝐶𝑚𝛼
𝛼 𝐶
2𝑉𝑜+ 𝐶𝑚𝑢
𝑢
𝑉𝑜+ 𝐶𝑙𝛿𝑒𝛿𝑒
𝐶𝑛 = 𝐶𝑛𝛽𝛽 + 𝐶𝑛𝑝𝑃𝑏
2𝑉𝑜+ 𝐶𝑛𝑟𝑅
𝑏
2𝑉𝑜+ 𝐶𝑛𝛿𝑎𝛿𝑎 + 𝐶𝑛𝛿𝑟𝛿𝑟
𝐷 = 𝑞 𝑆𝐶𝐷
𝐿 = 𝑞 𝑆𝐶𝐿
Y = 𝑞 𝑆𝐶𝑌
Flight Test
• The airplane flown for doublet inputs in aileron, rudder,
and elevator
• The data is used for the model validation
• Validated model is then used for control system design
11
662 664 666 668 670 672 674-10
-5
0
5
Aile
ron
(d
eg
)
Roll Doublet
662 664 666 668 670 672 674-50
0
50
100
Ro
ll A
ng
le (
de
g)
Time (sec)
Model Validation
12
14 16 18 20 22 24-10
0
10
A (
de
g)
Airplane Lateral-Directional Response
14 16 18 20 22 24-100
0
100
p (
de
g/s
ec)
Flight Data
Simulation
14 16 18 20 22 24-50
0
50
r (
de
g/s
ec)
Time (sec)
Airplane Longitudinal Response
13
16 18 20 22 24 26 28 30-10
0
10
Time (sec)
E (
de
g)
Airplane Longitudinal Response
16 18 20 22 24 26 28 30-100
-50
0
50
q (
de
g/s
ec)
Time (sec)
Flight Data
Simulation
FlightGear Model
14
Nonlinear Dynamic Inversion
• The nonlinear dynamic system can be represented as the
first order model
• 𝑥 = 𝑓 𝑥 + 𝑔 𝑥 𝑢
• Both functions f(x) and g(x) are nonlinear in x
• If the system is affine in the controls, then solving
explicitly for the control vector yields
• 𝑢 = 𝑔−1 𝑥 𝑥 − 𝑓 𝑥
• Replacement of the inherent dynamics with the desired
dynamics results in the control that will produce the
desired dynamics
• 𝑢 = 𝑔−1 𝑥 𝑥 𝑑𝑒𝑠 − 𝑓 𝑥
15
Time-Scale Separation
• Standard nonlinear equations of motion cannot be directly
used because the A matrix (system matrix) is not square
• The original dynamic model is formulated as two lower-
order systems
• Translational mechanics
• Rotational dynamics
• Four control inputs = four variables in each time-scale
• Dynamics are separated into slow and fast dynamics
• Slow controlled states are the angle of attack, climb angle, bank
angle and sideslip angle (α, γ, φ, β)
• The fast controlled states are the three angular rates plus the
forward speed (V, P, Q, R).
16
𝛼 = 𝑄 − tan 𝛽 𝑃𝑐𝑜𝑠𝛼 + 𝑅𝑠𝑖𝑛𝛼 +1
𝑚𝑉𝑐𝑜𝑠𝛽−𝐿 +𝑚𝑔𝑐𝑜𝑠𝛾𝑐𝑜𝑠𝜙 − 𝑇𝑠𝑖𝑛𝛼
𝛾 =1
𝑚𝑉𝐿𝑐𝑜𝑠𝜙 − 𝑚𝑔𝑐𝑜𝑠𝛾 − 𝑌𝑠𝑖𝑛𝜙𝑐𝑜𝑠𝛽 +
𝑇
𝑚𝑉, 𝑠𝑖𝑛𝜙𝑠𝑖𝑛𝛽𝑐𝑜𝑠𝛼 + 𝑐𝑜𝑠𝜙𝑠𝑖𝑛𝛼
Φ = 𝑃 + tan 𝜃(𝑄 sinΦ + 𝑅 cosΦ)
𝛽 = 𝑃 sin 𝛼 − 𝑅 cos 𝛼 +1
𝑚𝑉cos 𝛾 sinΦ + 𝑌 cos 𝛽 − 𝑇 sin 𝛽 cos𝛼
Nonlinear Coupled Differential Equations of Motion
17
Slow Dynamics
Fast Dynamics
𝑃 = 𝑐1𝑅 + 𝑐2𝑃 𝑄 + 𝑐3𝐿 + 𝑐4𝑁 𝑄 = 𝑐5𝑃𝑅 − 𝑐6 𝑃2 − 𝑅2 + 𝑐7𝑀
𝑅 = 𝑐8𝑃 − 𝑐2𝑅 𝑄 + 𝑐4𝐿 + 𝑐9𝑁
𝑉 =1
𝑚[−𝐷 + 𝑌𝑠𝑖𝑛𝛽 −𝑚𝑔𝑠𝑖𝑛𝛾 + 𝑇𝑐𝑜𝑠𝛽𝑐𝑜𝑠𝛼]
Time-Scale Separation Cont.
• The outer-loop involves the translational dynamics
• In response to position and velocity commands, it
produces the δ command for the inner-loop to track
• The inner-loop involves the rotational dynamics
• Tracks the attitude reference by determining the δT, δE,
δA, and δR commands
18
Nonlinear Dynamic Inversion Model
19
Future Work
• Further refine the flight dynamics model
• Use flight data for the development of flight dynamics
models using Parameter Identification techniques
• Design nonlinear controllers using dynamic inversion
techniques for complete autonomous missions
• Use H technique to design robust controllers
• Take into account modeling uncertainties
20
Acknowledgements
• NSF Award No. 1102382
• Hovig Yaralian
• Matthew Rose
• Nigam Patel
• Luis Andrade
• Dr. Subodh Bhandari, Mentor
21
Questions?
22