autonomous vtol for avalanche buried searching - avionics (slides)

Outline Introduction to Mountain Rescue Drone Avionics Design of a Digital ARTVA Simulations and Conclusions Q&A

Autonomous VTOL for Avalanche Buried Searching

AvionicsMatteo Ragni

Ingegneria Meccatronica Robotica


Introduction to Mountain Rescue

Drone Avionics

Design of a Digital ARTVA

Simulations and Conclusions


Mountain Rescue Intervention

I Call from witnesses or hikers indanger

I Helicopter missionI Evaluation of critical riskI Searching on avalanche surfaceI Searching for ARTVA signal presenceI Fine ARTVA searchingI Buried extraction

2. Starting Point

Buried5. Pinpointing a victim: ~2min

3. Searching for a signal

4. Signal found

1. Helicopter drops the rescue team

0 50 100 150

20

40

60

80

100

Time (min)

Cha

nces

ofsu

rviv

al(%

)


ARTVA Beacons Overview

I A1A Signal:

I amplitude modulateddigital signal

I one carrierfrequency: 457kHz

I frequency error±80Hz

I H–field peak at 10m

I ≥ 0.5 µA m−1

I ≤ 2.23 µA m−1Time

x

Inte

llige

nce

0

1

y

≥ 70ms ≥ 400ms

1000± 300ms

Triple

Antennas

Frequency shift

Anti–alias filter

A–D

Conversion

Digital FilterSignal

Detection

H–field

Estimation

Analog

Digital

TX MODE

RX MODE


H–Field in Transmission

Field Complexity

; ;

Simplified Equations for H–field

B(r, m) =µ0

4πr5

2x2 − y2 − z2 3xy 3xz

3xy 2y2 − x2 − z2 3yz

3xz 3yz 2z2 − x2 − y2



Drone Avionics




Perception–Action Map

Litterature overview...

Model

Hypothesis

Emulator

Grounding

Environment

Agent

I Subsumption and groundingI Emulation

... applied to our agent

Perception

Dynamics and control

Tracking Problem

Obstacle Avoidance

Altitude Keeping

Sourcesearching

Emulation

Radar detect

Explo-ration

routines

Action


Dynamics, control and tracking

LQR Control

Li =mg6

x = f (x, u) 1s

u x

xfK −

−

u∗ e

Newton–Euler Equations

xg

yg

zgxb

yb

zb LiMi

π/3 x = [x, y, z, φ, θ, ψ, u, v, w, p, q, r]T

u = [Li : i = 1..6]

x = f (x, u)


Obstacle Avoidance

ui

v(di)

v

di

obstacle

v = R(φ, ψ, θ)6∑

i=1v(di)

cos

((i− 1)

π

3

)− sin

((i− 1)

π

3

)0

I Advantages

I low computation neededI minor constraint on upper layersI fit QFD constraints

I Drawbacks

I non–optimal pathsI limited reliability

Speed function example:

v(di) = p3

(1

1 + e4(

p12 −di

)p2p3

− 1

)


Altitude Keeping

Identification of the surface normal m −→ S.L.A.M. Problem

x

m

mt-1

mt

h

A

C B

mt =(A− B)× (B− C)|(A− B)× (B− C)|

Keep the VTOL at costant distance h along exstimated plane normal mt


Exploring and Searching Signal Presence

Explore the surface, starting from point p0, to the point pn

p0

pn

Plane dire

ction

Receiver range

We need a strategy to understand if there is a signal


Radar Detection Problem for Signal Presence

Signal Source

Z0Z1

Z

p(s|H1)p(s|H0)

Z0 Z1

s

p(s)p(s|H1)

p(s|H0)

← s→

PMPD

Z0 Z1

s

p(s)p(s|H1)

p(s|H0)

PCPF

Minimize the risk incurred due to erroneous decisions

min R = R(ci,j, PX) →Z0 = s ∈ Z : ∆(s) < η

Z1 = s ∈ Z : ∆(s) > η


Pinpointing Signal Source

Searching the Maximum H–field

H∇H

|H| cos θ

|H| sin θ

θ

vPrevious

knowledge

ψ

|v|

cos θ

∇Hsin θ

|H|

Emulation of an H–field

And for multiple burials?

The stimated position is given by thesolution of the optimization problem:

min δ =(H−H(pt, m, x)

)2

(pT − x)2 ≤ rmax

and treated as a stochastic variable

p(p) =1N

N∑

k=1

γ(p− pk, h)V(h)

from p(p) we extract mean andcovariance!



Drone Avionics





General Overview

Power supply

Tuned tank Filter stage Identification Filter stage

ADC

Triple antennas

x

y

z

Analogstage

Power supply

Analogstage

Analogstage

Digitalstage

Ferrite rod

Loop solenoid

Preamplifier

Amplifier

Identification

Tune

d Ta

nk


Schematics – Antenna and PreAmplifier

103 104 105 106 107 108−150

−100

−50

0

50

Frequency (Hz)

Mag

nitu

de(d

B)

PreAmplifier Characteristic


Schematics – Identification and Amplifier

100 101 102 103 104−150

−100

−50

0

Frequency (Hz)

Mag

nitu

de(d

B)

Amplifier Characteristic



Drone Avionics




Simulation Results (1)

0

5

10

15

20

−30

−25

−20

−15

−10

−5

0

5

0

5

10

x(m)

y(m)

z(m)

0 20 40 60 80 100 120 140−30

−20

−10

0

10

20

Time (s)

Posi

tion

(m)

x y z

0 20 40 60 80 100 120 140

0

1

2

3

Time (s)

Att

itud

e(r

ad)

φ θ ψ

0 20 40 60 80 100 120 140−1

0

1

2

3

Time (s)

Velo

city

(m/s

)

u v w

0 20 40 60 80 100 120 140

−4

−2

0

Time (s)

Ang

ular

rate

(rad

/s)

p q r

0 20 40 60 80 100 120 140

−1

−0.5

0

0.5

1

Time (s)

Lear

ned

orie

ntat

ion

cos(θ) cos(θ) real sin(θ) sin(θ) real

0 20 40 60 80 100 120 14010−6

10−5

10−4

10−3

Time (s)

Lear

ned

inte

nsit

y(A

/m)

|H| |H| real

Position found in 110s


Simulation Results (2)

−30 −20 −10 0 10 20 30 40 50 60−60

−50

−40

−30

−20

−10

0

10

20

30

x(m)

y(m)

Drone positionTransmitter positionOptimization resultsLatest optimization resultsObstacle

0.00 0.05 0.10 0.15−60

−50

−40

−30

−20

−10

0

10

20

30

p(ptx,y|H)

−30 −20 −10 0 10 20 30 40 50 600.00

0.05

0.10

0.15

p(p

tx,x|H

)

Further improvements: weight the solutionswith respect to time!

x

y

0 50 100 1500

2

4

6

Time (s)

Dis

tanc

ed 0

(m)

0

50

100

150

0

2

4

6

Tim

e(s

)

Distance dπ/3 (m)

0

50

100

150

0

2

4

6

Time

(s)

Distance d 2π/3

(m)050100150

0

2

4

6

Time (s)

Dis

tanc

ed π

(m)

0

50

100

150

0

2

4

6

Tim

e(s

)

Distance d4π/3 (m)

0

50

100

150

0

2

4

6

Time

(s)

Distance d 5π/3

(m)


Conclusions

ARTVA

I A review of the ARTVA protocol is strongly advisedI The ferrite antennas must be carefully modeledI Move from analog devices to software–defined–radio for better performance

Avionics

I Perception–Action map fits our problem requirementI A wiser emulator should be defined, with time related weightsI Performance can be improved by augmenting perception


Questions?


State of the Art

Projects

I SHERPA: Universita di BolognaI Universita di TorinoI Project Alcedo Eidgenossische Technische Hochschule Zurich

Digital searching algorithms

I H–Field Lobe Following and pinpointingI Fast identification with SLAM and sum of Gaussian


Maxwell Formulation

Application of potential vectors and recalibration map to Maxwell’s eq.

∇ · B = 0

∇× E = − ∂

∂tB

∇ · E =ρ

ε0

∇× B = µ0

(J + ε0

∂

∂tE)

∇2φ− 1

c2∂2φ

∂t2 = − ρ

ε0

∇2A− 1c2

∂2A∂t2 = −µ0J

B = ∇×A

E = −∇φ− ∂A∂t

A′ 7→ A +∇ψ

φ′ 7→ φ− ∂ψ

∂t

∇ ·A′ = − 1c2

∂2ψ′

∂t2

Application to our problem: integral formulation

φ(r, t) =1

4πε0

∫Ω

1|r− r′|ρ

(r′, t− |r− r′|

c

)dr

A(r, t) =µ0

4π

∫Ω

1|r− r′| J

(r′, t− |r− r′|

c

)dr


Magnetic dipole problem

For a magnetic dipole problem: φ = 0!

Solution for boundary condition problem

xy

z

ϕ′

J

dr

r

κ = |r− r′|

r′

θψ A =

µ0m0

4πrsin(θ)

(1r

sin (ω0(t− r/c))−

+ω0

rcos (ω0(t− r/c))

)φ

Under the hypothesis: r′ r and r′ λ

B–Field solution

τ = t− rc

Br =µ0m0

2πr2 cos(θ)(

1r

cos(ω0τ)− ω0

csin(ω0τ)

)Br =

µ0m0

4πr3csin(θ)

((c2 −ω2

0r2) cos(ω0τ)−ω0rc sin(ω0τ))


Simulating a Range Finder

√d2

i − u2

d i=|x

Ψi− x d|

ρ

hu

(xΨi − xd)

ui

ρ (maximum radius)

u = (xΨi − xd) · ui

h (maximum range)

Characteristic lobe


Simulink Implementations (1)

Hexacopter Model

Parameters

+Fi =mg6

LQR Controller 1s

z attitude

+SearchingAlgorithm

ObstacleAvoiding

x

x Range Finder Model di

Ψ = [xi : i = 1..M]

[h, ρ]

RT(φ, θ, ψ)

vb =6∑

i=1v(di)ui × v

[p1, p2, p3]


Simulink Implementations (2)

x H sensor

Magnetic Dipole m

TX position pT

|H|

cos(θ)

sin(θ)

α1s + 1β1s2 + β2s + 1

Explo-ration

directionv

Emulation(H− H)2 = 0

Optimized pT

Optimized m

Parameters

x H (x, pT, m)

Magnetic Dipole m

TX position pT

|H|

×N (0, Σ) SNR

+ H

autonomous vtol for avalanche buried searching - avionics (slides)

Documents

digital artva simulations

digital artvasimulations

rescue team0

conclusions qaintroduction

uoutline introduction

conclusions qahfield

conclusions qadynamics

conclusions qaautonomous