Content uploaded by Christian Ernst Siegfried Koch
Author content
All content in this area was uploaded by Christian Ernst Siegfried Koch on Nov 25, 2020
Content may be subject to copyright.
Underwater Demonstrator for Autonomous In-Orbit Assembly of Large Structures
Virtual Conference 19–23 October 2020
Christian Ernst Siegfried Koch1, Marko Jankovic1, Sankaranarayanan Natarajan1, Shubham Vyas1,
Wiebke Brinkmann1, Vincent Bissonnette2, Thierry Germa2, Alessio Turetta3, Frank Kirchner1,4
1German Research Center for Artificial Intelligence, Robert-Hooke-Straße 1, 28203 Bremen, Germany,
E-mail: christian.koch@dfki.de
2Magellium, 1 rue Ariane, 31750 Ramonville St Agne, France, E-mail: vincent.bissonnette@magellium.fr
3Graal Tech S.r.k., Via Tagliolini 26, 16152 Genova, Italy, E-mail: alessio.turetta@graaltech.it
4University of Bremen, Robotics Lab, Robert-Hooke-Straße 1, 28203 Bremen, Germany,
E-mail: frank.kirchner@dfki.de
ABSTRACT
The PULSAR project aims to develop key-
technologies to enable the autonomous assembly of
large structures in space. Similar to industrial appli-
cations, the assembly process relies on robotic sys-
tems capable of assembling modular elements to
form a complex structure. However, the in-space
assembly provides exceptional challenges necessitat-
ing innovation in fields such as free-floating manipu-
lation and autonomous robotics. This paper provides
details on the PULSAR project and, more specifical-
ly, on a hardware-in-the-loop demonstrator dLSAF-
FE, developed to show the assembly process of a
large telescope mirror in a micro-gravity environ-
ment.
1 INTRODUCTION
Autonomous assembly of large structures in space is
a key challenge for future missions that will necessi-
tate structures too large to be self-deployed as a sin-
gle piece. The James Webb Space Telescope [1] has
reached this limit with a mirror of 6.5 meter diameter.
Concepts for the next generation telescope LUVOIR
[2], expected by astronomers in the late 2030s, cur-
rently rely on the development of novel launch vehi-
cles with larger payload fairing, e.g. NASA’s Space
Launch System. On-orbit assembly technologies will
deliver from these restraints: Large structures can be
assembled in-situ from modular parts, allowing for
smaller and less expensive launch systems. On-orbit
assembly, however, poses many challenges in fields
such as assembly automation and autonomous robot-
ics.
The EU funded project PULSAR (Prototype of an
Ultra-Large Structure Assembly Robot) aims to de-
velop and demonstrate technology for an on-orbit
precise modular assembly of very large structures by
an autonomous robotic system. In this context, PUL-
SAR focuses on the assembly of the primary mirror
of a space telescope with a diameter of 8 meters. Key
challenges, e.g. high precision and control of the free-
floating manipulation are addressed in software simu-
lations and hardware-in-the-loop (HIL) demonstra-
tors. One HIL demonstrator is dLSAFFE, the demon-
strator of Large Structure Assembly in a Free-
Floating Environment. In the context of PULSAR,
dLSAFFE aims to demonstrate the assembly at near
1:1 scale with representative hardware and software.
Figure 1: Visualization of dLSAFFE in the large test
basin at DFKI Bremen
The need for large structures in space goes beyond
telescopes and also concerns solar arrays for power
plants, light sails to reach outermost regions of the
solar system, or heat shields to land on Mars.
The field of space robotics was identified as key to
increase Europe’s competitiveness in the space sec-
tor. Consequently, the European Commission de-
ployed a Strategic Research Cluster (SRC) on “Space
Robotics Technologies” within the Horizon 2020
research program [3]. The SRC is divided into three
stages, each comprising of several projects, or Opera-
tion Grants (OGs). In the first stage, common build-
ing blocks for space robotics were developed, three
of which are ESROCOS [4], I3DS [5], and SI-
ROM.[6] European Space Robotics Control and Op-
erating System (ESROCOS) is an open-source robot
control operating system (RCOS) targeted towards
space robotics. It was based around the ASSERT Set
of Tools for Engineering (TASTE) toolchain devel-
oped by the European Space Agency (ESA) [7]. I3DS
is a suit comprising perception sensors selected for
space robotics and data processing hardware. SIROM
is an electro-mechanical-thermal interface for cou-
pling modular payloads. The PULSAR project be-
longs to the second stage that aims to further develop
and validate the common building blocks. For exam-
ple, a standard interface based on the SIROM design
is used for coupling the modular mirror tiles, and the
visual perception hardware is selected from the I3DS
sensor suit. Furthermore, the software architecture is
developed and implemented using the ESROCOS
framework.
2 PULSAR SCENARIO
The scenario in PULSAR is the assembly of the pri-
mary mirror of a space telescope. The mirror has a
diameter of 8 meters and consists of 36 hexagonal
mirror tiles (SMT – segmented mirror tiles) arranged
in three rings around a central tile. The assembly
process utilizes an autonomous robotic system, the
Robotic Assembly System (RAS). The RAS will
extract individual SMTs from a storage unit and
assemble them at the pre-defined position. The SMTs
are coupled with neighboring SMTs via the electro-
mechanical and thermal Standard Interface (SI). The
same interface is used as an end effector (EE) by the
RAS to manipulate the SMTs.
The RAS consists of a 6 degree-of-freedom (DoF)
articulated robot manipulator and a rail system acting
as prismatic joint at the base of the manipulator. The
rail system extends the manipulator workspace and
allows it to move between the storage unit and the
assembly area. However, the reach of the RAS is
limited and the outermost ring of SMTs in the mirror
assembly is unreachable. One key-concept in PUL-
SAR is Extended Mobility, allowing the assembly of
structures larger than the workspace of the robot. The
limitation is overcome by implementing a secondary
assembly site, where a set of 5 SMTs is pre-
assembled. This Pre-assembly is then subsequently
manipulated as a single unit as visualized in Fig. 2.
Figure 2: Manipulating and assembling the pre-
assembly (green) to build the outer rings
The assembly process consists of several repeating
steps: (1) Mating of two SIs. Two cases can be con-
sidered: the RAS EE mates with a target SMT, e.g. to
extract it from the storage, or an SMT is assembled at
a desired location. (2) Un-mating of two SIs. Again,
two cases can be considered: the RAS releases a
previously grasped SMT, or SMTs are released from
the spacecraft (S/C) structure, e.g. the storage. (3)
Manipulation of SMTs. The RAS manipulates one or
multiple SMTs towards a target pose. The different
steps are commanded sequentially according to a pre-
defined assembly plan. In each step, the configuration
of the system, including location of the SMTs, is
known. However, to account for uncertainties, the
mating process is guided by visual feedback: Two
cameras, one at the RAS EE and one external camera
observing the mirror assembly, are used for visual
pose estimation based on the detection of fiducial
markers.
The assembly process is demonstrated within the
PULSAR project by three demonstrators, covering
different aspects of the scenario. The demonstrator
presented in this paper, dLSAFFE, aims to demon-
strate the assembly process at near 1:1 scale with a
realistic workspace representation. Emphasis is put
on the Extended Mobility functionality. Three
demonstration scenarios with up to five SMTs are
considered: (1) assembly of the inner ring, (2) assem-
bly of the pre-assembly and (3) assembly of the outer
ring, i.e. the concept of Extended Mobility. For the
final demonstration, a subset of the pre-assembly is
considered. The three scenarios are shown in Fig. 3.
Figure 3: Final stages of the demonstration scenari-
os: assembly of inner ring (left), pre-assembly (mid-
dle), assembly of outer ring (right)
The demonstration is set in a zero-buoyancy envi-
ronment, the large water basin at DFKI Bremen.
Underwater experiments in the context of space ro-
botics, allows to simulate micro-gravity by carefully
balancing weight and buoyancy, components or mis-
sions can be put into a floating state, so that in-orbit-
like conditions can be simulated. For dLSAFFE, RAS
and SMTs are designed to be neutrally buoyant. With
this set-up, the assembly and manipulation of large
structures with a representative lightweight robot can
be performed.
The large basin at DFKI Bremen is used as testing
facility. The basin covers an area of 23 meters length
by 19 meters width with a depth of 8 meters. The
facility is equipped with a 12 tons gantry crane allow-
ing to handle large experiment set-ups [8].
Setting large scale experiments in the field of space
robotics in an underwater environment poses several
challenges. For example, the submerged equipment
must be adapted and simplified for underwater usage.
Further, the large scale of the experiments poses
logistic challenges that need to be addressed within
the project. One adaptation regards the SIs and af-
fects the mating process. To reduce complexity, the
SIs on the SMTs rely on permanent magnets and do
not require electricity for mating. The mating be-
tween SMTs is thus fully passive. On the S/C struc-
ture and the RAS EE, actively controllable electro-
magnetic interfaces are used. An admittance control
scheme is used to exploit the magnetic forces experi-
enced during the mating process. The forces, detected
by the RAS force/torque sensor, are felt as a pull
towards the mating position. Further, the force read-
ing is used to detect the successful mating.
3 DLSAFFE DEMONSTRATOR
The aim of dLSAFFE is to demonstrate the assembly
process at near 1:1 scale with each SMT measuring
about 1.3 meters in diameter. During the demonstra-
tions, one section of the mirror is assembled, stretch-
ing approximately 4 meters across. dLSAFFE is
designed to accommodate such a large experiment
and the demonstration is placed in the large water
basin at DFKI. During the demonstration, a total
height of 6 meters will be reached.
Figure 4: dLSAFFE Hardware Architecture.
(1) Final Assembly Site, (2) Pre-Assembly Site,
(3) cameras, (4) RAS arm, (5) RAS linear rail,
(6) SMT storage
However, the mock-up is equipped with the RAS, a 6
DoF robotic manipulator sitting on a 1 DoF linear
rail. For dLSAFFE, simplified SMTs are designed,
characterized by a low drag and neutral buoyancy.
Each SMT measures 1.3 meters in diameter. The
demonstration scenario considers one section of the
mirror consisting of five modules, stretching approx-
imately 4 meters across, reaching a total height of 6
meters.
3.1 Hardware Architecture
Fig. 4 shows the hardware architecture of dLSAFFE.
As can be seen, only a subset of components needs to
be in the water while, for example, computational
units can remain on land. The submergible part of the
demonstrator features the S/C mock-up measuring
approximately 3 meters in height with a 5 square
meter footprint. The structure of the S/C mock-up is
made of aluminum strut profiles by Bosch Rexroth.
This allows for short-term adaptations if needed. A
platform serves as footing and ensures the stability of
the structure on ground. A tower is mounted on the
platform serving as a supporting element for the
RAS, the pre-assembly site (PAS), the final assembly
site (FAS) and the SMT storage.
Figure 5: CAD visualization of dLSAFFE. In the
back, the SMTs in their storage location can be seen.
The FAS is the central tile of the mirror assembly. It
is designed as a hexagonal prism in the same dimen-
sions as the SMTs. Three sides of the FAS feature
active (controllable) SIs to which the SMTs are as-
sembled. The FAS is mounted on the front side of the
tower whereas the SMT storage and the PAS are
located on the back of the tower. The SMT storage
provides five active SIs to secure the SMTs in their
initial configuration. The PAS provides an additional
SI to support the pre-assembly process. The RAS is
mounted to one side of the tower. It consists of a 6
DoF manipulator arm and a linear rail. The manipula-
tor arm can be moved by the linear rail in order to
reach all the SMTs in their storage as well as the
target assembly locations on the FAS and PAS. The
manipulator arm is equipped with a camera and a six-
axis force/torque sensor. The CAD model of the
dLSAFFE S/C mock-up is presented in Fig. 5.
The demonstrator is controlled by three computation-
al units: The On-board-computer (OBC), the Plat-
form Control Unit (PCU), and a sensor data acquisi-
tion unit from the I3DS project referred to as RCU.
The OBC serves as the main computational unit and
is located on land. The RCU, also located on land,
interfaces the cameras. The PCU, on the other hand,
is located on the submergible platform and interfaces
the active SI, actuators, and remaining sensors. For
communication and power supply, the underwater
demonstrator is connected to land via an umbilical,
i.e. power- and fiber-optical data cables.
3.2 Robotic Assembly System
The RAS consists of two components as shown in-
Fig. 6: a 6 DoF manipulator arm and a 1 DoF linear
rail. The system is mounted vertically to the side of
the S/C tower structure.
Figure 6: Robotic Assembly System (RAS): rail (left)
and 6 DoF manipulator (right)
The manipulator has been made by exploiting the
available design of the Underwater Modular Arm
(UMA) commercialized by Graal Tech. It is charac-
terized by three identical joint modules, with two
motion axes each, for a total of 6 DoF, providing a
sufficient dexterity during the various manipulation
phases. The joints are made of anodized aluminum,
while the connecting bodies are made of carbon fiber
tubes filled with air. One of them is a cylinder, while
the other one is L-shaped (with a 90 ° bend) for min-
imizing the arm size when stowed. The overall length
when totally stretched is slightly more than 2 meters.
The weight in air is around 20 kilograms, while it has
been designed to result neutrally buoyant in water.
The rail acts as a prismatic joint allowing to vertically
move the base of the manipulator along a distance of
1.3 meters. The rail is implemented as a spindle drive
and carriage system, tailored to the requirements of
underwater usage. A commercial off-the-shelf geared
motor inside a custom waterproof housing is used as
actuator. It is designed to support the manipulator
arm both in water as well as in air and emits self-
braking behavior when not powered.
3.3 Mirror Tiles and Standard Interfaces
Each SMT is shaped as a hexagonal prism, as shown
in Fig. 7. The SMTs consist of a skeleton made of
carbon fiber tubes glued together with 3D printed
components providing a proper sealing for ensuring
water tightness. The SIs are mounted on each lateral
side of the tiles to H-shaped carbon fiber plates
clamped to the skeleton through 4 anodized alumi-
num fixing elements. Every SI is attached to its plate
with screws allowing an easy removal and replace-
ment, if necessary. The SMTs are equipped with
fiducial markers for vision passed pose estimation.
To aid the assembly process in the presence of uncer-
tainties, the markers are used to detect the precise
location of the SI and the entire SMT.
The SMTs have been designed to be neutrally buoy-
ant in water. Moreover, its symmetric design guaran-
tees an even mass distribution, which is necessary to
avoid hydrostatic moments in water, while the open
framed structure (without planar faces) significantly
reduces the hydrodynamic drag during motions in
water.
Figure 7: Segmented Mirror Tile (SMT) showing SI
and AprilTag [9] placement
Every SI is characterized by a 90 ° symmetry, with
the two mating parts that need to be have a relative
angle of 45 ° to properly engaged into each other.
The connections are established using magnetic cou-
pling, as it allows a simple enough design compatible
with the underwater environment. To allow the disas-
sembly of engaged interfaces three different kinds of
SIs have been designed: 2 passive and 1 active. In
one of the passive configurations there is a disc of
Neodymium N45 permanent magnet, coated with a
nickel layer as a protection against corrosion. In the
other passive configuration, the magnet is substituted
by an iron cylinder, mounted in the same position. In
the active SI, around the iron cylinder, an electrical
coil is introduced. This way, whenever required, an
electric current within the coil originates an electro-
magnetic force with the opposite polarity than those
acting between magnet and iron. For this reason, the
net intensity of the adhesive force between the two
surfaces results significantly reduced and the disas-
sembly can be operated by a simple motion of the
arm. The design of the SI is shown Fig. 8.
Figure 8: Standard Interfaces (SI) for coupling SMTs
3.4 Software Architecture
Fig. 9 visualizes the simplified software architecture
of dLSAFFE. The architecture is divided in three
layers, Sequencing, Functional and Hardware. The
Sequencing layer comprises high-level control
components. The central component is the
Sequencer. It monitors and orchestrates the assembly
process. The Ground Control Interface provides
feedback and intervention options for the human
operator. The Functional layer comprises perception
and mid-level control functions, i.e. the RAS motion
planning and control components. The Mating Detec-
tor uses the force/torque sensor to detect the
successful mating of SIs. The Hardware layer
comprises all the drivers of the actuators and sensors.
They are interfaced by components from both other
layers.
While some functionality, e.g. drivers, are specific to
dLSAFFE, several functions are shared as Core
Components among the three demonstrators of
PULSAR. For example the RAS Motion Planner and
Controller and the Perception Functions are
considered Core Components and thus designed with
generality in mind and not tailored to the specifics of
each demonstrator.
The software stack is deployed on a distributed
system among three different computational units:
OBC, PCU and RCU. The OBC runs the majority of
higher-level components, while the PCU acts as an
interface to the individual components and runs
drivers and the RAS Controller. The RCU interfaces
the cameras and runs preprocessing functions.
Figure 9: dLSAFFE Software Architecture. The col-
ors of the components denote the deployment: OBC
(blue), PCU (yellow), RCU (green)
The implementation of dLSAFFE software architec-
ture, especially on a representative hardware, is
bound to require a significant software engineering
effort which might be mitigated by using one of the
popular RCOS, such as the Robot Operating System
(ROS) [10], the Open Robot Control Software (Oro-
cos) [11] or the Robot Construction Kit (Rock) [12],
just to name a few. However, none of these frame-
works have been developed with critical applications
in mind, and therefore lack the Reliability, Availabil-
ity, Maintainability and Safety (RAMS) characteris-
tics required for space applications. For this reason,
adapting the existing frameworks to comply with the
mentioned requirements is deemed impractical. Fi-
nally, the existing solutions currently used in the
space industry are usually tied to a specific robotic
system and are mostly closed-source [4].
To bridge this gap, in recent years the ESROCOS
RCOS [4] has been developed as an open-source
framework specifically for space robotics and with
RAMS requirements in mind. It builds upon the
TASTE toolchain [7]. and provides a set of tools and
software components to support the development and
deployment of robotics applications with demanding
RAMS requirements [4].
The PULSAR project relies on the heritage of
ESROCOS by developing and implementing the
dLSAFFE software architecture (see Fig. 9) within
the mentioned framework, as outlined in what fol-
lows.
3.5 RAS Motion Planner and Controller
The objective of the RAS Motion Planner component
is to generate trajectories in joint space for the RAS
to accomplish desired manipulation tasks. The gener-
ated trajectories should fulfill constraints such as
collision avoidance, joint limits, shortest path, etc.,
based on the desired input from the Ground Control
Interface and/or the Sequencer. The component uses
a Motion planner framework [13] to generate an
optimal collision-free path in a cluttered
environment. The framework provides an interface to
different state-of-the-art motion planners [14]–[16],
kinematic [17], [18] and collision libraries [19], [20].
Moreover, the framework is designed using a Factory
method pattern design [21], which provides a user
with a possibility to easily experiment with different
combinations of motion planner, kinematic and
collision libraries to solve complex planning
problems by simply changing a configuration file.
Finally, the framework is platform-independent and
has been already used on real systems within the
Rock framework [12]. An example implementation
of the RAS Motion Planner component within the
functional layer of ESROCOS is displayed in Fig. 10,
illustrating the logical interaction between various
components of the example RAS. Each visible com-
ponent is modelled as a separate TASTE function
using C++ programming The logical dependencies
between the components are defined using a set of
provided and required interfaces which are to be
specified by a user via a GUI and are handled sub-
sequently by TASTE without any user input.
Figure 10. RAS Motion Planner example implemen-
tation in ESROCOS/TASTE
A brief description of the implemented functions and
their expected purpose are outlined as follows:
• Ground_Control_Interface queries the user for
input every 20 seconds and provides the motion
planner with goal pose or joint states.
• Motion_Planner plans an optimal, collision free
motion of the manipulator based on the user
input and current state of the robot.
• Trajectory_Interpolator interpolates the planed
trajectory (whenever available) and supplies the
robot with the joint commands.
• Fake_Robot provides the pervious functions
with the current state of the joints and uses the
command from the trajectory interpolator to
update its current state.
The objective of the RAS Controller component is to
generate joint positions and velocities to be sent to
the manipulator and rail drivers to achieve the desired
motion despite disturbances that will affect the exe-
cution of the planned motion. Therefore, while the
expected input data are the current and desired posi-
tions and velocities of manipulator joints in time, the
expected output data are positions and velocities of
manipulator joints in time to be sent to the drivers of
the rail and manipulator joints.
Due to the very nature of the envisioned mating pro-
cess and the magnetic properties of the interfaces to
be used in the dLSAFFE, physical interaction be-
tween a robot manipulator and the environment is to
be expected and needs to be properly accounted for
within the control scheme of the manipulator during
the mating process of two interfaces. However, the
same control scheme needs to be able to perform the
planned motion despite the hydrodynamic effect that
will occur during the motion of the RAS and tiles
through the water. To account for these requirements
the current implementation of the dLSAFFE control
architecture is envisioned to be composed out of two
separate components: the Feedback Control and Ad-
mittance Control, which are activated by the user
depending on the task to be completed.
The purpose of the Feedback Control is to use the
measured joint positions and velocities to achieve the
desired trajectories calculated by the motion planner
despite the external disturbances or modelling errors.
Therefore, it is envisioned to date to be a PID
controller that derives the necessary joint trajectories
of the RAS rail and manipulator to achieve the
planned motion.
The function of the Admittance Control is instead to
use the measured joint positions and velocities, along
with the measured force/torque applied on the end-
effector to derive the appropriate joint trajectories to
achieve a compliant behaviour of the robot during the
mating process of interfaces.
The implementation of the RAS Controller within the
ESROCOS framework is still an ongoing process and
is therefore not presented in this paper.
3.6 Perception Functions
The perception functions process images captured on
the demonstrator cameras to increase the autonomy
level and the safety of the assembly process.
Input images are first triggered and recovered from
the cameras by the perception component using the
Integrated 3D Sensors (I3DS) framework [5]. The
software framework of I3DS is a set of C++ libraries
for creating sensor interfaces (using their drivers) in a
standardized way. The interfaces use ASN.1 mes-
sages and transport sensor data through a ZeroMQ
communication layer.
Sensor acquisition is based on a subscriber/publisher
pattern to handle frame acquisition at a given fre-
quency, and a client/server pattern for command
request. The EE and external cameras are each con-
nected to a server and a publisher whereas perception
functions define subscribers and callback functions to
manage frame reception.
In dLSAFFE, two primary perception functions are
used: Tile Localization and Assembly Monitoring.
Using the AprilTag [9] fiducial markers mounted on
the sides and back of the segmented mirror tiles, as
shown on Fig. 7, Tile Localization will detect the
tiles, based on their unique IDs, in images captured
by the External or EE-mounted cameras. Tag corners
coordinates are then extracted, and the pose of the tile
with respect to the camera frame is computed using
Perspective-n-Point methods. The pose of the tile is
finally computed using the 2D/3D correspondences,
between 2D coordinates extracted from AprilTags
detection and 3D coordinates of the tile model. This
localization is used to enable a safe, step-by-step
approach to grasp the tiles with the manipulator end-
effector, instead of relying only on open-loop locali-
zation based on manipulator forward kinematics.
Further emphasizing safety, the Assembly Monitor-
ing function enables automatic detection of invalid
conditions during the assembly. By comparing the
currently localized tiles in the observed scene with an
internally held ground truth model of each assembly
step, it raises an error whenever an out-of-bounds tile
of subassembly pose is detected. The error can then
be handled by the autonomy component to perform a
user-defined recovery or wait for operator inter-
vention.
4 CONCLUSION AND FUTURE WORK
This paper provides an overview about the PULSAR
project and, more specifically, presents the under-
water demonstrator dLSAFFE. With dLSAFFE, the
process of on-orbit modular assembly can be demon-
strated in a zero-buoyancy environment, mimicking
zero-gravity conditions.
Currently, the dLSAFFE demonstrator is being as-
sembled at DFKI. Parallelly, tests are being carried
out for the use of admittance control for mating of the
magnetic interfaces. The results from these tests will
lead to the development of the admittance control
strategy as well as the development of algorithms for
detection of successful mating between 2 interfaces.
After the assembly of the demonstrator, it will be
tested under water in the large water basin at DFKI.
These tests will be used to characterize the influence
of hydrodynamic forces on the system and thus, de-
velop methods to exert the required restoring force to
carry out the tile assembly as required. Following
this, the complete assembly of a large segment of the
mirror will be carried out. Subsequently. as the con-
troller for the robot manipulator system is being de-
veloped in ESROCOS, an evaluation of ESROCOS
will be carried out to validate its applicability in such
simulated scenarios. Having a functioning system
also provides a testbed for performing the tile assem-
bly with a free-floating base in the future.
Acknowledgement
The PULSAR project is funded under the European
Commission’s Horizon 2020 Space Strategic Re-
search Clusters Operational Grants, grant number
821858.
References
[1] J. P. Gardner, “About Webb Space telescope,”
James Webb Space Telescope, Sep. 18, 2020.
http://tinyurl.com/y3col462 (accessed Sep. 18, 2020).
[2] The LUVOIR Team, “The LUVOIR Mission
Concept Study Interim Report,” ArXiv180909668
Astro-Ph, Sep. 2018, Accessed: Sep. 18, 2020.
[Online]. Available: http://arxiv.org/abs/1809.09668.
[3] W. Brinkmann et al., “Space robotic systems and
artificial intelligence in the context of the European
space technology roadmap,” in Proceedings of Space
Tech Conferences 2019, Bremen, Germany, Nov.
2019, p. 12, Accessed: Sep. 14, 2020. [Online].
Available: http://tinyurl.com/yxf99a9k.
[4] M. M. Arancón et al., “ESROCOS: Development
and Validation of a Space Robotics Framework,” in
Proceedings of the 15th Symposium on Advanced
Space Technologies in Robotics and Automation
(ASTRA 2019), Noordwijk, The Netherlands, 2019, p.
8, Accessed: Sep. 09, 2020. [Online]. Available:
http://tinyurl.com/yynta4zr.
[5] V. Dubanche and S. Andiappane, “Development
of I3DS: An Integrated Sensors Suite for Orbital
Rendezvous and Planetary Exploration,” in Proceed-
ings of 'The 14th International Symposium on Artifi-
cial Intelligence, Robotics and Automation in Space -
iSAIRAS’, Madrid, Spain, Jun. 2018, p. 8, Accessed:
Sep. 15, 2020. [Online]. Available:
http://tinyurl.com/y6ovyu75.
[6] J. Vinals et al., “Future Space Missions with
Reconfigurable Modular Payload Modules and
Standard Interface – An Overview of the SIROM
Project,” in In Proceedings of the 69th International
Astronautical Congress (IAC-2018), Bremen, Ger-
many, 2018, p. 11, Accessed: Oct. 11, 2019.
[Online]. Available: http://tinyurl.com/y6xdryg9.
[7] A. Perrotin, E. Conquet, P. Dissaux, T. Tsiodras,
and J. Hugues, “The TASTE Toolset: turning human
designed heterogeneous systems into computer built
homogeneous software,” presented at the ERTS2
2010, Embedded Real Time Software & Systems,
Toulouse, France, May 2010, Accessed: Dec. 13,
2019. [Online]. Available:
http://tinyurl.com/y3v4zkaq.
[8] DFKI, “Maritime Exploration Hall,” Robotics
Innovation Center - DFKI GmbH, Sep. 18, 2020.
http://tinyurl.com/v5qmlgt (accessed Sep. 18, 2020).
[9] J. Wang and E. Olson, “AprilTag 2: Efficient and
robust fiducial detection,” in 2016 IEEE/RSJ Interna-
tional Conference on Intelligent Robots and Systems
(IROS), Oct. 2016, pp. 4193–4198, doi: 10/ghbm3h.
[10] M. Quigley et al., “ROS: an open-source robot
operating system,” in Proceedings of the IEEE In-
ternational Conference on Robotics and Automation
(ICRA), Kobe, Japan, May 2009, vol. 3, p. 5, Ac-
cessed: Sep. 16, 2020. [Online]. Available:
http://tinyurl.com/yysd3nhj.
[11] H. Bruyninckx, “Open robot control software:
the OROCOS project,” in Proceedings 2001 ICRA.
IEEE International Conference on Robotics and
Automation (Cat. No.01CH37164), May 2001, vol. 3,
pp. 2523–2528 vol.3, doi: 10/fc8w2w.
[12] DFKI, “the Robot Construction Kit (Rock),”
2019. https://www.rock-robotics.org/ (accessed Sep.
10, 2020).
[13] B. Asadi and S. Natarajan, “Motion Planning
for Manipulators,” presented at the RIC Project Day
“Framework & Standardization” and “Manipulation
& Control,” Bremen, Germany, Jun. 19, 2014, Ac-
cessed: Sep. 16, 2020. [Online]. Available:
http://tinyurl.com/y63e9h7g.
[14] I. A. Şucan, M. Moll, and L. E. Kavraki, “The
Open Motion Planning Library,” IEEE Robot. Autom.
Mag., vol. 19, no. 4, pp. 72–82, Dec. 2012, doi:
10/gdnbz5.
[15] M. Kalakrishnan, S. Chitta, E. Theodorou, P.
Pastor, and S. Schaal, “STOMP: Stochastic trajectory
optimization for motion planning,” in 2011 IEEE
International Conference on Robotics and Automa-
tion, May 2011, pp. 4569–4574, doi: 10/fxbq67.
[16] J. Schulman et al., “Motion planning with se-
quential convex optimization and convex collision
checking,” Int. J. Robot. Res., vol. 33, no. 9, pp.
1251–1270, Aug. 2014, doi: 10/f6j8k3.
[17] P. Beeson and B. Ames, “TRAC-IK: An open-
source library for improved solving of generic in-
verse kinematics,” in 2015 IEEE-RAS 15th Interna-
tional Conference on Humanoid Robots (Human-
oids), Nov. 2015, pp. 928–935, doi: 10/ghbtq4.
[18] R. Smits and E. Aertbelien, “Kinematics and
Dynamics Library (KDL),” Orocos Kinematics and
Dynamics, 2020. https://orocos.org/kdl (accessed
Sep. 10, 2020).
[19] J. Pan, S. Chitta, and D. Manocha, “FCL: A
general purpose library for collision and proximity
queries,” in 2012 IEEE International Conference on
Robotics and Automation, May 2012, pp. 3859–3866,
doi: 10/ggs7v9.
[20] E. Coumans and Y. Bai, “PyBullet, a Python
module for physics simulation for games, robotics
and machine learning,” 2019 2016. http://pybullet.org
(accessed Sep. 14, 2020).
[21] E. Gamma, R. Helm, R. E. Johnson, and J. Vlis-
sides, Design patterns: Elements of reusable object-
oriented software. Reading, MA: Addison-Wesley,
1995.
