Apresentação do PowerPoint - Sitraer 2014

Report
Modeling and Simulation of
Autonomous UAS for Risk Evaluation of
Collision with Other Aircraft
(Modelagem e Simulação de VANT Autônomo para
Avaliação de Risco de Colisão com Outras Aeronaves)
Safety Analysis Group (GAS)
Computer and Digital Systems Engineering Department (PCS)
School of Engineering (Escola Politécnica - Poli)
University of São Paulo (USP)
São Paulo, Brazil
Universidade de São Paulo
Brasil
1
http://www.gas.pcs.poli.usp.br/
Introduction
• About GAS (Safety Analysis Group).
• Safety issues of (non-military) UAS operation.
– Focus on collision avoidance (sense and
avoid).
– Pilot responsibility: “Nothing in these rules shall
relieve the pilot-in-command of an aircraft from
the responsibility of taking such action (...)”
(ICAO Rules of the Air [1]).
– Control and Communication (C2) link failure.
• Academia, foreseeable future and autonomous
UAS.
Universidade de São Paulo
Brasil
2
http://www.gas.pcs.poli.usp.br/
The manned aerial vehicle paradigm
• Civil Aviation: only a Remotely Piloted
Aircraft (RPA) “will be able to integrate into
the international civil aviation system in the
foreseeable future” (ICAO Circular 328 [3])
Human pilot in command.
 Pilot skill, insight, experience.
 Mood, stress.
Universidade de São Paulo
Brasil
3
http://www.gas.pcs.poli.usp.br/
The pilot in UAS operation
Link Failure
Ergonomics
Human Factors
Pilot
Remote
Piloting
Interface
Control and
Communication
(C2)
UAV
• C2 link failure: research needed (FAA, 2013 [3]).
• Human factors and Ergonomics (Waraich et al., 2013
[4]).
– E.g.: human pilot in “Automatic pilot” mode (Catino
& Patriotta, 2013 [5]).
Universidade de São Paulo
Brasil
4
http://www.gas.pcs.poli.usp.br/
Granting autonomy to the UAV
Link Failure
Autonomous
UAV
Ergonomics
Human Factors
Pilot
Remote
Piloting
Interface
Control and
Communication
(C2)
• Safety premise: an additional autonomous
UAV in the ATC system must not increase
collision risk more than an additional manned
aircraft would.
Universidade de São Paulo
Brasil
5
http://www.gas.pcs.poli.usp.br/
Hypotheses for an autonomous UAV
Main hypothesis: Granting the UAV an
autonomous piloting capability that ‘imitates’
the human pilot would make the autonomous
operation to present safety levels similar to
those observed in the operation of manned
aircraft.
Secondary hypothesis: If the UAV imitates only
correct maneuvers, the safety levels of an
autonomous operation would be greater as
compared to human-piloted aircraft (including
RPA).
Universidade de São Paulo
Brasil
6
http://www.gas.pcs.poli.usp.br/
‘Imitating’ - Learn from
Demonstration
• Learn from Demonstration (Argall et al., 2009 [6])
– Subset of Supervised Learning.
– Human pilot as a Demonstration Teacher.
– Piloting Autonomous System (PAS, Gimenes et al.,
2013 [7]) must ‘imitate’ the Demonstration Teacher.
• Learning culture is a characteristic of High Reliability
Organizations such as the ATC (Brooker, 2013 [8]).
Universidade de São Paulo
Brasil
7
http://www.gas.pcs.poli.usp.br/
UAV Model and CNS/ATM
Inertial
Measurement
Unit (IMU)
Motor
Speed
Controller
Motor
Speed
Controller
Positioning
sensor (GPS)
Central Processing Unit
Data link
interface
Camera
Motor
Speed
Controller
Universidade de São Paulo
Brasil
Motor
Speed
Controller
• Human pilot → see and
avoid.
• UAV → sense and avoid.
• Use of CNS/ATM
technologies for Sense
capability.
Automatic Dependent
Surveillance –
Broadcast (ADS-B).
8
http://www.gas.pcs.poli.usp.br/
PAS Model
Central Processing Unit
Feedback to PID controllers
IMU data
GPS data
Cooperative data
(data link
communication)
PAS
0
Throttle
Pitch
Flight plan
Roll
1
Camera data
Commands from
remote pilot (data
link
communication)
Universidade de São Paulo
Brasil
PID controllers
To motor
speed
controllers
Yaw
RPAS mode
on
9
http://www.gas.pcs.poli.usp.br/
Model Evaluation
LfD-based
UAV
(autonomous)
• Proof of concept
– UAV assembly and training.
– Qualitative evaluation.
• Simulation
– Aircraft encounter models (Kochenderfer et al.,
2010 [9])
RPA
Source: [9]
Universidade de São Paulo
Brasil
10
http://www.gas.pcs.poli.usp.br/
Model Evaluation (cont.)
• Simulation (cont.)
– Adaptation of (trained) PAS to the simulation
environment.
– Monte Carlo simulation (quantitative evaluation):
Trained PAS x Encounter models aircraft
Random
Nx
Universidade de São Paulo
Brasil
(
PAS
VS
)
11
http://www.gas.pcs.poli.usp.br/
Related Works
• Unmanned aircraft control:
Learning-based: Neural Networks, Support Vector
Machines, Learning from Demonstration (Ross et
al., 2013 [10]).
Focus on the computational problem only.
• Simulation-based collision risk estimation:
Aircraft encounter models (Kochenderfer et al.,
2010 [9]).
Comparison of collision avoidance methods (Holt
et al., 2013 [11] e por Alexopoulos et al., 2013
[12]).
Universidade de São Paulo
Brasil
12
http://www.gas.pcs.poli.usp.br/
Future Work
• Execution of training and proof of concept:
– UAV assembly.
– PAS training.
– Verification of collision avoidance capability.
• Collision risk evaluation:
– Selection of the encounter model.
– Adaptation of PAS to the simulation
environment.
– Evaluation of the results.
Universidade de São Paulo
Brasil
13
http://www.gas.pcs.poli.usp.br/
Final Considerations
• Learn from Demonstration concept applied to
autonomous UAS.
• Evaluation criterion: autonomous UAS safety
levels must be equal or greater as compared to
manned aircraft.
• Metrics for collision avoidance capability.
• Definition of next steps to obtain results.
• Computation problem vs air traffic problem.
Universidade de São Paulo
Brasil
14
http://www.gas.pcs.poli.usp.br/
References
[1] International Civil Aviation Organization, Annex 2 to the Convention on International Civil Aviation - Rules of the Air, no. July.
2005, pp. 1–74.
[2] International Civil Aviation Organization, “ICAO Cir 328, Unmanned Aircraft Systems (UAS),” 2011.
[3] Federal Aviation Administration, “Integration of Civil Unmanned Aircraft Systems (UAS) in the National Airspace System
(NAS) Roadmap,” 2013.
[4] Q. R. Waraich, T. a. Mazzuchi, S. Sarkani, and D. F. Rico, “Minimizing Human Factors Mishaps in Unmanned Aircraft Systems,”
Ergon. Des. Q. Hum. Factors Appl., vol. 21, no. 1, pp. 25–32, Jan. 2013.
[5] M. Catino and G. Patriotta, “Learning from Errors: Cognition, Emotions and Safety Culture in the Italian Air Force,” Organ. Stud.,
vol. 34, no. 4, pp. 437–467, Mar. 2013.
[6] B. D. Argall, S. Chernova, M. Veloso, and B. Browning, “A survey of robot learning from demonstration,” Rob. Auton. Syst., vol.
57, no. 5, pp. 469–483, May 2009.
[7] R. A. V. Gimenes, L. F. Vismari, V. F. Avelino, J. B. Camargo, J. R. Almeida, and P. S. Cugnasca, “Guidelines for the Integration of
Autonomous UAS into the Global ATM,” J. Intell. Robot. Syst., vol. 74, no. 1–2, pp. 465–478, Sep. 2013.
[8] P. Brooker, “Introducing Unmanned Aircraft Systems into a High Reliability ATC System,” J. Navig., vol. 66, no. 05, pp. 719–735,
Jun. 2013.
[9] M. J. Kochenderfer, M. W. M. Edwards, L. P. Espindle, J. K. Kuchar, and J. D. Griffith, “Airspace Encounter Models for Estimating
Collision Risk,” J. Guid. Control. Dyn., vol. 33, no. 2, pp. 487–499, Mar. 2010.
[10] S. Ross, N. Melik-Barkhudarov, K. S. Shankar, A. Wendel, D. Dey, J. A. Bagnell, and M. Hebert, “Learning monocular reactive
UAV control in cluttered natural environments,” 2013 IEEE Int. Conf. Robot. Autom., pp. 1765–1772, May 2013.
[11] J. Holt, S. Biaz, and C. A. Aji, “Comparison of Unmanned Aerial System Collision Avoidance Algorithms in a Simulated
Environment,” J. Guid. Control. Dyn., vol. 36, no. 3, pp. 881–883, May 2013.
[12] A. Alexopoulos, A. Kandil, P. Orzechowski, and E. Badreddin, “A Comparative Study of Collision Avoidance Techniques for
Unmanned Aerial Vehicles,” in 2013 IEEE International Conference on Systems, Man, and Cybernetics, 2013, pp. 1969–1974.
Universidade de São Paulo
Brasil
15
http://www.gas.pcs.poli.usp.br/
THANK YOU FOR YOUR ATTENTION
For more information:
www.gas.pcs.poli.usp.br
Thiago T. Matsumoto, M.Sc. student
[email protected]
Phone: +55 11 3091-5323
Universidade de São Paulo
Brasil
16
http://www.gas.pcs.poli.usp.br/

similar documents