Networks of Autonomous Unmanned Vehicles - Department of ...

loutclankedAI and Robotics

Nov 13, 2013 (3 years and 8 months ago)

118 views

Networks of Autonomous
Unmanned Vehicles

Prof. Schwartz

Prof. Esfandiari

Prof. P. Liu

Prof. P. Staznicky

Slide Name

Research and Development Areas


Autonomous Robot Construction.


Cooperating Mobile Autonomous Robots.


Vision Systems.


Robot Flocking and Swarming


Robot swarms that adapt and learn (game theory and evolution).


Robot teams and learning.




Slide Name

Autonomous Vehicles

Built from low cost robot kit.

HandyBoard HC11 controller

Bluetooth communication channel.

Sonar sensor.

Able to control over internet.

On board navigation control
.


Slide Name

Robotic Tracking


Activmedia PeopleBot Robot


2 DOF camera


Optical flow
-
based target detection and
verification


Target’s motion is estimated using a particle
filter


Laser rangefinder


It is used to determine distance between
robot and target


Slide Name

Robotic Boat


developed by 4
th
-
year students


The boat can be controlled over a
wireless network


User with a PC and a web browser
can control the boat from anywhere


The web server is placed on the
on
-
board microcontroller, which
has not be done before by others


Slide Name

Unmanned Aircraft System (UAS) Development


UAS for geophysical surveys is being developed in the Mechanical
and aerospace Engineering department (M&AE), with industry
partner, an Ottawa company and with support from Systems and
Computer Engineering (SCE)


UAS has a demanding mission


8
-
hours endurance


Airspeed between 60 and 100 kts


Low altitude down to 30 ft above terrain; terrain following is required


Sensitive magnetometers are mounted on the wingtips


Magnetic signature of the air vehicle must be minimized

Slide Name

UAS Development: Status


Air Vehicle prototype is being built


Size: Wing span 16 ft, weight 200 lb, engine power 30 hp


Start of flight testing: spring of 2009


Four research projects are underway, collaboration between M&AE
and SCE:


Autonomous operation


Obstacle detection and avoidance


Magnetic signature control


Low
-
cost non
-
magnetic airframe


Slide Name

UAS Development: Experimentation


Two small aircraft, avionics test
beds, are being used for testing


Autopilot system


Telemetry system


Communication system


Iridium satellite system selected and
being tested


Altimeter system


A laser altimeter has been purchased for
testing



Slide Name

Robots leaving a room using game theory

Slide Name

Modelling

Robots leaving a room

Player B

Walk

Wait

Player A

Walk

-
1

X

Wait

-
1

0

1
,


X
Z
X
where
Slide Name

Cooperative robots and intelligence



Robots have own control and navigation algorithms



Robots only know their position and others

Slide Name

Video Processing and Understanding

Tracking of video objects

Slide Name

Intelligent Video Object Tracking

Tracking, counting and timing of video
objects.

Slide Name

Networks of Robots and Sensor Swarms

Slide Name

Vehicles Playing the Evader


Pursuit Game

Research Topics



Vehicles Learn Each Others
Dynamics.



Vehicles Adapt Behaviour
.



Coalition and Team Formation

Slide Name

Soccer Playing Robots


We are interested in imitating agent behavior that is space and time dependent


RoboCup is a good environment for such exploration


Our methodology:


1.
Perform data capture from logs generated by existing RoboCup clients

2.
Transform the captured data into a spatial knowledge representation format (a
scene
)

3.
Game
-
time: pick closest (or one of k
-
closest) captured scene to current one
and perform corresponding action


Slide Name

Robots Learning How to Play Soccer

Slide Name

Scene Recognition


Find best match(es) between current situation and stored scenes (k
-
nearest
-
neighbor search)


Perform associated action

-
> accuracy of the distance calculation function between two scenes is
crucial

“What should I do in this situation?”

“What did the observed agent do when faced with a
situation like this?”

Slide Name

The Robots Have Learned the Game

Slide Name

Limitations and Future Work


Short term


consider object velocity;


weigh the importance of an object based on its proximity to the player;


scene prototyping to reduce duplication and introduce more scene variation; (done!)


CBR
-
style adaptation of the action;


automatic weight determination is very time consuming: more tests required here. (done!)


Long term


Need to take into account state and context
-
based behavior:


non
-
visual info: body state, game state...


actions as part of a plan or succession of scenes


a clue: two similar scenes leading to different actions


might need to remember and backtrack to previous scene(s)


Higher
-
level representation for scenes


conversion to spatial and/or temporal logic?

Slide Name

Multiple Robots

Slide Name

Actions


Follow up
-
hill gradient





Or follow the down
-
hill gradient











i
i
j
i
j
i
i
i
i
j
i
j
i
i
i
y
x
R
y
x
T
y
x
R
y
x
T
u
,
,
,
,












i
i
u
d




Slide Name

Personality rewards


Courage:




Fear




Cooperation





2
1
,
,
1













i
i
i
i
i
u
y
x
T
y
x
T






2
2
1
shot
P
K
shot








2
1
,
,
1





















i
i
i
j
k
j
i
i
j
k
j
u
y
x
R
y
x
R


Slide Name

Algorithm


Calculate up
-
hill and down
-
hill gradients


Calculate personality rewards


Calculate if robot has been shot. If so, go back to base.


Update personalities:



Where
η

is the learning rate and the step size is:







t
t
t
i
i
i








1






t
t
t
j
j
i
i



3
1





Slide Name

Personalities dynamics

Slide Name

Swarm Intelligence and Personality Evolution


Game Theory, Coalition formation.


Evolutionary Game Theory.


Learning (fuzzy, adaptive, genetic).


Personality Traits.

Slide Name

This is a smart robot

Slide Name

Conclusion


Capability in Building Autonomous Vehicles


Autonomous Vehicle Control


Swarming


Evader/Pursuer





Learning and Adapting Networks



Robots leaving a room



Learning to play soccer



adapting personalities