ROBOTIC DEPLOYMENT OF EXTRATERRESTRIAL SEISMIC NETWORKS

Daniel Leidner1, Selma Music2, and Armin Wedler1

1German Aerospace Center (DLR), Institute of Robotics and Mechatronics, Germany, 82234 Weßling,daniel.leidner@dlr.de, armin.wedler@dlr.de

2Technische Universitat Munchen (TUM), Institute for Information-Oriented Control, Germany, 80333 Munchen,music@lsr.ei.tum.de

ABSTRACT

Manual installation of seismic networks in extraterres-trial environments is risky, expensive and error-prone.A more reliable alternative is the automated depositionwith a light-weight robot manipulator. However, insert-ing a spiked sensor into soil is a challenging task for arobot since the soil parameters are variable and difficultto estimate. Therefore, we investigate an approach toaccurate insertion and positioning of geophones using aCartesian impedance controller with a feed-forward forceterm. The feed-forward force component of the controlleris either estimated using the Fundamental Earth-MovingEquation, the Discrete Element Method or empirically.For the first time, both the geological aspects of the prob-lem as well as the aspects of robotic control are consid-ered. Based on this consideration, the control approachis enhanced by predicting the resistance force of the soil.Experiments with the humanoid robot Rollin’ Justin in-serting a geophone into three different soil samples vali-date the proposed method.

Key words: Robotic Space Exploration, Seismic Net-works, Compliant Manipulation, Soil-Tool Interaction.

1. INTRODUCTION

To investigate sub-surface properties and seismic activ-ity of extraterrestrial environments, such as the Moon,seismic networks need to be deployed at the surface on-site. However, the manual installation of extraterrestrialseismic networks, i. e. geophones, is risky, expensive anderror-prone. Evidence for this can be found in the reportsof the Active Seismic Experiment (ASE) conducted duringthe Apollo missions [1]. During the Apollo 14 mission,the astronauts emplaced the so-called Apollo Lunar Sur-face Experiments Package (ALSEP), which constitutes aseismic network consisting of a seismometer and a stringof three spiked geophones. The mission report states thatthe lunar soil gave little resistance to hold the geophonesin place causing them to tilt after inserting them into theground. The astronauts where forced to repeatedly adjustthe sensors to guarantee good coupling with the soil andan upright position with less than seven degree tilt error.

Figure 1. The humanoid robot Rollin’ Justin inserting ageophone dummy into Martian soil simulant.

As a consequence the astronauts took longer than plannedto setup the experiment. Moreover, The Lunar Recon-naissance Orbiter (LRO) recently captured images of ad-ditional seismic instruments deployed during the Apollo17 mission. Czeluschke et al. [2] found out that theseimages show significant differences between previouslypublished and LRO-based source-receiver distances of upto 40 m, resulting in inaccurate assumptions of the lunarsub-surface properties.

A more reliable solution is the automated sensor deposi-tion by a robotic manipulator, allowing for efficient, pre-cise and repeatable task completions. Sensing capabili-ties to measure force and torque enable a compliant robotto fulfill this task w. r. t. the requirements of the geolog-ical experiments. This paper describes the experimen-tal validation of this issue as part of the ROBEX project[3]. In a laboratory environment, a suite of experimentsis conducted to validate the automated deployment pro-cedure under varying condition. These experiments serveas proof of concept for the upcoming field mission to beheld at a planetary analogue site. During this mission, theLRU Rover [4] equipped with a light weight manipulatorwill deploy an active seismic network which resemblesthe ALSEP setup deployed during the Apollo missions[1]. The geophones of the seismic network have to bealigned precisely w. r. t. the vertical and lateral directions.

Moreover, specific forces have to be applied to ensuregood coupling between the geophone spike and the soil,which is important for the geological measurements.

In this work we propose a control strategy to insert aspiked tool into soil with a compliant light weight robot.Our method is based on Cartesian impedance control withan additional feed-forward force term, as described inSec. 3. This term is calculated w. r. t. the soil propertiesbased on three methods, namely the Fundamental Earth-Moving Equation and the Discrete Element Method andan empirical method. We compare our approach witha state-space control strategy, and Cartesian impedancecontrol without feed-forward force term in a set of elab-orate experiments in Sec. 4. We evaluate our approachwith three different soil samples, possibly found in ex-traterrestrial environments.

2. RELATED WORK

Robot interaction with the environment has been a con-stantly expanding area of research. Constrained robotmovement poses some very challenging tasks mainlyconcerned with maintaining stability in cases of transi-tion from unconstrained to constrained motion, unknownenvironments or environments with variable properties.This problem becomes all the more difficult if the robotinteracts with the environment while using a tool, suchas a geophone, to be inserted into soil with possibly un-known properties. This problem can be solved using ei-ther hybrid force/position control or impedance control.The concept of impedance control was first proposed byHogan [5] and was motivated by the fact that separatecontrol of position and force is not sufficient in case ofdynamical interaction between a manipulator and the en-vironment. The operational space formulation by Khatib[6] enabled control formulation in the task space insteadof in the joint space. This led to the introduction of Carte-sian impedance controllers, as described by Ott [7]. In[8], a modified controller for work in environments withunknown properties is proposed by introducing an esti-mated feed-forward force term.

Analysis of soil-tool interaction is a very wide researcharea in geophysics, but little work was done in combin-ing this work with robotic manipulation for the purposeof process automation. Basics of soil mechanics for co-hesionless and cohesive soils are presented in [9]. In[10], Reece argues that all soil forces can be describedby a single equation. He studied a problem of bladecutting into the soil and proposes a Fundamental Earth-Moving Equation (FEME). This equation has become astarting point for many soil-tool interaction analysis. Anadvanced modification of Reece’s fundamental equationwas developed by Chung and Sudduth [11] w. r. t. a conepenetrometer, traveling vertically through soil.

Nowadays, there are many software tools for simulatingsoil behavior and computing force as a result of soil-toolinteraction. Most work in this area is based on the Dis-

crete Element Method (DEM) developed by Cundall andStruck [12]. The open-source software YADE-DEM im-plements the Discrete Element Method [13] for the sim-ulation of granular materials. It is frequently used for theanalysis of physical parameters of the soil and often inthe analysis of soil-tool interaction. Obermeyer et al. pre-dict horizontal draft forces for a thin metal plate movingthrough cohesionless soil, using YADE-DEM [14]. Mod-enese et al. analyzed lunar soil behavior using DEM mod-eling in YADE-DEM [15]. The conclusion was made thatlunar soil shows unusual cohesion in comparison to theterrestrial soil of the same mineralogy. That is explainedby higher surface energies resulting from different envi-ronmental conditions, i. e. low gravitational acceleration,very low pressure, and very high temperatures. There-fore, higher forces need to be exerted in order to breakthe soil while inserting a tool.

The closest related research to our work can be found inthe field of automated excavation and digging processes.In [16], Luengo et al. suggest to model closed loop be-havior based on predicted soil resistance forces. Robot-soil interaction was studied by Hong [17]. The authorpresented different soil models and different robot controlmethodologies, implemented for the purpose of obtainingsoil parameters from the interaction of robot manipulatorwith the soil. In [18], ground coupling of a spiked geo-phone with the ground was studied. The geophone wasconsidered as a cylinder and it was shown that groundcoupling is dependent on the length of spike.

3. AUTOMATED GEOPHONE DEPLOYMENT

Inserting a spiked geophone sensor into soil is a challeng-ing task for a robot since the soil parameters are variableand difficult to estimate. Therefore, we investigate theproblem of soil-tool interaction w. r. t. both the geologi-cal aspects of the problem as well as the aspects of therobotic control system. This is done by using a Cartesianimpedance controller with a feed-forward force term out-lined in Sec. 3.1. The feed-forward force component ofthe controller is estimated with two traditional methodsdealing with tool-soil interaction, namely the Fundamen-tal Earth-Moving Equation (FEME) and the Discrete El-ement Method (DEM). The FEME method is a simplifiedanalytical approach to desribe continuous tool-soil inter-action according to the mathematical model of a blade in-serted into soil. This method can be applied to cohesion-less and cohesive soil which enables the simulation offine-grained, cohesive lunar regolith, which is outlinedin Sec. 3.2. On the contrary the simulation based DEMmethod is able to simulate larger, irregular particles andresulting discontinuities such as the medium-sized, lightlunar basalt rocks. This method is detailed in Sec. 3.3.Additionally, a third empirical method is proposed to de-scribe the soil resistance force if the soil parameters areunknown in Sec. 3.4. The robot itself is thereby utilizedto explore the soil properties by inserting the geophonewith a stiff position control strategy to measure the exter-nal forces.

3.1. Cartesian Impedance Control with Feed-Forward Force Term

Impedance control enables simultaneous control of forceand motion by defining a virtual, spatial impedance be-tween the current and the desired robot configuration.Therefore, it is applicable for constrained tasks as it lim-its environmental force via the constant relation betweenforce and motion. The impedance control action in theCartesian space for the regulation task and with the as-sumption of a quasi-static task execution, i.e. x ≈ 0, canbe defined as following:

f c = −Dxx−Kx(x− xd) (1)

where the vector f c ∈ R6 is the wrench exerted by therobot, Dx ∈ R6x6 is the virtual damping matrix of themanipulator in the Cartesian space and Kx ∈ R6x6 thevirtual stiffness matrix of the robotic manipulator in theCartesian space. A desired robot configuration is denotedas xd ∈ R6, while a current robot configuration is de-noted as x ∈ R6. Accordingly, x ∈ R6 is a current robotvelocity. The virtual damping matrix, Dx, is computedusing the double diagonalization approach [19]. The ro-tational component of the impedance equation is basedon quaternions [20]. At this stage, only a decoupled be-havior is considered, i.e. the matrices Dx and Kx arediagonal. In the task of inserting a spiked geophone,precise positioning and good coupling with the soil aremandatory. The Cartesian impedance control should beextended w. r. t. to these requirements to allow for an ex-plicit bound on the interaction force. Therefore, we pro-pose a feed-forward force term to overcome the soil re-sistance force. The dynamical equation of the roboticsystem in the Cartesian space, with the included feed-forward force term, is described as follows:

M(x)x+C(x, x)x+ g(x) = f c + fd + fe (2)

where the components M(x) ∈ R6x6, C(x, x) ∈ R6x6

and g(x) ∈ R6 represent the inertial matrix, the matrix ofCoriolis and centrifugal contributions to the force and thegravitation vector of the robotic system in the Cartesianspace, respectively. Furthermore, f c ∈ R6 is the con-trol action of the Cartesian impedance controller (i. e. thewrench exerted by the robot on the environment), whilethe fd ∈ R6 is the feed-forward force term (i. e. the de-sired force exerted by the robot) and fe ∈ R6 is thevector of external wrenches exerted by the environment(soil). Please note that the vertical motion of geophoneinsertion is studied in this paper solely. Accordingly, onlythe vertical, translational component fd,3 is of interest.Hence, in the remaining paper only the force componentfd,3 (and analogously fe,3) will be considered.

In order to predict fd,3, detailed knowledge of the inter-acting environment is required. In case of the task of geo-phone insertion, the prediction of the interaction force be-tween the soil and the spike depends mainly on the soilproperties. Three prediction methods are described in thefollowing sub-sections.

3.2. Fundamental Earth-Moving Equation

One way to predict the soil resistance force is the so-called Fundamental Earth-Moving Equation (FEME) byReece [10]. It describes a mathematical model in 2D, de-veloped to theorize the design of machines used for mov-ing soil, e. g. in excavation or digging applications. Theauthor developed the equation based on the example ofa blade cutting into soil with low velocity until completesoil failure. A resistance force exhibited between the soiland the moving tool can generally be represented as fol-lows:

F = f(γ, q, c, ca, φ, δ, d, α, θ), (3)

where F is a scalar of the total resistance force magnitudeof the soil, acting on the tool during its insertion, γ isthe unit weight of the soil, q represents any additionalsurcharge effects, c is the cohesion of the soil, ca is theadhesion between the soil and the tool, φ is the internalfriction angle of the soil, δ is the friction angle betweenthe soil and the tool, d is the depth to which the tool hasbeen inserted into the soil, and α is the angle of the toolinclination when inserted into the soil. θ has to be definedw. r. t. the geometric properties arising from the shape ofthe tool, such as the diameter of a geophone spike wt.

In a nutshell, the resistance force of the soil acting on thetool depends on the soil properties (density, internal an-gle of friction, cohesion) and the tool properties (width,length, angle of inclination). The FEME equation, pro-posed by Reece, is adapted in this paper by assuming atriangular failing surface instead of a logarithmic one, aswe operate in shallow depths. Since only low velocitiesof insertion are considered, the damping contribution tothe reaction force of the soil during the insertion can beneglected. In this case, the intensity of the reaction forceexerted on the tool by the soil can be calculated as fol-lows:

F = γd2wtNγ + qdwtNq + cdwtNc + cadwtNa (4)

0

2

4

6

8

10

Lunar Regolith

Dense Earth Sand Martian Soil Simulant

fe,3

[N]

0 0.02 0.04 0.06 0.08 0.1

d [m]

Figure 2. The FEME Method applied to three exemplarysoil samples. Note that the lunar regolith (blue, dot-ted), was computed w. r. t. lunar gravitational conditions,where g = 1.62519m/s2

The four main terms in (4) represent the effects of theweight of the soil, any surcharge, the cohesion, and theadhesion between the soil and the tool. TheN factors aredimensionless numbers describing the shape of the soilfailure surface. Based on (4), the intensity of the verticalforce is obtained by a vector decompositon:

fe,3 = F ∗ cos(α+ δ) (5)

It is important to make a difference between cohesionlessand cohesive soils at this point, since these properties af-fect resistive forces to the greatest extent. Cohesionlesssoil, e. g. dry sand, shows no bond between particles. Onthe contrary, cohesive soil, e. g. clay, shows high bondbetween the particles.

This analytical method can be utilized to predict the soilresistance force of cohesive, small to medium grain soilwith varying properties as illustrated in Fig. 2. It is suit-able to simulate the conditions of lunar regolith (simu-lant) and Martian soil (simulant), which constitutes soilsample number three in our experiment suite.

3.3. Discrete Element Method

One approach to predict the interaction forces for morecomplex scenarios, is the simulation based Discrete El-ement Method (DEM). Unlike the Fundamental Earth-Moving Equation, DEM is a three dimensional approachwhich can be applied to more realistic tool shapes sinceit does not require any assumption of the soil failure sur-face shape. It is therefore often used to simulate wheel-soil interaction for extraterrestrial rovers [21]. Particlesof a granular material, such as sand or gravel, are usuallysimulated as a set of spheres. To simulate more complexsoil structures, several spheres can be combined to ap-proximate poly-ellipsoidal particles. The tool interactingwith the soil is defined as a set of vertices forming a rigidbody. As a result, the method is able to compute interac-tion forces between the soil particles, and reaction forces

Figure 3. The simulated DEM tool-soil interaction com-puted and visualized with YADE-DEM for gravel.

0 0.02 0.04 0.06 0.08 0.10

5

10

15

20

d [m]

DEM

fe, 3

[N]

Figure 4. The predicted soil resistance force for gravelsimulated with YADE-DEM.

between the soil and the tool. The discretized methodis especially suitable to predict discontinuities resultingfrom soil which consists of larger particles, such as gravelor lunar basalt rocks. Simulating lunar regolith with theDEM approach is only applicable to a certain extent (withvery small simulation steps), since the lunar regolith par-ticles are very small compared to the surface of the geo-phone spike.

We use the open-source software YADE-DEM [13] tosimulate the insertion of geophones into big grained soil,e. g. gravel or basalt rocks, as illustrated in Fig. 3. Thesimulation results unveil the discontinuities in the forceprofile. The spike moving with a constant velocity pushesthe spheres until they move aside. That causes a freespace in front of the spike allowing it to fall through thatpart of the soil almost undisturbed by the particles. Thisbehavior is not recorded for soils with smaller particles,e. g. in the case of sands this effect is completely miti-gated. A quadratic interpolation of the data can be usedto approximate the feed-forward force of the impedancecontroller. This method is suitable to simulate lunarbasalt rocks, which constitutes soil sample number onein our experiment suite.

3.4. Empirical Soil Parameter Prediction

Soil parameters are hard to predict in general. However,they are required to calculate the resistance force withthe FEME and the DEM method. To overcome this issue,we propose an additional method to empirically relate thesoil resistance force to a variable stiffness behavior. Thisis done by inserting the geophone with the robot withvarying velocities by utilizing a position controller thatis assumed to react infinitely stiff. The stiffness of thegeophone spike is also assumed to be infinitely stiff andtherefore negligible. Fig.5 shows an equivalent mechan-ical structure of the task set-up in the Cartesian space,where the end-effector, the geophone and the ground arerepresented by mechanical elements. The end-effector isreplaced by a mass-damper-spring system, the geophone

End-Effector Geophone Environment (Soil)

Figure 5. Equivalent representation of the end-effector,geophone and soil using mathematical elements mass,damper and spring.

by a spring only and the soil is considered as a visco-elastic material, represented by a damping and a variablestiffness. Note that the effect of the soil damping is ne-glected as only low velocities are considered. The massof the soil is neglected as its quasi-static behavior is as-sumed.

The vertical component of the environment force is mea-sured during the empirical estimation procedure with theposition controller and can be related to the FEME equa-tion as fe,3 = Fcos(α + δ), if and only if d > 0. TheFEME equation is a nonlinear function of the insertiondepth d. All remaining parameters are constant for theproposed task. Consequently, it is possible to representthe FEME equation as a force-to-displacement equationwhere the force is related to displacement via a variablestiffness:

fe,3 = γwtNγcos(α+ δ)︸ ︷︷ ︸C1

d2 +

(qwtNq + cwtNc + cawtNa)cos(α+ δ)︸ ︷︷ ︸C2

d

= C1(xe,3 − x3)2 + C2(xe,3 − x3)= Ke(x3)(xe,3 − x3)

(6)

where d = xe,3 − x3 with xe,3 being the vertical lo-cation of the soil surface and x3 being the vertical lo-cation of the geophone spike tip. The soil stiffness is:Ke(x3) = C1(xe,3 − x3) + C2, where C1 and C2 areconstant parameters of the soil variable stiffness. This al-lows for an estimation of the constants C1 and C2 offlineusing the least-squares method. Analogously to (6), thedesired force exerted by the robot in the vertical, trans-lational direction, in the constrained and unconstrainedsubspace, can be defined as follows:

fd,3 =

{Ke(xd,3)(xe,3 − xd,3), xd,3 ≤ xe,30, xd,3 > xe,3

(7)

where xd,3 is a translational, vertical component of a de-sired end-effector motion and Ke(xd,3) = C1(xe,3 −xd,3) + C2. In general, a well estimated feed-forwardforce compensates for the soil reaction force, while theimpedance control action mitigates the effect of modeluncertainties and external disturbances.

4. EXPERIMENTAL EVALUATION

We conduct a suite of experiments to evaluate the pro-posed Cartesian impedance control strategy with the mo-bile humanoid robot Rollin’ Justin [22] of the GermanAerospace Center (DLR). It is equipped with a sensorizedhead, two DLR LWR III light weight arms and the DLRHAND II as end-effector. This robot is representative forthe LRU rover [4] currently under development for theupcoming field mission of the ROBEX project. The ex-perimental setup includes a geophone dummy, consistingof an inertial measurement unit (IMU) in a custom hous-ing mounted on a cylindrical metal spike (90 mm longand 8 mm diameter), and three boxes filled with differ-ent soil samples: Big grained basalt rocks (22 - 8 mm),medium grain clay particles (8 - 2 mm), and small graincohesive Martian soil simulant (< 2 mm). For each soilsample the force profile can be predicted with the predic-tion methods described in Sec. 3.2 and Sec. 3.3 if the soilparameters are available. In particular, the big grainedbasalt rocks are best simulated with YADE-DEM, and theFEME method is applicable to the two smaller grains. Ifthe soil parameters are unknown, the empirical methoddescribed in Sec. 3.4 can be utilized to predict the soilstiffness offline with the least-squares approach. The pa-rameters given in Table 1 represent the average values ofthe identified parameters.

For each soil sample, the robot has to insert the geo-phone dummy by utilizing three different control strate-gies: State-space control, Cartesian impedance control,and Cartesian impedance control with feed-forward forceterm according to the predicted soil resistant force. Foreach trial the robot is commanded to grasp the geophoneand hold it above the sample container. In this position,the IMU data of the geophone dummy is used to estimatethe geophone orientation and to realign it w. r. t. the sur-face normal of the soil below. Afterwards, the geophonespike is placed on top of the soil, not yet penetrating it,to mark the starting position for the trial. The geophoneis inserted by executing a vertical straight line trajectoryuntil the spike is fully inserted into the soil layer to ensuregood coupling and no tilting (see Fig. 1). Note that notall soil samples have the same fill level, which results indifferent penetration depths. Before each trial, the soil ismechanically loosened to ensure a similar ground com-paction. During the insertion procedure, the desired andmeasured Cartesian position (see Fig. 6), as well as thesoil reaction forces are recorded (see Fig. 7). The experi-ments where executed with an average desired Cartesianvelocity of 38 mm/s.

Soil Sample C1 C2

Basalt Rocks 232.0 12.3Clay Particles 444.5 23.9Martian Soil Simulant 269.9 9.7

Table 1. Average stiffness parameters of the soil samples.

Clay Particles (8 - 2 mm)

0 1 2 3 4 5 6

0.02

0.00

0.04

0.06

0.08

0.10

t [s]

d [m

]

Martian Soil Simulant (< 2 mm)

0 1 2 3 4 5 6

0.02

0.00

0.04

0.06

0.08

0.10

t [s]

d [m

]

Basalt Rocks (22 - 8 mm)

0 1 2 3 4 5 6

0.02

0.00

0.04

0.06

0.08

0.10

t [s]

d [m

]DesriredState SpaceCart. Imp.Cart. Imp. w/ fd

Figure 6. The desired penetration depth (black, solid)compared to the measured trajectories executed withdifferent control strategies, namely state-space con-trol (blue, dotted), Cartesian impedance control (red,dashed), and Cartesian impedance control with feed-forward force term (green, chain-dotted).

A displacement plot for all control strategies and all soilsamples is shown in Fig. 6. In general the state-spacecontroller performs well in the tracking task (low trackingerror) as well as in the regulation task (low steady-stateerror, < 1 mm) for all types of soil. The default Cartesianimpedance controller shows the highest steady-state errorfor all types of soil (up to 15 mm for the basalt rocks). Byincluding the soil model, the Cartesian impedance con-troller with feed-forward force term performs compara-ble to the state-space controller in terms of tracking errorand steady-state error (< 3 mm) for all types of soil. Thestate-space controller is stiff in all Cartesian dimensions.In contrary both impedance control strategies act com-pliant in the direction of motion, as well as in the otherdimensions. This way the robot is able to adapt the mo-tion w. r. t. unknown obstacles, such as larger rocks in theground. This behavior can be observed in the displace-

0 0.02 0.04 0.06 0.080

2

4

6

8

10

d [m]

State Space

Basalt Rocks (22 - 8 mm)

0 0.02 0.04 0.06 0.080

2

4

6

8

10

Clay Particles (8 - 2 mm)

d [m]

0 0.01 0.02 0.03 0.04 0.05 0.060

2

4

6

8

10

Martian Soil Simulant (< 2 mm)

d [m]

fe, 3

[N]

fe, 3

[N]

fe, 3

[N]

Figure 7. Comparison of the soil reaction forces ob-served while inserting the geophone with the state-spacecontroller (blue) and the Cartesian impedance controllerwith feed-forward force term (green).

ment plot for the basalt rocks. As the coupling betweenthe geophone and the ground is best when it is entirelyinserted [18], the Cartesian impedance controller withoutfeed-forward force term is not suited for the task of geo-phone insertion.

The soil reaction force is illustrated in Fig. 7 for the state-space controller and the Cartesian impedance controllerwith feed-forward force term. The state-space controllerincreases the force to penetrate the soil with the geo-phone. The resulting force plot for the first sample con-tainer filled with basalt rocks is comparable to the sim-ulated scenario of gravel in YADE-DEM illustrated inFig. 4. The force oscillates between zero and relativehigh values all the way through the soil, starting alreadyat the soil surface. A similar, but more moderate oscilla-tion can be observed for the medium sized clay particles.The oscillation for the small grained Martian soil simu-lant is negligible, however, the force increases constantly.

In comparison the Cartesian impedance controller withfeed-forward force term limits the soil reaction force. Itis always close to zero in the direction of motion.

We have observed tilting of the geophone after it was re-leased by the robot when using either of the two controlstrategies. Both strategies failed in some trials, as theyexceeded the maximum allowed tilting angle of 7◦ [1].In average, the Cartesian impedance controller with feed-forward force term showed improved coupling betweenthe geophone spike and the ground. Consequently thegeophone tilted less in most of the trials. However, wecould not generate reproducible measurements as severetilting was randomly introduced by the adhesion betweenthe rubber of the fingers and the housing of the geophonewhen releasing it. This effect will be compensated bya dedicated docking interface design, which is currentlyunder development for the upcoming ROBEX field mis-sion.

5. DISCUSSION

In this work we have shown that a light weight robot isable to precisely deploy a seismic sensor in various typesof soil. We utilize Cartesian impedance control with afeed-forward force term to ensure precise positioning andgood coupling. The feed-forward force term is designedto overcome the soil resistance force which can be pre-dicted, analytically with the FEME approach, numeri-cally with the DEM approach, or empirically by explor-ing the soil properties with the robot in advance. The pro-posed control strategy reduces the tilting of the geophoneas it is required by the seismic network to guarantee ac-curate geological experiments.

Different types of soil and different tools, i. e. seismome-ters, may require different deployment strategies. Espe-cially lunar conditions may lead to different requirementsfor the deployment. Therefore, the insertion strategies areintegrated in so-called Action Templates which provide aflexible way to develop process models for arbitrary ma-nipulation tasks [23]. The object-centered approach ofAction Templates make them robot independent per def-inition. They can be utilized by the LRU rover in theupcoming field mission without adaption. This rover willbe equipped with a docking interface instead of a robotichand to further reduce the positioning and tilting error.As a result, the robotic deployment of seismic networkscan be executed more accurate, more reliable, and with-out endangering astronauts. As conclusion we encourageto consider automated seismic sensor deployment for fu-ture extraterrestrial exploration missions.

ACKNOWLEDGMENTS

This work was supported by the Helmholtz Association,project alliance ROBEX, under contract number HA-304.

The authors would like to thank Alexandra Czeluschke ofthe DLR Institute of Planetary Research for the discus-sions about her ongoing research on the re-evaluation ofthe Apollo 17 Lunar Seismic Profiling Experiment basedon new recordings captured with the Lunar Reconnais-sance Orbiter Camera.

REFERENCES

[1] T. Sullivan, “Catalog of apollo experiment oper-ations,” NASA Reference Publication, no. 1317,1994.

[2] A. Czeluschke, M. Knapmeyer, J. Oberst, andI. Haase, “New lunar depth profiles derived fromlroc-based coordinates of apollo 17 seismic equip-ment,” in In Proc. of the European Lunar Sympo-sium, May 2015.

[3] A. Wedler, C. Waldmann, G. Meinecke, M. Wilde,L. Witte, C. Lange, N. Schmitz, M. Knapmeyer,O. Pfannkuche, L. Thomsen, S. Flgel, R. Rosta,T. Bellmann, M. Hellerer, B. Rebele, H. Gmeiner,B. Vodermayer, and Y. Takei, “ROBEX - compo-nents and methods for the planetary explorationdemonstrationmission,” in Proc. of the 13th Sympo-sium on Advanced Space Technologies in Roboticsand Automation (ASTRA), May 2015.

[4] A. Wedler, B. Rebele, J. Reill, M. Suppa,H. Hirschmller, C. Brand, M. Schuster, B. Vo-dermayer, H. Gmeiner, A. Maier, B. Willberg,K. Bussmann, F. Wappler, and M. Hellerer, “LRU -lightweight rover unit,” in Proc. of the 13th Sympo-sium on Advanced Space Technologies in Roboticsand Automation (ASTRA), May 2015.

[5] N. Hogan, “Impedance control: An approach tomanipulation,” ASME Journal of Dynamic Systems,Measurement and Control, vol. 107, pp. 17–24,June 1985.

[6] O. Khatib, “A unified approach for motion and forcecontrol of robot manipulators: The operationalspace formulation,” IEEE Journal on Robotics andAutomation, vol. RA-3, pp. 43–53, February 1987.

[7] C. Ott, Cartesian Impedance Control of Redun-dant and Flexible-Joint Robots, vol. 49 of SpringerTracts in Advanced Robotics. Springer PublishingCompany, Berlin Heidelberg, 2008.

[8] C. Yang, G. Ganesh, S. Haddadin, S. Parusel,A. Albu-Scheffer, and E. Burdet, “Human-likeadaptation of force and impedance in stableand unstable interactions,” IEEE Transactions onRobotics, vol. 27, pp. 918–930, October 2011.

[9] K. Terzaghi, Theoretical Soil Mechanics. NewYork: John Wiley and Sons, 1943.

[10] A. Reece, “The fundamental equation of earth-moving mechanics,” Proceedings of the Institutionof Mechanical Engineers, vol. 179, pp. 16–22, June1964.

[11] S. Chung and K. Sudduth, “Soil failure modelsfor vertically operating and horizontally operatingstrength sensors,” American Society of Agriculturaland Biological Engineers, vol. 49, no. 4, pp. 851–863, 2006.

[12] P. Cundall and O. Strack, “A discrete numeri-cal model for granular assemblies,” Geotechnique,vol. 29, pp. 47–65, 1979.

[13] J. Kozicki and F. Donze, “A new open-source soft-ware developed for numerical simulations using dis-crete modeling methods,” Computer Methods in Ap-plied Mechanics and Engineering, vol. 197, no. 49,pp. 4429–4443, 2008.

[14] M. Obermayr, K. Dressler, C. Vrettos, and P. Eber-hard, “Prediction of draft forces in cohesionless soilwith the discrete element method,” Elsevier Journalof Terramechanics, vol. 48, pp. 347–358, Septem-ber 2011.

[15] C. Modenese, S. Utili, and G. Houlsby, “DEM mod-elling of elastic adhesive particles with applicationto lunar soil,” in Proc. of the Earth and Space 2012Conference, pp. 45–54, April 2012.

[16] O. Luengo, S. Singh, and H. Cannon, “Modelingand identification of soil-tool interaction in auto-mated excavation,” in Proc. of the IEEE/RSJ Inter-national Conference on Intelligent Robots and Sys-tems, pp. 1900–1906, 1998.

[17] W. Hong, Modeling, Estimation and Control ofRobot-Soil Interactions. PhD thesis, MassachusettsInstitute of Technology, September 2001.

[18] G. Drijkoningen, F. Rademakers, E. Slob, andJ. Fokkema, “A new elastic model for groundcoupling of geophones with spikes,” Geophysics,vol. 71, pp. Q9–Q17, March-April 2006.

[19] A. Albu-Schffer, C. Ott, and G. Hirzinger, “A pas-sivity based cartesian impedance controller for flex-ible joint robots - part ii: Full state feedback,impedance design and experiment,” in Proc. of theIEEE International Conference on Robotics andAutomation (ICRA), vol. 3, pp. 2666–2672, April2004.

[20] F. Caccavale, C. Natale, B. Siciliano, and L. Villani,“Six-dof impedance control based on angle/axisrepresentations,” IEEE Transactions on Roboticsand Automation, vol. 2, pp. 289–300, April 1999.

[21] M. A. Knuth, J. Johnson, M. Hopkins, R. Sullivan,and J. Moore, “Discrete element modeling of a marsexploration rover wheel in granular material,” Jour-nal of Terramechanics, vol. 49, no. 1, pp. 27–36,2012.

[22] C. Borst, T. Wimbock, F. Schmidt, M. Fuchs,B. Brunner, F. Zacharias, P. R. Giordano, R. Koni-etschke, W. Sepp, S. Fuchs, et al., “Rollin’justin-mobile platform with variable base,” in Proc. of theIEEE International Conference on Robotics and Au-tomation (ICRA), pp. 1597–1598, 2009.

[23] D. Leidner, C. Borst, and G. Hirzinger, “Things aremade for what they are: Solving manipulation tasksby using functional object classes,” in Proc. of theIEEE/RAS International Conference on HumanoidRobots (ICHR), pp. 429–435, 2012.