sensor driven online coverage planning for autonomous...

Introduction Background Proposed Methods Experimental Setup Results Conclusion

Sensor Driven Online Coverage Planning forAutonomous Underwater Vehicles

Liam Paull1, Sajad Saeedi Gharahbolagh1, Mae Seto2, HowardLi1

1 Collaboration Based Robotics and Automation (COBRA)Group in the Department of Electrical and ComputerEngineering at the University of New Brunswick.

2 Mine and Harbour Defense Group at Defense R&DCanada-Atlantic.

October 9, 2012

Paull et al. Sensor Driven Online Coverage Planning for AUVs


Outline

1 Introduction

2 Background

3 Proposed Methods

4 Experimental Setup

5 Results

6 Conclusion



IntroductionUnderwater Mine Countermeasures

Mine Countermeasures

AUVs have manyadvantages for counteringundersea threats, such as:

Increased covertness

Reduced number ofpersonnel required

Safety of qualifiedpersonnel

Increased efficiency

Moored Mine



Autonomous Underwater Vehicles (AUVs)

Autonomous Underwater Vehicles

Autonomous underwatervehicle (AUV) researchbegan in the 1970s

AUVs are used for surveying,mine countermeasures(MCM), bathymetric datacollection and more

Challenges: No GPS,communication verychallenging, environmentquite unstructured

The IVER2 AUV



Sidescan Sonar Sensor

Sidescan Sonar Sensor

Lateral

RangeAltitude

AUV T

rajector

y

Data

Target

P(y) Characteristic Curves

0 10 20 30 40 50 60 700.5

0.6

0.7

0.8

0.9

1

Lateral Range (m)

Pro

babi

lity

of D

etec

tion

Sand 10m DepthClay 10m DepthCobble 10m Depth



Objective of Work

Current AUV seabed survey plans are generated manually byan operator before the start of the survey.

These plans are usually highly structured (“lawn mower or zigzag”)

The performance of the sonar used for seabed survey is highlydependent on many parameters that are not necessarily knownbeforehand but can often be measured in situ.

Can we develop survey planning strategies that do not requireoffline survey plans to be generated, that aren’t restricted tothe structured paths, and can adapt to parameters measuredon the fly.



Outline

1 Introduction

2 Background

3 Proposed Methods


5 Results

6 Conclusion



Path Planning and Coverage Path Planning

Free Configuration space (Cfree): The set of all validconfigurations that the robot can achieveWorkspace (W ): The world that the robot exists in

General Path Planning

τ : [0, 1] → Cfree

τ(0) = xi , τ(1) = xg

Tasks: Navigation, coverage,localization, mapping

Past Approaches: Bug,potential fields, celldecomposition,sampling-based...

Coverage Path Planning

N sensor readings: {A1, ...,AN}

N⋃i=1

Ai ⊇ W

Heuristic methods

Cell decomposition



Information Theory

The Shannon Entropy of an RV X : H(X ) = E [logP(X )]The Expected Conditional Entropy of an RV X given Z :

H(X |Z ) = Ez{H(X |Z )}

= −

∫P(Z )

∫P(X |Z ) logP(X |Z )dXdZ .

(1)

The mutual information I or expected entropy reduction

(EER):I (X ,Z ) = H(X ) − H(X |Z ), (2)

The information gain B of a control action U that will result in n

independent measurements {Z1,Z2, ...,Zn}:

B(U) =

n∑k=1

I (X ,Zk). (3)



Outline

1 Introduction

2 Background

3 Proposed Methods


5 Results

6 Conclusion



Multi-Objective Function

The backbone of the proposed approach is an objective functionthat is evaluated over the domain of all possible desired headings:ψ = {0..360}:

ψd = argmaxψ

R(ψ) = wBB(ψ) + wGG (ψ) + wJJ(ψ), (4)

Where:

R is the total utility

B is the information gain

G is the branch entropy

J is the benefit of maintaining the current heading(∝ −|ψc − ψ|, ψc is current heading)

wB , wG , and wJ are the weights tuned manually or with somemetaheuristic method



Information Gain Behavior

Define a proposed track starting from the AUVs current location(x , y) and extending out a distance r and angle ψ:

C : [0, 1] → Cfree , s → C(s)

C(0) = (x , y)

C(1) = (x + r cos(ψd ), y + r sin(ψ))

(5)

Then evaluate the expected information of the paths:

H(Tij |Zijk ) = Ezk{H(Tij |Z

ijk )} (6)

I (Tij ,Zijk ) = H(Tij )− H(Tij |Z

ijk ) (7)

I (W ,Zk) =∑

(i ,j)on C⊥

I (Tij ,Zijk ) (8)

B(ψ) =

n∑k=1

I (W ,Zk) (9)



The Branch Entropy BehaviorOverview

Behaviour Objectives

finish sections beforeit leaves them.

find the areas of theworkspace that arenot covered.

not get stuck ininfinite loops.

Algorithm Flowchart

Compute Branch Entropy for Each Neighbour of Current Cell

Generate Directed Acyclic Graph

Perform Exact Hexagon Decomposition

Compute Objective Function

W, H(W)

G(ψd)



The Branch Entropy BehaviorThe Hexagon Decomposition

1

{0,1,2}

1

{0,1,5}

1

{0,4,5}

1

{3,5}

1

{2,4}

1

{1,2,3}

2

{1,2}

2

{1,2,3}

3

{}

2

{}2

{1,2}

3

{1}

4

{}

3

{1,2}

4

{}

3

{2}

2

{0,1,2}

3

{0}

3

{1}

4

{}

2

{0,1}

3

{}

2

{0,1,5}

3

{}

3

{}

2

{0,5}

3

{}

2

{0}

2

{}

Level

{Children}

0

1

2

3

4

5



The Branch Entropy BehaviorHexagon Decomposition → Directed Acyclic Graph

Cp

0.6

0.5

0.1

0.2

0.95

0.9

Cp

0.6 0.5 0.2

0.1 0.9 0.95

g4 g3 g2

0.6 0.5

g3

0.1

g3g3

0.2

0.9 0.95



Calculating the Branch Entropies

Cp

0.6

0.5

0.1

0.2

0.95

0.9

Cp

0.6 0.5 0.2

0.1 0.9 0.95

g4 g3 g2

0.6 0.5

g3

0.1

g3g3

0.2

0.9 0.95

The Branch Entropy Equation

gk =

L∑l=2

(L− l + 1)

mlk∑i=1

Hi

mlk

L−1∑l=1

l

Simple Example

g4 = 1/3((2)(0.6) + (1)(0.1))

= 0.433,

g3 = 1/3((2)(0.5) + (1)(0.1))

= 0.367,

g2 = 1/3((2)(0.2)

+ (1)(1/2)(0.95 + 0.90))

= 0.442.



The Combined Objective Functions

Simulated Path

Multi-Objective Optimization

0 50 100 150 200 250 300 3500

20

40

60

80

100

120

140

160

180

200

Desired Heading (Degrees)

Util

ity

Expected Information GainBranch EntropyMaintain HeadingCollectiveBest Heading

= 94o



Outline

1 Introduction

2 Background

3 Proposed Methods


5 Results

6 Conclusion



Experimental Setup

DVLINS

Sensor Drivers

Front Seat / Inner Loop Control

Back Seat /Outer Loop Control

Navigation &Localization

ActuatorDrivers

DC/ServoMotors

Base Station

OtherAUVs

AUV Hardware orDynamics Simulator(HWIL)

High LevelPlanning

AUV

Autonomy

Water

Base

Acoustic

Sonar

Mine Detection

Acoustic

GPS Compass



Hardware Trials



Outline

1 Introduction

2 Background

3 Proposed Methods


5 Results

6 Conclusion



Hardware Trials

0 50 100 1500

50

100

150

200

250

Meters

Meters



Hardware Trials

0 50 100 150 200

0

50

100

150

200

250

300

Meters

Meters



Outline

1 Introduction

2 Background

3 Proposed Methods


5 Results

6 Conclusion



Advantages of Proposed Approach

The proposed approach has the advantages that:

1 The total paths and times required to cover a workspace areshorter in many cases.

2 There is no need for pre-programmed waypoints.

3 The AUV will maintain heading for better data mosaicing inthe presence of currents or erratic waypoint tracking behaviorcaused by poor navigation or controller performance.

4 It is adaptive to any changes in environmental conditions thatcan be detected in situ.

5 It is able to generate paths for complex and non-convexenvironment shapes such as would typically found in harbours.

6 Fast and scales linearly with environment size (afterinitialization)



Thank you!

This research is supported by Natural Sciences and EngineeringResearch Council of Canada (NSERC) and Defense R&D Canada -Atlantic.


sensor driven online coverage planning for autonomous...

Documents