Task-Oriented Mobile Actuator/Sensor Networks

cuckootrainΜηχανική

31 Οκτ 2013 (πριν από 3 χρόνια και 10 μήνες)

70 εμφανίσεις

Task
-
Oriented Mobile Actuator/Sensor Networks
:


Distributed Measurement for Distributed Control
” and/or

Distributed Control for Distributed Measurement
”?

YangQuan Chen


Center for Self
-
Organizing and Intelligent Systems (CSOIS),

Dept. of Electrical and Computer Engineering

Utah State University

E: yqchen@ece.usu.edu; T: (
435)797
-
0148; F: (435)797
-
3054

W: http://www.csois.usu.edu/people/yqchen



October 22, 2004, CSOIS Bi
-
Weekly Research Seminar Series

10/19/2004

SDL "Skunk Works" Project

Slide
-
2

Mobile Actuator
-
Sensor Network

(MAS
-
net)


Tasks


Efficiently deploy a group of mobile sensors to
characterize the dynamically evolving diffusion
boundary


Using the same mobility platform, mobile actuators
can actively control the formation of the diffusion
boundary to a desired zone/shape



Application scenarios
-

10/19/2004

SDL "Skunk Works" Project

Slide
-
3

MAS
-
net: Three Application Scenarios


Application Scenario 1 (land):

The safe ground boundary determination of the radiation
field from multiple nuclear radiation sources
.

In this case, each networked sensor is
mounted on a ground mobile robot. The mission is to determine the safe radiation boundary
of the radiation field from possibly multiple nuclear radiation sources. Each robot is actuated
according to spatial and temporal sensed information (radiation gradient, spatial position etc.)
from more than one actuated or mobile sensors.


Application Scenario 2 (water):

The nontoxic reservoir water surface boundary
determination and zone control due to a toxic diffusion source
.

Similar to Application
Scenario 1 if the toxic diffusion source is a one
-
time pouring and the diffusion is in steady
state. However, the boundary may be dynamically evolving if the toxic source keeps polluting
the reservoir. The
actuated or mobile sensors

are autonomous boats mounted with toxic
chemical concentration sensors. The boats are commanded according to the spatial
-
temporal
sensed information from more than one sensor. Furthermore, assume that some of the boats
(not all of the boats) are equipped with the relevant neutralizing chemicals to make the water
detoxified. By a proper design of distributed sensing and actuation/control strategies, it is
possible to control the zone or shape of the toxic region to match the given desirable
zone/shape. Now we have a complex distributed feedback control system that is more
challenging than the networked actuators and sensors themselves.


Application Scenario 3 (air):

The safe nontoxic 3D boundary determination and zone
control of biological or chemical contamination in the air
.

This scenario is similar to the
above water case, but it is more complicated since 3D space must be explored. Here, the
actuated or mobile sensors

are unmanned aerial vehicles (UAVs) equipped with
concentration detectors and anti
-
contamination chemical agent(s) distributors.


10/19/2004

SDL "Skunk Works" Project

Slide
-
4

MASNET Experimental Platform

(Conceptual Block Diagram)

10/19/2004

SDL "Skunk Works" Project

Slide
-
5


Actuated sensors


(mote
-
based robots)


take “plume” samples


Wireless communication


system broadcasts commands


to actuated sensors


Base station makes


plume prediction and


computes sensor locations


Vision system for


locating sensors


Air outlet



Fog “Contaminant”


(orange) introduced


into air stream


Fan blows


air (green)


through


system

2
-
D
System
Testbed
Concept

10/19/2004

SDL "Skunk Works" Project

Slide
-
6

MAS
-
net Platform Development


System architecture


Hardware configuration


Robot chassis


MICA board & circuit system


Camera system


Software configuration


System diagram


pGPS


Mote Software

10/19/2004

SDL "Skunk Works" Project

Slide
-
7

MAS
-
net

Key Sub
-
Systems


Mobility Platform (small mobile robots)


Sensors on Each Mobile Robot


Actuators on Each Mobile Robot


Based Station


Diffusion Generating Environment

10/19/2004

SDL "Skunk Works" Project

Slide
-
8

MAS
-
net

Mobility Platform


Mote Based Control, Wireless Communication,
and Interfacing Unit


Chassis, Wheel Assembly


Servo


Encoders


IR’s

10/19/2004

SDL "Skunk Works" Project

Slide
-
9

The Test Bed: Motes

GUI

Camera Driver

Serial Cable

Parallel Cable

Programming
Board

Mote
(MICA Board)

TinyOS

Wireless Communication

Motes and Robots

Camera

10/19/2004

SDL "Skunk Works" Project

Slide
-
10

MICA2

(Berkeley)

Control Board (USU)

AVR Atmega
128 (CPU)

CC1000 (Comm.)

2
Encoders

3
IR

(Sharp GP2D12)

2 Photo
-

Resistors

2
Servos

Sensors

3
V Power

6
V Power

2
ADC

2
PWM

3
ADC

2
ADC

Hardware Configuration of the
Mobility Platform

10/19/2004

SDL "Skunk Works" Project

Slide
-
11

Software on Mobile Mote

Stack/Xnp

(Comm.)

TinyOS

User
Applications

TinyDB

TinySchema

Low Level
Lib

2 Encoders

2 Servos

Other Sensors/Actuators

Other
Utilities of
TinyOS

10/19/2004

SDL "Skunk Works" Project

Slide
-
12

1
st

Prototype Photos

Mote
-
based Robot: USU MASmote

With Cover

Tag on top for
pGPS

10/19/2004

SDL "Skunk Works" Project

Slide
-
14

10 MASmotes

10/19/2004

SDL "Skunk Works" Project

Slide
-
15

The Trend of MAS
-
net Control


Symbolic+continuous dynamics


Distributed, asynchronous, networked environment


High
-
level coordination and autonomy


Automatics synthesis of control algorithm


Reliable systems made up of unreliable parts


-
>
Huge system modeling and control


IEEE Control Systems Magazine 2003 Apr + J.Song

10/19/2004

SDL "Skunk Works" Project

Slide
-
16

Basic Questions


Q1: Given the accuracy requirements, what is the
minimum number of robots?


Q2: How to drive the robots (differential two
-
wheels
drive and generic nonholonomic) to estimate the fog
diffusion.


Q3: How to control the robots to eliminate the fog.
(optimized with certain criterion)

10/19/2004

SDL "Skunk Works" Project

Slide
-
17

One Problem: Photoresistor (PR)


Problem description


Large derivative of PR characteristics


Max R: 6K~90K


Min R: 28 ~120 omega


Mapping (by Op Amp analog computation): v
o
=2(
(Rp
-
5K)/65 ) v
i ,
v
i
=1.5 Volt


10/19/2004

SDL "Skunk Works" Project

Slide
-
18

Sensor calibration (3 Qs)


Q4: Calibrate the PRs with the visual information
from the camera. After that, the camera is used
only for localization.


Q5: Very likely, the characteristics of the PR are
nonlinear. How to fit?


Q6: Reject the background light disturbance effect

10/19/2004

SDL "Skunk Works" Project

Slide
-
19

Optimization


Q7: What is the relationship between the number
of robots and the variance of the sensing errors?
Given the cost of a robot and a sensor, together
with the sensor characteristics distribution
function and the properties of the fog, can you tell
me the optimum number of robots and sensors to
purchase in order incur the minimum cost?

10/19/2004

SDL "Skunk Works" Project

Slide
-
20

Robust control


Q8: Infinite dimensional robust control. We have a group
of robots to observe an infinite dimensional system (fog).
Given a polynomial with interval coefficients as the
characteristic of the sensor, what is the minimum number
of robots we need? Using the robot control theory to
design a H_inf or H_2 controller to observe the system
with the minimum number of robots.


10/19/2004

SDL "Skunk Works" Project

Slide
-
21

Interval control


Q9: Infinite dimensional interval control. Answer
the same question in above with the frame work of
interval computation theory.


10/19/2004

SDL "Skunk Works" Project

Slide
-
22

Adaptive control


Q10: Infinite dimensional adaptive control. In case the
calibration is not possible, for example, a sensor like the
camera is not possible, can we design an adaptive
controller which does not require calibration at all? If yes,
what is the cost of performance? Or, we can assume the
calibration is not thoroughly, does the adaptive controller
help? The controller need to be adaptive to the (1) slowly
changing environment. (2) the assumed time
-
varying
characteristics of each PR.



10/19/2004

SDL "Skunk Works" Project

Slide
-
23

Logic+PDE


Q11:
Since the base station only communicate with
MASmote by high
-
level command, we need to merge
logic with PDE at this stage. How to find the minimum
set of logics that sufficient for low
-
level control? What is
the minimum sample rate?


PDE
-
>Logic


Base station

Logic
-
>ODE


MASmote

10/19/2004

SDL "Skunk Works" Project

Slide
-
24

Communication+real
-
time


Q12:


(1) Wireless communication collision avoidance. (bad
assumptions for CSMA/CA)


(2) The max bandwidth for asynchronies delay critical
communication.


(3) “wireless fieldbus” by CC1000

10/19/2004

SDL "Skunk Works" Project

Slide
-
25

Comm+Ad
-
hoc network


Q13:


Homogeneous vs. heterogeneous (multi
-
hop routers)


What is the proper ad
-
hoc network configuration for
the best communication performance


Routing algorithm: energy+speed



10/19/2004

SDL "Skunk Works" Project

Slide
-
26

Interdisciplinary modeling


Q14:


robot collusion/exception handling (FSM/DES) + fog
estimation (PDE) + robot inverse kinematics (ODE)


How about Petri
-
Net based model fusion?

10/19/2004

SDL "Skunk Works" Project

Slide
-
27

Real
-
time code automation


Q15:


Like for Gitto, but consider asynchronies ad
-
hoc
network environment.


Simulate with “player and stage,” the result should be
close to the performance of real hardware platform.


10/19/2004

SDL "Skunk Works" Project

Slide
-
28

(High
-
level) control algorithm
automation


Q16: robocup scenario


Strategies learning (centralized or distributed)


Run time strategy update at MASmote, or flash
memory download update (using XNP)



10/19/2004

SDL "Skunk Works" Project

Slide
-
29

CSP+real
-
time


Q17


Port CSP from Java to nesC


Automatic (semi
-
automatic) dead lock, live lock
checking (one robot) for nesC


Automatic dead lock, live lock checking for
heterogeneous robot groups with nesC


How to cooperate CSP with ODE/PDE control laws?

10/19/2004

SDL "Skunk Works" Project

Slide
-
30

Fundamental limitations



Q18


Characterize the chaos/bifurcation properties of the
fog/air flow.


What is the limitation of observation?


What is the limitation of control?


Respect the instability?

10/19/2004

SDL "Skunk Works" Project

Slide
-
31

Ad
-
hoc network localization


Q19: rescue robot scenario


Unreliable indoor communication


Less cost sensors


Locate each robot by ad
-
hoc network.


Semi
-
3D localization.

10/19/2004

SDL "Skunk Works" Project

Slide
-
32

Regional analysis


Regional stability and stabilizability


Regional state observer design for DPS (parabolic)


Regional detectability


Regional gradient observer


See



A. El Jai and A. J. Pritchard,
Sensors and Actuators in Distributed Systems
Analysis
, Ellis Horwood Series in Applied Mathematics, Ellis Horwood, John
Wiley, Chichester, West Sussex: Ellis Horwood, 1988.


A. E. Jai, M. C. Simon, E. Zerrik, and A. J. Pritchard, ``
Regional controllability
of distributed parameter systems
,'' International Journal of Control, vol. 62,
1995.


M. Amourous, A. E. Jai, and E. Zerrik, ``
Regional observability of distributed
systems
,'' International Journal of Systems Sciences. vol. 25, 1994.

10/19/2004

SDL "Skunk Works" Project

Slide
-
33

Optimal policies


Sensing policy


Sensor scheduling


Motion planning


Actuation policy


Actuator scheduling


Motion planning


Collaborative sensing


Collective actuation

10/19/2004

SDL "Skunk Works" Project

Slide
-
34

Research Output so far (08/2003
-
10/2004)


Papers published:


Kevin L. Moore*, YangQuan Chen, and Zhen Song. "
Diffusion
-
based path planning in mobile actuator
-
sensor networks (MAS
-
net): some preliminary results
". INTELLIGENT COMPUTING:
THEORY AND APPLICATIONS II (OR53).
SPIE Defense and
Security Symposium 2004.

April 12
-
16, 2004, Gaylord Palms
Resort and Convention Center, Orlando, FL, USA. (PDF)
SPIE5421
-
08. (PDF)


YangQuan Chen*, Kevin L. Moore, and Zhen Song. "
Diffusion
boundary and zone control via mobile actuator
-
sensor networks
(MAS
-
net): challenges and opportunities
." INTELLIGENT
COMPUTING: THEORY AND APPLICATIONS II (OR53).
SPIE Defense and Security Symposium 2004.

April 12
-
16,
2004, Gaylord Palms Resort and Convention Center, Orlando,
FL, USA. (PDF) SPIE5421
-
12. (PDF)

10/19/2004

SDL "Skunk Works" Project

Slide
-
35

Papers published (continued)


Zhongmin Wang, Zhen Song, Peng
-
Yu Chen, Anisha Arora,
Kevin L. Moore and YangQuan Chen. "
MASmote
--

A Mobility
Node for MAS
-
net (Mobile Actuator Sensor Networks)".

IEEE
Int. Conf. on Robotics and Biomimetics (RoBio04),

August
22
-
25, Shengyang, China. (PDF
-
robio2004
-
330)


Kevin L. Moore* and YangQuan Chen. "
MODEL
-
BASED
APPROACH TO CHARACTERIZATION OF DIFFUSION
PROCESSES VIA DISTRIBUTED CONTROL OF ACTUATED
SENSOR NETWORKS
".
The 1st IFAC Symposium on
Telematics Applications in Automation and Robotics.

Helsinki University of Technology Espoo, Finland, 21
-
23 June
2004.


10/19/2004

SDL "Skunk Works" Project

Slide
-
36

Papers submitted.


Zhongmin Wang, Zhen Song, Peng
-
Yu Chen,
YangQuan Chen and Kevin L. Moore. "
Formation
motion control methods in mobile actuator/sensor
networks
"
SPIE Defense and Security Symposium
2005.

April 2005


Zhen Song, Pengyu Chen, Zhongmin Wang, Anisha
Arora, Yangquan Chen. “
MAS
-
net: a Mobile Actuator
-
Sensor Network System for Diffusion Observation and
Control
”,
IEEE Communication Magazine
.

10/19/2004

SDL "Skunk Works" Project

Slide
-
37

Others


Establish a world reputation in sensor
-
networks
with a strong “
control
”/“
closed
-
loop
” flavor


IEEE/RSJ Int. Conf. on Intelligent Robotics and
Systems. (www.IROS2005.org)


Member, Organizing Committee, Invited Session co
-
Chair


Plan:
to organize a tutorial workshop on “
Task Oriented
Mobile Actuator and Sensor Networks
” at IROS2005 with
other leading players in the field (under planning, going well
so far, workshop proposal due March 1, 2005)

10/19/2004

SDL "Skunk Works" Project

Slide
-
38

Invited Talk


08/17/2004. “
Mobile actuator and sensor
networks for diffusion boundary determination
and zone control
”, Invited talk (75 minutes) at the
Institute of Intelligent Machines of Chinese
Academy of Sciences (IIM of CAS) in Hefei, the
capital city of Anhui Province, China.

10/19/2004

SDL "Skunk Works" Project

Slide
-
39

In 2 years


CSOIS is the earliest to initiate the research on MAS
-
net.
So
far, MAS
-
net is still unique and novel.


CSOIS will still be the leader in this field, specifically:


Distributed control of distributed parameter systems using
networked moving sensors and moving actuators


Dynamic boundary determination/tracking and zonal control using
networked moving sensors and moving actuators


Regional observation and state reconstruction with networked
moving sensors and active formation sampling


… my PhD students are working hard on the above theoretical and
practical problems.

10/19/2004

SDL "Skunk Works" Project

Slide
-
40

Mind
-
Storming Session


Demos so far show that


pGPS working (yes but)


Issues: optimal patterns? Not systematic designs (Lili). Orientation/position accuracy,
balanced accuracy?
Better lens
-

$200?


LLC servo algorithms (reliable but not accurate)


Issue: position loop only. Encoder resolution: 32 sectors. Dan is trying 128. Anisha: better
servo motor (with minimum changes, 10/31)


Deadzone, quantitative result? (Stiction + PW)


Data logging, w/time stamp (send in batch, not on
-
the
-
fly)


Saturation


(but, integral, we need AW)


LFFC helps on servo calibration


(systematic, deterministic, recurrent)


(in need: more
automatic calibration procedure). Think about “recalibration state/on demand”.


IR (working)


Issues: consistency? In need:
characterization

and then autocalibration.


PR (no big confidence now)


Issues: ibid.
Use pGPS to help on the calibration.

Or,
use gray
-
level template
. Or buy
better PRs (?)


GUI commands robots (kind of joystickable)


Issues: Real
-
time grouping, formation nicely. Calibration command (servo, IR, PR), Data
Logging etc.
Characterization tools
.

10/19/2004

SDL "Skunk Works" Project

Slide
-
41

MAS
-
net Tasks (Demo Scenarios)


Basic Behaviors


Obstacle/collision avoidance, E
-
stop, tracing behavior


Collective Behaviors (for what?)


Leader
-
follower, VIP/BG (pattern formation, either
static or dynamic


“collective tracing behavior”),
formation movement (regulation vs. tracking), …


Task
-
Oriented Behaviors


Adaptive spatial sampling, (Anisha: spatial sampling)

10/19/2004

SDL "Skunk Works" Project

Slide
-
42

Task
-
Oriented Behaviors



Distributed Measurement for Distributed Control
” and/or “
Distributed Control
for Distributed Measurement
”?




Distributed Control for Distributed Measurement
”!


? Scenarios: Think about this.


Scanning sensor problem in DPS (groups)


Periodic scanning sensor problem in DPS (groups)






10/19/2004

SDL "Skunk Works" Project

Slide
-
43

Task force


Anisha: spatial sampling (open loop)


Peng
-
Yu: pattern formation (static and dynamic)


Zhongmin: formation movement (regulatory and
tracking),


Zhen Song: DPS measurement, system ID and state re
-
construction using networking mobile sensors.


Jinsong: DPS with (networked!) moving sensors and
moving actuator. (1D and 2D simulation platforms)


Hyosung: TBD.