Sky Savers: Leveraging Drone Technology for Victim Localization

in Avalanche Rescue via Transceiver Signal Analysis

Robin Vetsch

1,3 a

, Samuel Kranz

2 b

, Tindaro Pittorino

2 c

, Peter de Baets

, Martial Ch

ateauvieux

Christoph W

ursch

3 d

, Daniel Lenz

and Sebastien Gros

1 e

Department of Engineering Cybernetics, Norwegian University of Science and Technology, Høgskoleringen 1, Trondheim,

Norway

Institute for Electronics, Sensors and Actuators, Eastern Switzerland University of Applied Sciences, Werdenbergstrasse 4,

Buchs, Switzerland

Institute for Computational Engineering, Eastern Switzerland University of Applied Sciences, Werdenbergstrasse 4, Buchs,

Switzerland

ANAVIA, Bahnhofstrasse 37, N

afels, Switzerland

Keywords:

Avalanche Rescue, Victim Localization, Vertical Take-Off and Landing (VTOL) System, Avalanche

Transceiver.

Abstract:

In modern avalanche rescues, the search for buried victims is carried out primarily using a state-of-the-art

handheld transceiver. However, in situations where the rescuers do not have the necessary experience, or

if the victims are buried in areas that can be dangerous for the rescuers, e.g. due to the risk of secondary

avalanches, this search process can be time-consuming, complex and dangerous. To overcome these chal-

lenges, we propose a proof-of-concept (PoC) of a search system based on an autonomous vertical take-off

and landing (VTOL) aircraft that could signiﬁcantly reduce search time, even in the case of multiple over-

lapping signal sources attributable to multiple victims or where conventional methods are not sufﬁciently

efﬁcient, e.g. in the case of large-scale avalanches. Electric drones or VTOL systems cannot be used because

electro-magnetic interference (EMI) blocks the signal from the sending avalanche transceiver. By replacing

electromagnetically noisy DC motors with a turbine, we effectively reduce electro-magnetic interference in

the signal stream and demonstrate sub-meter localization accuracy under realistic ﬁeld conditions. We employ

a two-stage Extended Kalman Filter (EKF) approach to estimate the stationary target coordinates. Eventually,

a VTOL system also allows for operations in adverse weather and rugged alpine terrain, greatly extending the

practical capability of search and rescue missions.

1 INTRODUCTION

On average, around 100 people die in avalanche ac-

cidents every winter in Europe (Association, 2025).

This high number of fatalities motivates the develop-

ment of an automated system to locate buried victims

as quickly as possible.

In modern avalanche rescues, the search for buried

victims is carried out primarily using commercially

available handheld avalanche transceivers (Mammut,

2024b). The clinical study conducted by Brugger

et al. (Brugger et al., 2007) showed how the use of

an avalanche transceiver or an airbag could reduce

https://orcid.org/0009-0007-8099-9457

https://orcid.org/0009-0006-6766-6458

https://orcid.org/0009-0005-7548-1130

https://orcid.org/0000-0002-1337-3477

https://orcid.org/0000-0001-6054-2133

the probability of mortality. Rescuers rely on these

transceivers to locate individuals by following the

strength and direction of the signals, moving closer to

the strongest signal to pinpoint the victim’s location.

Although this method can be effective, the rescuer

must be proﬁcient in using the device and interpret-

ing the visualizations on the graphical user interface

(GUI). In scenarios where rescuers have limited or no

experience, or when there are multiple victims buried

under the avalanche, the search process can become

time consuming and complex, increasing the chance

of deadly fatalities (Falk et al., 1994).

To address these challenges, we propose a VTOL

search platform designed for fully automated opera-

tion, capable of drastically reducing victim localiza-

tion times over vast avalanche terrains. With the com-

bination of slow ﬂight, vertical take-off and landing,

and fast cruise ﬂight to the accident site, our system

Vetsch, R., Kranz, S., Pittorino, T., de Baets, P., Châteauvieux, M., Würsch, C., Lenz, D. and Gros, S.

Sky Savers: Leveraging Drone Technology for Victim Localization in Avalanche Rescue via Transceiver Signal Analysis.

DOI: 10.5220/0013736400003982

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 22nd International Conference on Informatics in Control, Automation and Robotics (ICINCO 2025) - Volume 1, pages 469-476

ISBN: 978-989-758-770-2; ISSN: 2184-2809

469

can reach conﬁned and remote regions that are inac-

cessible to ﬁxed-wing aircraft or ground vehicles, thus

speeding up search operations in large areas.

As a sensor for localization, we used a commer-

cially available MAMMUT Barryvox Pulse avalanche

transceiver (Mammut, 2024a), which was integrated

as an external payload onto the VTOL system. Our

custom data analysis pipeline fuses the transceiver

output with real-time GPS and heading information

from the VTOL onboard navigation suite. As a

backup, the sensor system also had a GPS module on

board, which provided a data quality comparable to

the GPS of the VTOL system. As input signals for

the localization algorithm in this PoC, we used dis-

tance r and direction ϕ (in 15

◦

increments) data of

the transceiver device. This data is also available on

the display of the device and is used today for mate

searches by the user.

Figure 1: The proposed localization system (sensor sys-

tem) consists of a 3-axis magnetic ﬁeld sensor, an MAM-

MUT Barryvox Pulse (yellow), an onboard computer for

data collection (red) and a GPS module (blue). During

ﬂight, the VTOL system provides GPS and heading infor-

mation which is used to compare the position estimations of

the GPS module of the sensor system in blue.

1.1 Function of an Avalanche

Transceiver

An avalanche transceiver essentially consists of an ar-

rangement of three coils in the x, y and z directions, as

shown in (Ayuso et al., 2015). The device can operate

in two modes, send / transmit and search / receive.

In send mode, the coil in the x direction, which has

the highest sensitivity, and thanks to its size, gener-

ates a magnetic ﬁeld. In search mode, the magnetic

ﬁeld emitted by the transmitter is coupled to the re-

ceiver via the x- and y-antennas and processed on-

board. The z-antenna, which has the lowest sensitiv-

ity due to its limited length, is only used for the ﬁne

search. The ﬁne search is deﬁned as a speciﬁc search

pattern (r < 5 m) that is used to locate the person as

precisely as possible, before shoveling the person out.

In send mode, the device sends a pulsed electro-

magnetic signal that is generated with an oscillation

Figure 2: EN-normed pulse signal pattern including the

minimal and maximal time length of the on- and off pe-

riod. The pulse length and the off-time between two pulses

is used for the separation of multiple targets.

frequency of f

= 457 kHz and is approved worldwide

for avalanche rescue. The signal is pulsed at regu-

lar intervals deﬁned by the EN norm industry stan-

dard (ETSI, 2017), but with a unique repetition fre-

quency to distinguish multiple signals when several

transceivers transmit simultaneously. To minimize

power consumption, the frequency of the signal is

limited to approximately 1 Hz. The normed signal is

shown in Figure 2.

1.2 Related Work

Recent research efforts in the development of auto-

mated systems for avalanche rescue have increasingly

focused on the integration of sensor modalities into

unmanned aerial vehicles (UAVs), such as drones for

the rapid location of victims. However, commercial

drones often exhibit signiﬁcant drawbacks when op-

erating in extreme environments, particularly with re-

gard to electro-magnetic interference caused by con-

ventional electric motors, which adversely impacts

the performance of onboard avalanche transceivers.

In addition, lithium batteries used in electric drones

exhibit a loss of capacity in the cold winter weather,

signiﬁcantly limiting endurance.

A notable commercial initiative was the Powder-

Bee system, a lightweight (600g) battery-powered

drone designed for rapid search operations. Powder-

Bee was engineered to autonomously execute prede-

ﬁned search patterns and land upon detection of a

buried victim. Despite its innovative design, the com-

pany (Bluebird Mountain Inc.) was dissolved in 2021

and the EMI challenges inherent to its commercial

platform remained unresolved (Bluebird, 2021).

Similarly, the Alcedo project from ETH Zurich

represented an early student-driven effort to develop

a foldable and easily transportable UAV for avalanche

search and rescue. The system aimed to localize

the buried victims according to the dimensions of

the avalanche supplied by the user. However, the

project did not progress beyond the prototype stages

ICINCO 2025 - 22nd International Conference on Informatics in Control, Automation and Robotics

470

due to robustness limitations and EMI issues (Alcedo,

2010).

The start-up Nivitec pursued an approach using

a DJI Matrice 210 drone (DJI, 2025) equipped with

an avalanche transceiver and a camera system. Their

concept included plans for automated navigation to

accident sites. However, the company stopped its

activities in 2020, facing challenges related to lim-

ited ﬂight endurance and operational speed (Nivitec,

2020).

An alternative mitigation strategy is employed by

the Atlas AVALANCHE PRO system, which sus-

pends the avalanche transceiver on a ﬂexible pole 1

to 2 meters below the UAV to physically reduce EMI.

Although this approach helps minimize interference,

the uncontrolled orientation of the transceiver com-

promises the precision of the location, leading to the

potential loss of critical positional data (ATLASUAS,

2025).

Beyond these commercial initiatives, a growing

body of academic research has advanced UAV-based

avalanche victim localization using transceiver sys-

tems. Silvagni et al. (Silvagni et al., 2017) developed

a UAV system that integrates a commercial avalanche

transceiver and thermal camera, demonstrating auto-

mated localization tested in Alpine conditions.

Azzollini et al. (Azzollini et al., 2020) pro-

posed an extremum-seeking control strategy to au-

tonomously steer a transceiver-equipped UAV toward

the strongest signal, validated through simulation and

hardware-in-the-loop tests.

The AVERLA project by Toson et al. (To-

son et al., 2021) presented a custom high-gain an-

tenna system for UAV-mounted transceiver tracking,

achieving preliminary ﬁeld success despite challenges

in system miniaturization.

However, Janovec et al. (Janovec et al., 2022) re-

ported severe EMI issues between UAV electronics

and the beacon, concluding that conventional electric

UAV platforms are ill suited for avalanche searches

without extensive modiﬁcations. For the tests, they at-

tached the avalanche transceiver with a 2m long rope

underneath the drone.

Ricciardi (Ricciardi, 2017) showed that motor-

induced EMI limits the sensitivity of avalanche

transceivers to approximately 6 m with spinning pro-

pellers. EMI-grade aluminum shielding around the

motors and arms extended reliable detection to ap-

proximately 7 m.

In contrast to these previous systems and research,

our proposed VTOL-based solution offers several

critical advances. Using a state-of-the-art avalanche

transceiver integrated with real-time GPS and head-

ing data fusion, we substantially improve the robust-

ness of the localization. Furthermore, the use of a ro-

bust, professional unmanned VTOL system powered

by a turbine engine markedly reduces electromag-

netic interference compared to electric motor plat-

forms. This conﬁguration enables high-speed auto-

mated operation in challenging and rough mountain

environments, positioning our system as a signiﬁcant

step forward in UAV-based avalanche rescue technol-

ogy in practice usage.

1.3 Methodology

The following subsections describe in detail the hard-

ware setup of the payload, as well as the VTOL sys-

tem setup and the localization algorithms.

1.3.1 Hardware Setup VTOL System

A VTOL system, a helicopter from ANAVIA, was

used as a carrier system for the sensor system.

ANAVIA’s HT-50 helicopter (ANAVIA, 2025b) fea-

tures a carbon ﬁber composite airframe and dual inter-

meshing Flettner rotors, emulating the full-scale ﬂight

dynamics of the larger HT-100 system (ANAVIA,

2025a). Its maximum take-off weight (MTOW) is

50 kg with a payload capacity of 20 kg. Power is pro-

vided by a 7 kW shaft-driven microturbine coupled to

a high-performance gear. This type of drive is a ma-

jor difference from all the other approaches and pro-

totypes presented in1.2.

Fuel is stored in a 17 L tank. The helicopter con-

sumes approximately 10 L/h under cruise conditions.

HT-50 achieves a maximum ﬂight time of 90 min and

has a top speed of 100 km/h. The system will be

able to ﬂy a pre-planned mission fully autonomously.

However, the ﬁnal version of the autopilot was not

yet installed during the tests, so all ﬂights were ﬂown

manually for our data collection.

Figure 3: HT-50 from ANAVIA was developed as a training

platform for the larger HT-100. It has a payload of 20 kg, a

ﬂight time of 90 min and a maximum speed of 100 km/h.

Sky Savers: Leveraging Drone Technology for Victim Localization in Avalanche Rescue via Transceiver Signal Analysis

471

1.3.2 Hardware Setup Sensor System

Our system consists of an MAMMUT Pulse Barryvox

avalanche transceiver, a Raspberry Pi, a GPS-module,

an absolute Inertial Measurement Unit (IMU) and

a Power Supply. The design and arrangement of

the hardware components were chosen to achieve the

best possible sensitivity of the avalanche burial sig-

nal and to ensure precise localization through accu-

rate GPS data and efﬁcient data processing. The

avalanche transceiver was mounted on the tailboom

of the VTOL system - a location with minimal possi-

bility of EMI. The Raspberry Pi serves as the central

processing unit (CPU) of the system. The CPU is re-

sponsible for managing the Universal Asynchronous

Receiver/Transmitter (UART) connection with the

avalanche transceiver, which allows real-time data

transfer from the transceiver to the central processing

unit.

The data received from the avalanche transceiver

include the direction and distance of the target, which

are used in the subsequent localization algorithms.

This data is then merged with the current GPS loca-

tion and heading information provided by the VTOL-

system. In this state of research, all computationally

expensive tasks, such as source localization or visu-

alization, are performed ofﬂine on a laptop after the

ﬂight. The GPS data are then fused with the distance

and direction data from the avalanche transceiver to

locate the buried avalanche transceiver. As the current

aircraft did not yet have a fully functional autopilot

with real-time kinematics (RTK) on board, a HILTI

PLT-400 (Hilti, 2025) total station was used to track

the target on the ground. The corner points of the

search area were obtained with cm-accurate position

data.

1.4 Target Localization Approach

The magnetic ﬁeld is generated by a static magnetic

dipole based on the analytical formulation shown in 1.

A magnetic dipole can be described by its dipole mo-

ment m. In our case, the dipole moment describes the

strength of the dipole and depends on the properties

of the coil.

B(r) = ∇ × A =

4π



3r(m · r)

−



(1)

where B(r) is the magnetic ﬁeld generated by the

transceiver at distance r, m = (m

, m

)

is the

magnetic moment vector of the dipole that is given by

the speciﬁcations of the avalanche transceivers coils.

The constant µ

is the vacuum permeability. In the

current state of research, we do not have available B

ﬁeld data from the avalanche transceiver, but only a

distance r and a direction value processed internally.

The use of the B ﬁeld data would allow us to solve

the inverse problem via constrained optimization for

a more accurate localization.

We employ a two-stage Extended Kalman Filter

(EKF) approach to estimate the (assumed stationary)

target coordinates (x

, y

). The ﬁrst EKF (Smooth-

ing Filter) combines the raw directions ϕ and dis-

tance r measurements from the avalanche transceiver

with GPS position and heading (x

, y

, h

) of the

drone for noise reduction. The second EKF (Track-

ing Filter) uses the smoothed measurements to update

the target estimate using the non-linear measurement

model, shown in 2. In Figure 4 we can ﬁnd the high-

level schematic of the setup.

(

+ r cos ϕ = x

+ r sin ϕ = y

(2)

Figure 4: We have used one transceiver device in send

mode, as a target. A second system is used in search mode

on the VTOL-system to collect data and used for localiza-

tion.

1.4.1 Smoothing EKF

To robustly fuse coarse avalanche transceiver data

with precise GPS position and heading, we imple-

mented a simple EKF that smoothed outliers in direc-

tion and distance before propagating a second ﬁlter to

estimate the absolute location of the target.

State and Process Model

We deﬁne the state vector at time k as:





, (3)

where ϕ

is the estimated direction to the buried

transceiver in degrees, and r

is the estimated distance

to the target in decimeters. The covariance P

∈ R

2×2

encodes the uncertainty.

ICINCO 2025 - 22nd International Conference on Informatics in Control, Automation and Robotics

472

Upon receiving new GPS data, indicating a dis-

placement of (∆x, ∆y) in meters from successive lat-

itude and longitude readings, we predict the next

state. The prediction model, which accounts for the

searcher’s movement, is given by:

= ϕ

k−1

+ α∆θ, (4)

ˆr

= max



k−1

− β∥[∆x, ∆y]∥, r

min



, (5)

where:

• ∆θ ∈ [−180

◦

, 180

◦

] is the angular difference be-

tween the GPS-derived motion heading and the

previously estimated direction of the transceiver

k−1

• α and β are adaptive gain factors that reﬂect how

strongly the searcher’s motion is expected to in-

ﬂuence the transceiver estimates. These gains

are dynamically adjusted on the basis of observed

movement.

• r

min

> 0 is a minimum distance threshold, pre-

venting non-physical negative ranges.

The predicted state covariance matrix

evolved by adding the process noise covariance ma-

trix Q

= P

k−1

+ Q

k−1

. (6)

The noise of the process Q

is dynamically tuned

to allow moderate drift in the estimated direction and

distance, its magnitude being adjusted based on the

speed of the searcher’s movement.

Measurement Update

When a new transceiver measurement (ϕ

meas

, r

meas

)

becomes available, we use a measurement matrix H =

I, as the measurements directly correspond to the state

variables. The residuals (innovations) are formed by

comparing the measurement with the predicted state,

with a proper angle wrapping in [−180

◦

, 180

◦

] for the

direction:

= ((ϕ

meas

−

+ π) mod 2π) − π, (7)

= r

meas

− ˆr

. (8)

The measurement noise covariance matrix R

adaptively adjusted based on two primary factors:

• Range-dependent direction noise: A larger mea-

sured distance r

meas

implies higher directional un-

certainty, leading to an increased variance for

meas

• Low-motion penalty: If the searcher’s movement

between updates is minimal (≪ 1 m), both direc-

tional and distance variances in R

are scaled up

to counteract potential spurious swings or static

measurement inaccuracies.

The Kalman gain K

and the subsequent state and

covariance updates are calculated as follows:

= H

+ R

, (9)

−1

, (10)



ˆr



+ K





, (11)

= (I − K

(I − K

+ K

. (12)

1.4.2 Tracking EKF

We employ another Extended Kalman Filter (EKF)

to estimate the ﬁxed avalanche transceiver’s abso-

lute position (x

, y

) in a local Cartesian frame.

This ﬁlter takes advantage of the smoothed direction

and distance measurements provided by the avalanche

transceiver.

State and Process Model

The state vector at time k is deﬁned as the target’s

estimated position:



TX,k



, (13)

with its associated covariance matrix P

Cov(x

). Assuming the target remains stationary, the

prediction step simply propagates the state estimate

without change.

= x

k−1

. (14)

However, the predicted state covariance evolves

by adding a small noise covariance matrix in the pro-

cess Q

= P

k−1

+ Q

. (15)

The magnitude of Q

is adaptively reduced as

the search progresses through its phases (coarse, ﬁne,

pinpoint), reﬂecting an increasing conﬁdence in the

target’s immobility.

Non-linear Measurement Model

At each time step k, the searcher, located at

, y

) in the same local cartesian frame, obtains

smoothed distance r

and direction ϕ

measurements

Sky Savers: Leveraging Drone Technology for Victim Localization in Avalanche Rescue via Transceiver Signal Analysis

473

from the avalanche transceiver. The non-linear mea-

surement function h : R

→ R

relates the target’s true

position (x

, y

) to these expected measurements:

− x

)

+ (y

− y

)

, (16)

= atan2(y

− y

, x

− x

). (17)

Thus, the measurement vector is z

= [ r

, ϕ

]

⊤

Jacobian Linearization

To incorporate these non-linear measurements into

the EKF framework, the measurement function h is

linearized about the predicted state

. This yields the

measurement Jacobian matrix H

∂h

∂x







−x

ˆr

−y

ˆr

−

−y

ˆr

−x

ˆr





, (18)

where ˆr

is the predicted smoothed distance cal-

culated using the predicted state

and the searcher’s

position (x

, y

). To avoid singularities when the

searcher is very close to the estimated target position,

for very small ˆr

(speciﬁcally, ˆr

< 0.01 m), we sub-

stitute H

≈ I.

Measurement Noise Adaptation

The measurement noise covariance matrix R

is de-

ﬁned as:



0 σ



, (19)

where:

• σ

is the standard deviation of the distance mea-

surement. Scales with the measured range r

(e.g.

≈ 0.1 r

in phase 1, with a minimum of 1 me-

ter) and is signiﬁcantly reduced in later search

phases (ﬁne, pinpoint).

• σ

is the standard deviation of the direction mea-

surement. It is also distance-dependent (inversely

proportional to range) and is further inﬂated if the

searcher is receding from the target. The vari-

ances for both distance and direction are dynam-

ically adjusted based on the current search phase,

with higher conﬁdence assigned to measurements

received during later phases.

Update Equations

The innovation vector (residual) y

is formed by com-

paring the actual measurement z

with the predicted

measurement h(

), ensuring proper angular wrap-

ping for the direction component:



− ˆr



−

+ π



mod 2π− π



. (20)

Here, ˆr

and

are the predicted distance and di-

rection based on

The subsequent EKF update equations are ap-

plied:

= H

⊤

+ R

, (21)

⊤

−1

, (22)

+ K

, (23)

= (I − K

)

(I − K

)

+ K

. (24)

The Kalman gain K

is further modulated based

on the search phase and whether the searcher is ap-

proaching or receding from the target.

2 RESULTS

The performance of the proposed VTOL-based local-

ization approach was evaluated in an open ﬁeld test

site of about 50m × 50m. We adopted a horizon-

tal sweep search pattern as described in (Blankenship

et al., 2022), varying the spacing between the lines,

the ﬂight height and the ﬂight speed of the helicopter.

In total, ten ﬂights were executed, not all on the same

day and on the same location.

The helicopter was manually controlled by an ex-

perienced pilot, who was informed of current altitude

above ground (AGL) and current ﬂight speed at reg-

ular intervals from another operator reading out ﬂight

information from a ground station terminal.

2.1 Ground Tests

Prior to ﬂight tests, we performed stationary ground

tests to assess the impact of sensor placement on the

sensitivity of localization during turbine running. By

mounting the avalanche transceiver on the VTOL tail-

boom, thus increasing distance from the gas turbine,

fuel pump and onboard electronics, we improved the

effective detection range from approximately 10 m to

up to 30 m. This threefold gain demonstrates that

simple mechanical separation from high-EMI compo-

nents can markedly enhance transceiver signal qual-

ity.

2.2 Sensitivity Experiments

As an initial test, the sensitivity of the sensor system

was analyzed during ﬂying. For this purpose, the tar-

get was placed with an offset of 3 m in the x and y

directions to the helicopter. The VTOL system then

ﬂew vertically upward during data recording. The re-

sults in Figure 5 showed that it is possible to track the

ICINCO 2025 - 22nd International Conference on Informatics in Control, Automation and Robotics

474

target up to a distance of 30 m. Compared to the tests

by (Ricciardi, 2017), we achieved a range that was

three to four times higher.

Figure 5: The collected distance data from the transceiver

to the target is shown in blue. The dashed black line shows

the height of the VTOL system in meters above sea level

(m.a.s.l.). After the system had reached an altitude of 30

m, the helicopter turned off and then began tracking another

target (shown in red and green), but this is not relevant to the

test result.

2.3 Test-Flight Experiments

Following ground validation and sensitivity analysis,

we conducted in total ten ﬂights to investigate the lim-

its of the system.

Figure 6: Results of a test-ﬂight: raw transceiver direction

(dotted gray arrows) and distance (colored markers), current

transceiver ID (markers shape), smoothed direction and po-

sition (black arrows), and ﬁnal estimated target location (red

cross) and ground truth (green star). The dotted line around

the ﬂight path marks the search area boundary.

For the test ﬂight in Figure 6 we chose a spacing

between the lines of 5 m. The pilot ﬂew at a height

of 3 m above ground and a speed of 2 m/ s. Both,

altitude and airspeed, varied slightly due to the lack

of an autopilot. The green and red squares show the

start and stop points of the helicopter.

2.4 Flight-Path and Target Localization

The direction readings (dotted gray arrows) and dis-

tance measurements (colored markers) in Figure 6are

the noisy raw data collected from the avalanche

transceiver. After smoothing the direction and dis-

tance, the estimated target location (red cross) con-

verges near the true target location (green star). The

Euclidean distance between the ground truth and the

estimation was 0.63,m in this experiment. Using RTK

positioning of the drone could further increase the lo-

calization accuracy.

Figure 7: Evolution of the Tracking EKF’s estimated target

position (in green) in the horizontal plane over time. The

blue lines show the ﬂight path. The numbers indicates the

current index of the datapoint.

Figure 7 depicts the Tracking EKF’s state estimate

of the target’s (x

, y

) coordinates. The plot shows

how the position of the target estimated by the EKF

converges in the proximity of the ground truth posi-

tion.

2.5 Discussion

The purpose of the paper was to demonstrate the func-

tionality of a helicopter-based localization system.

The current state-of-the-art is and remains companion

rescue.

The added value of the system comes into play

in dangerous conditions or in very large avalanche

cones, where it is difﬁcult to move on foot. Field tests

must be carried out to compare such scenarios, but

this has not yet been done and will only be the case

when the system is further developed.

Sky Savers: Leveraging Drone Technology for Victim Localization in Avalanche Rescue via Transceiver Signal Analysis

475

3 CONCLUSION

In this work, we have presented a novel VTOL-based

avalanche victim localization system that overcomes

key limitations of existing brushless motor-driven

platforms presented in section 1.2. By replacing noisy

DC motors with a turbine, we effectively avoided EMI

issues in avalanche transceiver signals and demon-

strated submeter localization accuracy under realistic

ﬁeld conditions. The use of a turbine-powered VTOL

system allows operations in cold weather conditions

and rugged alpine terrain, greatly extending the prac-

tical utility of search and rescue missions.

In the future, we plan to integrate raw B ﬁeld

vector measurements directly from the avalanche

transceiver coil to reﬁne position estimates and im-

prove the overall localization precision. We will also

explore inverse optimization methods to enable si-

multaneous tracking of multiple buried subjects. To

realize a fully end-to-end rescue solution, future work

will focus on:

• Developing (semi-) autonomous ﬂight trajectories

for rapid deployment to the accident site,

• Incorporating on-board camera and LiDAR sen-

sors for real-time estimation of the avalanche cone

geometry and safe landing zones,

• Designing a reliable target-marking mechanism

such as visual markers to guide ground teams to

located victims.

These advances will be critical to the delivery

of a complete and life-saving system capable of au-

tonomous search, pinpoint location, and target mark-

ing in harsh mountain environments.

ACKNOWLEDGEMENTS

The authors appreciate the support provided by GPV

AG and MAMMUT AG. Their provision of avalanche

transceivers, as well as the corresponding interfaces

and communication protocols, was essential for the

development and execution of this research. The au-

thors also thank HILTI AG for providing the PLT

400 total station and support with data extraction.

The project was funded by Innosuisse’s Innovation

Booster Robotics program.

REFERENCES

Alcedo (2010). Alcedo: A drone-based avalanche rescue

prototype. Unpublished student project report, ETH

Zurich.

ANAVIA (2025a). Ht-100.

ANAVIA (2025b). Ht-50.

Association, A. A. (2025). Avalanche fatalities statistics.

https://avalanches.org/fatalities/fatalities-statistics/.

[Accessed: April 23, 2025].

ATLASUAS (2025). Atlasavalanche pro system overview.

Product brochure; employs a ﬂexible transceiver

mounting to reduce EMI.

Ayuso, N., Cuch

ı, J., Lera, F., and Villarroel, J. (2015). A

deep insight into avalanche transceivers for optimiz-

ing rescue. Cold Regions Science and Technology,

111:80–94.

Azzollini, I. A., Mimmo, N., Gentilini, L., and Marconi, L.

(2020). Uav-based search and rescue in avalanches

using arva: An extremum seeking approach. In IFAC-

PapersOnLine, volume 53, pages 1627–1632.

Blankenship, R., Bluman, J., and Steckenrider, J. (2022).

An investigation of search algorithms for aerial recon-

naissance of an area target. In Proceedings of the An-

nual General Donald R. Keith Memorial Conference,

pages 72–77, West Point, New York, USA. IEWorld-

Conference.org. A Regional Conference of the Soci-

ety for Industrial and Systems Engineering.

Bluebird (2021). Powderbee drone system. Dissolved on

August 10, 2021.

Brugger, H., Etter, H. J., Zweifel, B., Mair, P., Hohlrieder,

M., Ellerton, J., Elsensohn, F., Boyd, J., Sumann, G.,

and Falk, M. (2007). The impact of avalanche rescue

devices on survival. Resuscitation, 75(3):476–483.

DJI (2025). Matrice 200 series.

ETSI (2017). Avalanche beacons operating at 457 khz;

transmitter–receiver systems; part 1: Harmonised

standard for access to radio spectrum.

Falk, M., Brugger, H., and Adler-Kastner, L. (1994).

Avalanche survival chances. Nature, 368(6466):21–

21.

Hilti (2025). Plt 400-4 – digitale absteckger

ate. https:

//www.hilti.ch/content/hilti/CH/de/home.html. Laser

measuring tool.

Janovec, M., Kandera, B., and

Sajbanov

a, K. (2022). Using

unmanned aerial vehicles during the search of people

buried in an avalanche. Transportation Research Pro-

cedia, 65:350–360.

Mammut (2024a). Barryvox pulse. Older generation sum-

mit beacon, discontinued 2017.

Mammut (2024b). Barryvox s. Avalanche transceiver with

3-antenna digital/analog, 70 m search strip width.

Nivitec (2020). Nivitec drone assisted avalanche rescue sys-

tem. project inactive since 2020.

Ricciardi, F. (2017). Automatic Search of Missing People in

Avalanches. Master’s thesis, Alma Mater Studiorum

– University of Bologna, Bologna, Italy.

Silvagni, M., Tonoli, A., Zenerino, E., and Chiaberge, M.

(2017). Multipurpose uav for search and rescue oper-

ations in mountain avalanche events. Geomatics, Nat-

ural Hazards and Risk, 8(1):18–33.

Toson, F., De Giudici, F., Piva, A., and Menegatti, E.

(2021). Averla: Autonomous drone for avalanche res-

cue. In Proceedings of the XXVI AIDAA Congress.

ICINCO 2025 - 22nd International Conference on Informatics in Control, Automation and Robotics

476