Rushana Karimova1, Vikram Unnithan1, Laurenz Thomsen1, Jakob Schwendner2, Mathias Hökelmann2, Andrea Pacifici3, Riccardo Pozzobon3 Angelo Pio Rossi1

(1) Jacobs University Bremen Jacobs University Bremen, Campus Ring 1, 28759 Bremen, (2) German Research Center for Artificial Intelligence (DFKI), Robert-Hooke-Straße 1, 28359 Bremen, (3)"G. d'Annunzio" University of Chieti-Pescara, Via dei Vestini, 66100 Chieti CH, Italy

1 to 100 cm

0.1 to 100 km

Results

Geomorphological Map10 to 0.1 km

IntroductionTesting rover equipment on analogue sites on Earth is

crucial for evaluating the rovers` performance on the soil

types which they will encounter on other planetary

bodies.

We tested the hybrid leg-wheel rover ASGUARD,

developed by the German Research Center for

Artificial Intelligence, during the ROBEX (Robotic

Exploration of Extreme Environments) field trip in June

2015, and collected drone imagery in June, 2016. The

test location was the Fossa cinder cone on the island of

Vulcano in Italy.

Objectives:

• Characterize the test site from mechanical planetary

analogue perspective

• Test and report rover performance on terrains of

various complexity

• Come up with a simple and reliable method to quantify

terrain roughness on the rover test sites

Summary

Fig. 1. ASGUARD rover with a mounted camera on the test site

Methods and Data Collection

Test drive terrain settings: boulder field - loose ash -

solid duricrust - dry riverbed - gully

A GoPro Hero4 camera recorded the path during each

drive. A mobile phone GPS tracked the positioning.

Aerial imagery: collected with an aerial drone over the

boulder field.

Large scale: Geomorphological map, using google

satellite imagery (OpenLayers plugin in QGIS).

Medium to small scale: 3D models, orthomosaics, DEMs

generated from rover`s video feed and drone images

in PhotoScan and Pix4D software.

Roughness: compared for all rover test settings by

manual particle counting from images collected on

site

Automatic particle counting: performed on the

orthomosaic from drone imagery in ImageJ

Rover performance: analyzed using the GPS track and

observations on site

3D models and roughness

• Model resolution: 2.13 cm/pixel (PhotoScan), 1cm/pixel (Pix4D)

• Aerial model and orthomosaic: extensive coverage and very fine

details

• Models from oblique (on-rover) images: fine details but limited

coverage

Particle count for roughness

estimation:

a) A subset of the orthophoto

generated in Pix4D

10 cm to 10 m Limitations and Future Work

b) A Mean Square Error Map

(MSE) derived from the DTM

c) Map of particles

counted in ImageJ

Fig. 2. Screenshot of the 3D model of the boulder field from aerial images

Rover Performance

Fig.4. Rover speed on tracks A,B, C in m/s vs. track points

Fig.3. Locations and slopes of

rover tracks A, B, and C on the

boulder field

• Rover video feed was not usable in many

cases due to the instability of the camera

and its rolling shutter, which produced

blurry images.

• Collecting aerial images on the other test

drive locations would allow for roughness

comparisons using the automated

counting.

• For drive tests, using a differential GPS

devise could improve positioning and

elevation data.

• Additional sensors on the rover (e.g.

engine current sensor) could refine the

performance data considerably.

LiteratureMarlow, J. J., Martins, Z., & Sephton, M. A. (2008). Mars on Earth: soil analogues for future Mars

missions. Astronomy & Geophysics, 49(2), 2-20.

Eich, M., Grimminger, F., Bosse, S., Spenneberg, D., & Kirchner, F. (2008). Asguard: A Hybrid-Wheel Security

and SAR-Robot Using Bio-Inspired Locomotion for Rough Terrain. In International Workshop on Robotics for

Risky Interventions & Surveillance of Environment, Benicàssim Spain.

Kanzog, C. (2015). ROBEX-Robotic Exploration of Extreme Environments. Journal of Unmanned System

Technology, 3(2), 40-45.

Turner, D., Lucieer, A., & Watson, C. (2012). An automated technique for generating georectified mosaics

from ultra-high resolution unmanned aerial vehicle (UAV) imagery, based on structure from motion (SfM)

point clouds. Remote Sensing, 4(5), 1392-1410.

Ferreira, T., & Rasband, W. (2012). ImageJ user guide. IJ1. 46r. Natl. Inst. Health, Bethesda, MD.

http://rsb. info. nih. gov/ij/docs/guide/user-guide. pdf.

A

B

C

Track Slope (degrees) Speed (m/s)

A 25 0.276

B 25.6 0.251

C 12.5 0.448

Table1. Average slope and speed on tracks A,B,C

Flatter slopes →

higher speeds at

similar roughness

(track C).

Acknowledgements This work was supported by the Helmholtz Association,

project alliance ROBEX, under contract number HA-

304. We thank Sven Klüber for assisting with the drone

operation, Matthias Killian, Christian Bühler and Paulina

Prondzinsky for their help during field work

A

B

C

• Orthomosaics and models created in

Pix4D software from drone imagery

worked best

• Drone images → orthomosaic →

automatic particle counting in

imageJ most straightforward for

quantitative roughness

• Mean Square Error maps – boulder,

slope maps – channel related

roughness, use together for traverse

planning

• Minimum height of an obstacle that

could not be crossed =~25cm

• Slope - larger influence on

traversibility than roughness

• Loose soil + steep slope ~30°created too much wheel slippage