demining robotics case study - metr 4202 advanced control & robotics - l13-casestudy.v2

7/29/2019 Demining Robotics Case Study - METR 4202 Advanced Control & Robotics - L13-CaseStudy.v2

1/28

1

Demining Robotics Case Study

METR 4202: Advanced Control & Robotics

Drs Surya Singh, Paul Pounds, and Hanna KurniawatiLecture # 13 Part 2 October 22, 2012

[email protected]

http://itee.uq.edu.au/~metr4202/ 2012 School of Information Technology and Electrical Engineering at the University of Queensland

RoboticsCourseWare Contributor

First Off: Bang Bang Control!

22/ 10/ 2012 -METR 4202 2


2/282

Smart for one, dumb for all

22/ 10/ 2012 -METR 4202 3

Landmines:Worldwide Problem

22/ 10/ 2012 -METR 4202 4


3/283

Gryphon: Mine Scanning Robot

22/ 10/ 2012 -METR 4202 5

Generalized Mine & Placement

22/ 10/ 2012 -METR 4202 6


4/284

(Antipersonnel) Landmines are Challenging

Terrain Diversity

Counter-thwart mechanisms False Positive Rate >100:1

Variable & Changing

22/ 10/ 2012 -METR 4202 7

Ex: PMN-2 [1]

22/ 10/ 2012 -METR 4202 8


5/285

Ex: PMN-2 [2]

22/ 10/ 2012 -METR 4202 9

Ex: PMN-2 [3]: Mechanically Intricate

22/ 10/ 2012 -METR 4202 10


6/286

Land Mines: Highly Variable

Little metal

High-sensitivitydetectors / instruments

Highly Variable(Example: PMN-2):

3-stage detonation

Anti-thwart

All mechanical

Poor constructiondetectors / instruments

Focus on manipulatingsensor instead ofcomplex sensing ???,,,

22/ 10/ 2012 -METR 4202 11

Clearance & Breeching

Breeching: Line

Demining: Area

International Mine Action Standards (IMAS)

22/ 10/ 2012 -METR 4202 12


7/287

Humanitarian Demining Process

22/ 10/ 2012 -METR 4202 13

Humanitarian Demining Process

1. Level 1: Specification

Rough minefield location

2. Level 2: Clearing

Heavy machines(e.g., flails, grinders, rollers, ploughs, and sifters)

~ 90% clearance

3. Level 3: Confirmation

22/ 10/ 2012 -METR 4202 14


8/288

Level-3: Confirmation

22/ 10/ 2012 -METR 4202 15

Sensor Mobility Is Critical

22/ 10/ 2012 -METR 4202 16


9/289

Back to Gyrphon

22/ 10/ 2012 -METR 4202 17

Cleared area

Minefield

All terrain vehicle

Counter-weight

Stereo vision camera

Optional ground-penetrating radar

Metaldetector

Network camera

22/ 10/ 2012 -METR 4202 18


10/2810

Gryphon Schematic

Wrist joints

z

x

Manipulator

SensorTerrain

Camera

Ground frame FG

Joint1 (yaw)

Joint2

Joint3

Compliant base

Manipulator frame FMCounter-weight

22/ 10/ 2012 -METR 4202 19

Gryphon: Comparison to other tracked robots

Mechanical Robustness

ControlRobustness(Au

tonomy)

22/ 10/ 2012 -METR 4202 20


11/2811

Demining Terrain Not Always Rough

Unexploded BLU-97

22/ 10/ 2012 -METR 4202 21

Remote Terrain: 30% > 30 Grade

Source: US Army, Logistcal Vehicle Off-Road Mobility, 1967 (c/o Prof. K. Waldron)

22/ 10/ 2012 -METR 4202 22


12/2812

Multiple Inaccuracies

Sensing:

ATV Suspension:

Screenshots courtesy M. Freese, Tokyo Tech.

22/ 10/ 2012 -METR 4202 23

Operational Overview

Heightmap DetectingTerrain

scanning

Screenshots courtesy M. Freese, Tokyo Tech.

22/ 10/ 2012 -METR 4202 24


13/2813

Terrain Modeling & Following Overview I. Terrain Mapping II. Terrain Model

III. Path Generation IV. Scanning

Conditional PlannerFiltering & Stitching

Height map expansion& Path generation

Input-shaping

V. Evaluation& Marking

22/ 10/ 2012 -METR 4202 25

Terrain Mapping

Stereo depth maps (Pont Gray Bumblebee)

Kinematic calibration corrections

Ex1: Grassy areaEx2: Grassy areawith hill or bump

22/ 10/ 2012 -METR 4202 26


14/2814

Terrain Geometry Model: Heightmap Expansion

Scanning gap

x

z

x

z

As Surface Normals:

Raw Map Raw

Model

Filtered

ModelExpanded

and Offset

22/ 10/ 2012 -METR 4202 27

Terrain Geometry Model:

Conditional Planar Filter

Planarity: Found from plane eq. residuals for a surface

patch

Filter type and strength varied based on this

Goal: Reduce noise without feature degradation

Planarity measure (for local patch)

Gaussian-smoothed value

Median value *

*

+1

+Filtered

value

22/ 10/ 2012 -METR 4202 28


15/2815

Terrain MapModel: Conditional Planar Filter

ComputeNormals

Map:

TerrainMesh

Model

Applyfilter(s)

22/ 10/ 2012 -METR 4202 29

MapModel (II): Height Map Expansion

Envelope expansion:

Fenv = Fterr + scanning gap

Performed along the normals, more than vertical axis

addition:

Scanning gap d

22/ 10/ 2012 -METR 4202 30


16/2816

MapModel (III): Height Map Expansion

Terrainenvelope

Model of

terrainScanning gap

22/ 10/ 2012 -METR 4202 31

xz

y

Calibration Model

Height (z) Calibration:

Plane (x-y) Calibration

22/ 10/ 2012 -METR 4202 32


17/2817

Effect of Overall Calibration Matrix

-10

-5

0

5

10

15

20

25

30

0 0.5 1 1.5

Circular distance [m]

Deviation

from

idealpath

[mm]

With Ove rall Calibration Matrix

correction

Without Overall Calibration Matrix

correction

Scanning speed: 100 mm/s

Scanning gap: 100 mm

22/ 10/ 2012 -METR 4202 33

Path Generation

x-y: Scanning Scheme

Joint-space/Work-space?

Reduce excess work

z: Terrain Sampling (z)

Sample corresponding

point based on the local

patch & normals

ATV

Manipulator Detectorzpath fenv(xpath,ypath )

22/ 10/ 2012 -METR 4202 34


18/2818

Path Generation (II)

Orientation: Advanced Terrain Following

(a) (b) (c)

(d) (e)

Controlpoints

22/ 10/ 2012 -METR 4202 35

Contour Following

DetectorEnvelope

Terrain

Potential collisions

22/ 10/ 2012 -METR 4202 36


19/2819

Robust Control:

Command Shaping for Vibration Reduction

22/ 10/ 2012 -METR 4202 37

Command Shaping

Original velocity profile

Input shaper

Command-shaped velocity profile

Velocity

Velocity

Time

Time

Time*

22/ 10/ 2012 -METR 4202 38


20/2820

Command Shaping in Position Space

22/ 10/ 2012 -METR 4202 39

Command Shaping:

Zero Vibration and Derivative

2,1i

For Gryphon: At0=1.5 [m] At1=3.0 [m]

2.32 1.81Axis 1

0 0

3.3 3.0Axis 2 & 3

0 0

22/ 10/ 2012 -METR 4202 40


21/2821

Terrain Modeling:Find a good model to characterize

22/ 10/ 2012 -METR 4202 41

Scanning on ~ Level Terrain - Measurements

-10

-5

0

5

10

15

20

0 0.5 1 1.5


Deviation

from

idealpath

[mm]

Unfiltered

Gaussian filtered

Conditional Planar filtered



Manipulator

Laser range

finder

Scan pass

22/ 10/ 2012 -METR 4202 42


22/2822

Scanning on Rough Terrain - Measurements

-40

-30

-20

-10

0

10

20

30

0 0.5 1 1.5


Deviation

from

idealpath

[mm]

Unfiltered

Gaussian filtered

Conditional Planar filtered

Obstacle location

50cm

20cm

Scan passRough terrain

obstacle

~70 slope

Manipulator

Laser range

finder



22/ 10/ 2012 -METR 4202 43

Command Shaping Tests: Step-Response

Reduced Joint

Encoder Vibration Reduced Tip

AccelerationJoint 1 (ATV Yaw) Encoder:

Joint 3 (Arm Extend) Encoder:

22/ 10/ 2012 -METR 4202 44


23/2823

High-Level Control Software

22/ 10/ 2012 -METR 4202 45

Detector Imaging

Targets

Imaging

Target# Target type Depth [cm] MD GPR

1 PMA-2 5 Yes No

2 PMA-1A 12.5 Yes Yes



5 Fragment 5 Yes No

6 Stone ~10 No Yes

PMA-2PMA-1A Fragment

22/ 10/ 2012 -METR 4202 46


24/2824

Extensive Field Tests

2005: Kagawa, Japan

2007: Siem Reap, Cambod ia

2006: Benkovac, Croatia

22/ 10/ 2012 -METR 4202 47

Gryphon: Field Tests in Croatia & Cambodia

22/ 10/ 2012 -METR 4202 48


25/2825

February 2006: Tests in Croatia

Metal detector imaging

22/ 10/ 2012 -METR 4202 49

Terrain & Estimation

IF we know terrain Triangulation

IF we know depth SNR gives terrain characteristic

Estimate both simultaneously ( solution up to scale)

Detector

Terrain

t0 ti

Target

22/ 10/ 2012 -METR 4202 50


26/2826

End on a Bang, Bang

22/ 10/ 2012 -METR 4202 51

A Better (Controlled) Bang Bang

22/ 10/ 2012 -METR 4202 52


27/2827

Robotics, State-Space and The Meaning of Life

22/ 10/ 2012 -METR 4202 53

Seeing is forgetting the name of the thing one sees

What is it?

A Coca Cola Can

1122q3134234232342

(SIFT descriptor in HEX)

(The PHOTID) (The Filename)

22/ 10/ 2012 -METR 4202 54


28/28

Robotics & Uncertainty

University:

Critical questioning

Open exchange of ideas

Among equal participants.

Corporatocracy:

Ruled by control freaks Behavior modification

works best on dependent,

powerless, infantilized, boredand institutionalized people(which is almost inimical to )

Be careful of being overworked

Et Tu?

There are peddlers of

uncertainty

They want you to be

afraid

Thus, they sell fear

Yes, They live for your

schadenfreude

Insurance lets you riskand balance uncertainty

22/ 10/ 2012 -METR 4202 55

UQ Robotics: Dynamic Systems in Motion

PlanningMechanics

of motion

Aerial Systems

Bio-inspired

Systems

Hanna Kurniwati

(NUS/MIT)

Paul Pounds

(ANU/Yale)Surya Singh

(Stanford/Syd)

Diverse international

research group

22/ 10/ 2012 -METR 4202 56

demining robotics case study - metr 4202 advanced control & robotics - l13-casestudy.v2

Documents