sign in sign up
<<
Home , Robotic arm, Robotic spacecraft

SSC20-WKI-05

Ground-Based 1U CubeSat Robotic Assembly Demonstration

Ezinne Uzo-Okoro, Christian Haughwout, Emily Kiley, Mary Dahl, Kerri Cahoy Massachusetts Institute of 77 Massachusetts Ave. Cambridge, MA 02139, USA (301) 370-6888 [email protected]

ABSTRACT Key gaps limiting in-space assembly of small are (1) the lack of standardization of electromechanical CubeSat components for compatibility with commercial robotic assembly hardware, and (2) testing and modifying commercial robotic assembly hardware suitable for small assembly for space operation. Working toward gap (1), the lack of standardization of CubeSat components for compatibility with commercial robotic assembly hardware, we have developed a ground-based robotic assembly of a 1U CubeSat using modular components and Commercial-Off-The-Shelf (COTS) arms without -in-the-loop. Two 16 in x 7 in x 7 in dexterous robot arms, weighing 2 kg each, are shown to work together to grasp and assemble CubeSat components into a 1U CubeSat. Addressing gap (2) in this work, solutions for adapting power-efficient COTS robot arms to assemble highly-capable are examined. Lessons learned on thermal and power considerations for overheated motors and positioning errors were also encountered and resolved. We find that COTS robot arms with sustained throughput and processing efficiency have the potential to be cost-effective for future space missions. The two robot arms assembled a 1U CubeSat prototype in less than eight minutes.

INTRODUCTION (DARPA) Robotic Servicing of Today, as space becomes more accessible, Geosynchronous Satellites (RSGS) $400M there is a lack of affordable on-demand program aims to demonstrate that a robotic capability to address multiple government [1] servicing can perform safe, reliable, and commercial constellation needs for on- useful and efficient operations in or near the servicing and assembly. The industry’s Geosynchronous Orbit (GEO) first satellite life extension vehicle, Northrop environment. RSGS is using the custom- Grumman’s Mission Extension Vehicle-1 developed and large radiation-hardened Front- (MEV-1), completed its first docking to a client end Enabling Near-term satellite, Intelsat IS-901 on February 25, 2020. Demonstration (FREND) robot arm, which is a MEV-1 is designed to dock to geostationary 1.8 m arm from shoulder pitch to wrist pitch satellites whose fuel is nearly depleted and weighing 78 kg, with an additional 10 kg for does not make use of robot arms for its on-orbit electronics. servicing mission [2]. On-orbit robotic The need for low-cost, low-latency and agile assembly to date is costly, as evidenced by space infrastructure, which can reach strategic prior and current missions [3]. For example, the such as GEO and Low Earth Orbit (LEO) Defense Advanced Research Projects Agency in addition to polar and International Space

Ezinne Uzo-Okoro 1 34th Annual Small Satellite Conference Station (ISS) locations, could be realized using response time from a minimum of 30 days to a robotic assembly of modularized components less than 10 hours for a small satellite build and into CubeSats. A standardized modular deployment cycle. CubeSat and COTS-based robotic assembly This robotic-assembly based mission using could break the reliance on high-risk, high- propulsive “lockers” could help create a latency, high-cost legacy space hardware. resilient platform capable of rapidly Satellite cellularization [4][5][6][7] has made assembling and deploying scalable space incremental advances in the modularization of systems faster than NASA’s documented small satellite subsystems. However, this thesis minimum launch-on-demand response time (35 explores a new approach to CubeSat days) for the International (ISS) production based on the robotic assembly of crew rescue [8]. functional components. There are four phases necessary to successfully

Figure 1. Conceptual system design of the realize the mission concept. Phase 1 involves interior of the spacecraft locker. the ground-based robotic assembly of a CubeSat prototype using two dexterous arms We envision a new mission in which small and electromechanical components in a COTS robot arms are enclosed in free-flying laboratory environment and assessing different small spacecraft “lockers” of approximately 24 and options to optimize inches x 36 inches x 12.5 inches for the response time and sensing. This paper assembly of a new standard of small satellites. addresses the feasibility of Phase 1 and These mini-fridge-sized spacecraft “lockers” characterizes the systems engineering efforts with propulsion capability are intended to be required to develop in-space robotic assembly. orbit-agnostic in order to deploy on-demand Phase 2 involves the development and launch robot-assembled CubeSats where needed. The of a flight unit locker with robot arms and spacecraft locker houses two robotic arms, CubeSat modular components, including modular components including sensor and propulsion options for the CubeSats propulsion modules, and for 1U to themselves and not just the spacecraft locker. 3U-sized CubeSats. The mission is expected to The spacecraft locker would be hosted at the deliver an unprecedented improvement in ISS Japanese Experiment Module Exposed

Ezinne Uzo-Okoro 2 34th Annual Small Satellite Conference Facility (JEM-EF) [9][10][11], and house six Degrees of Freedom, and minimize on-orbit enough components to demonstrate the on- SmallSat assembly time by using the dexterous orbit assembly of five 1U CubeSats. The first robot arms while satisfying the given power prototype CubeSat would be a CubeSat consumption and weight requirements at a assembled on earth and deployed first in order given orbit. Thus, we employ multidisciplinary to test the deployment system. The four design optimization tools and methodologies remaining CubeSats - two with Radio Science focusing on the second key gap: testing and Experiments (RSE) and magnetometers, two modifying commercial robotic assembly with visible (VIS) sensors - will be robotically hardware suitable for small satellite assembly assembled on-orbit. The ISS Phase 2 in space. Given that the search parameters in technology demonstration is expected to prove the task for a robot with the on-orbit assembly of modular many degrees of freedom are constant, the reconfigurable CubeSats, increase Technology Genetic Algorithm approach in combination Readiness Level (TRL), and assess response with the robot simulation is used. We also time quantitatively. describe the technology choices and redundancy levels of the different subsystems Organization in this optimal on-orbit assembly design. Following a state-of-the-art review of current robot arms in space, a feasibility study on the COTS Robot Arm Flight Heritage use of dexterous COTS robot arms in space is To date, most on-orbit assembly missions are analyzed. We summarize the study results not for small satellite on-orbit assembly, but toward feasible on-orbit CubeSat robotic instead are designed to support ISS assembly. experiments, exploration, and servicing (refuel Next we describe the lab prototype or repair existing satellites) missions demonstration of the robotic assembly of a 1U [15][16][17][18]. Previous missions include CubeSat by two dexterous COTS robot arms. the U.S. Defense Advanced Research Projects The assembly process uses two COTS robot Agency’s (DARPA) Orbital Express program arms to assemble modularized boards fastened [19], the DARPA Program [20], and with magnets into a small satellite. The the Jet Propulsion Laboratory’s (JPL) assembly steps use open-loop control and a Insight mission [21]. Robotic manipulators, Python software program. We show that the important for scientific experiments and the assembly of modularized construction and maintenance of the ISS, have components is a viable option for a new class conducted on-orbit robotic assembly. of CubeSats. Examples include the Shuttle Remote Lastly, we provide a summary of the work and System (SRMS) [22], also known introduce the next steps for space qualification as Canadarm, which is a 16.9-meter, seven of the system. degree of freedom (DOF) manipulator with a relocatable base; the National Space FEASIBILITY OF ROBOT ARMS IN Development Agency of Japan’s (NASDA) SPACE Japanese Experiment Module (JEM) Remote We review the current state-of-the-art of robot Manipulator System (JEMRMS), which is a arms in space. We assess the feasibility of the 9.91-meter, six DOF manipulator; and lastly, on-orbit assembly of small satellites using the European Robotic Arm (ERA), which is an Commercial-Off-The-Shelf (COTS) hardware 11-meter, seven DOF manipulator [23]. without humans-in-the-loop. We select low- cost robot arms, LewanSoul xArm with

Ezinne Uzo-Okoro 3 34th Annual Small Satellite Conference Table 1. Shows the gaps in select current on- servicing. The National Aeronautics and Space orbit assembly/servicing space missions Administration (NASA) Goddard Space Flight

Center’s (GSFC) RESTORE-L servicing These manipulators employed very large mission [30], is also a robotic arms to deploy, maneuver, and capture equipped with the tools to rendezvous with, payloads. Hirzinger’s patent on a multisensory grasp, refuel, and relocate satellites to extend robot was tested aboard the Columbia shuttle, their lifespan. Lastly, NASA’s Dragonfly has which successfully worked in autonomous also recently demonstrated a ground-based test modes, and was teleoperated by , as of robotic satellite assembly [15][31] and Made well as in different telerobotic ground control In Space (MIS) received a large NASA contract modes [24][25]. In the area of autonomy, to demonstrate on-orbit assembly using three SPHERES Universal Docking Port (UDP) robot arms to assemble 3-D printed parts in demonstrates autonomous docking maneuvers space, called the Archinaut mission [32]. using small satellites [26] and AstroBees, the free-flying robots, provide a flexible platform Benefits and Implications for In-Space for research on zero-g free-flying robotics [27]. Manufacturing Current missions (on-orbit or in development) The transferable technology includes a new include the MEV-1 and CubeSat standardization of mechanical, DARPA Robotic Servicing of Geosynchronous electrical, power, and thermal components, the Satellites (RSGS) program [28]. RSGS aims to modularity of key spacecraft elements with demonstrate satellite servicing mission different selectable sensors and/or propulsion operations on operational Geosynchronous units and a custom-built spacecraft locker that Earth Orbit (GEO) satellites. can be deployed at various orbit-agnostic RSGS uses the FREND project, which locations such as in LEO and GEO for asset developed the state-of-the-art in autonomous monitoring and constellation reconstitution. A rendezvous and docking with satellites not pre- comparison with the alternative - placing designed for servicing, and was the precursor ready-made CubeSats in an on-orbit locker - and inspiration for the DARPA RSGS program has been designed with a focus on packaging [29]. The DARPA/Naval Research Lab (NRL) efficiency for launch. We show that a custom- team working on FREND focused on configured locker filled with components for autonomous rendezvous and docking with on-orbit assembly is more efficient by 2x. satellites, which were not designed for Reusability of CubeSat electrical/mechanical

Ezinne Uzo-Okoro 4 34th Annual Small Satellite Conference components and propulsion systems can help wait-time or launch from the ground. The systems evolve to create different form factors. spacecraft locker could assemble and deploy Reusability of CubeSat electrical/mechanical the needed CubeSat solution within hours for components and propulsion systems can help rapid-response. Note that if there exists pre- systems evolve to create different form factors. integrated spacecraft on-the-ground, a launch There have been subsystem-focused spacecraft manifest, with a minimum of 35 days, is still as part of a swarm, limited by pre-built or pre- required. integrated spacecraft components. No program Similarly, for constellation applications, currently offers the ability to respond to several future cloud networks like DARPA’s emerging needs with any of the above BlackJack intend to produce using a satellite configurations, which could result in high- constellation that makes use of several nodes. performance target acquisition and unmatched If a node goes down, there are two options for pointing and stabilization accuracy. recovery. The node needs to be replaced, or the The concept of operations for modular number of satellites on a plane would need to CubeSats assembled in space would partition be enlarged to close the link. The spacecraft the spacecraft into modules, which would be locker would be available to robotically configurable into a wide variety of assemble a CubeSat with provided payload applications. Future use cases using different requirements to replace the node within hours. sensors on the spacecraft with variations in , sensors, propulsion, etc. Robot Arms for Assembly would benefit multiple applications including Having purchased robot arms in the under $500 constellation reconstitution and LEO and GEO range, with damaged servo motors by the first asset monitoring. test of the concept, we assessed a list of For example, if there exists an issue with a LEO replacement servo motors (see Table 2). We asset, and an inspection is required quickly, the define low-cost for the lab prototype as under spacecraft locker in LEO could robotically $500 for all robots, boards, and parts. Thus, in assemble a CubeSat with an RF sensor to listen, order to stay within the bounds of a low-cost a radar, or optical capability to respond. There system, we selected the HiWonder servo may be a need for two propulsion systems or motors, which are used on the LewanSoul chemical propulsion to arrive at the LEO xArm robots. location as quickly as possible. The spacecraft locker in LEO - a smart locker with all Table 2. Select list of common low-cost motors components at the ready - would require no for robot arm use

Ezinne Uzo-Okoro 5 34th Annual Small Satellite Conference Low-cost robotic arms such as the LewanSoul anticipate we will encounter challenges xArm shown in Figure 2 are controlled using designing robot arms (and its environment) that servo motors, which lack the power and torque would match an existing product and does not required for space missions. require a custom build, given in-space satellite assembly requirements. To restrict the scope of the design optimization, we first test a model simulation of two degrees of freedom. Figure 3 shows a 2-DOF simulation of the xArm robot arm in PyBullet. We know that the simulation framework (PyBullet) is able to accommodate this level of fidelity in simulations because the framework has precedent [33].

Figure 2. (a) LewanSoul xArm Robot with 6-

DOF (b) Robot Arm with dimensions. (Source: LewanSoul)

Given that the robots will be housed in and perform functions in a spacecraft locker, we continue to use the servo motors as an adequate lab prototype test for feasibility. As we progress to space qualification, appropriate Figure 3. Simulation of two xArm robots in motors for space operations will be used. PyBullet with 2 DOF Robotic arms of similar size and weight come in a range of prices. A major difference The model simulation returns 19.09 seconds, between the robot arms available off-the-shelf from a starting point of 91 seconds. While not is the use of powerful motors and sophisticated optimal, the output value is reasonable because control systems. The more sophisticated robot the robot arm requires 5 seconds to grasp and arms are able to move with greater precision 10 seconds to move an object to a drop-off due to these characteristics. location, and less than 5 seconds to snap- We recognize that these (and similar) COTS assemble the part. Therefore, in order to grasp robotic arms can be customized by adding and move an object, the robot arm, which is additional sensors or swapping particular positioned within 2 mm of the target and drop- components such as a motor or link. The off locations, 19 seconds to grasp, move and software used to control the arms is also usually drop-off an object is correct. However, the supplied with customization for controlling global optimal value might be out of reach due system parameters; however, a different to power constraints on the servo motors on the control computer and real-time operating robot arms as PyBullet did not find an system could be used. The interaction between optimum. the control parameters and the physical During simulation with different parameters, dexterity can be complex due to we see that using a powerful motor at high latencies and multi-tasking proportional gains results in faster (more using the operating system. Thus, we make optimal) time values, but consumes more assumptions and verify using the existing power. Conversely, using a high gain value physical prototype. Since most space robot with a weak motor results in oscillations that arms used in space tend to be custom-built, we take time to dampen out to within the 2 mm

Ezinne Uzo-Okoro 6 34th Annual Small Satellite Conference tolerance and hence result in longer vs Robot Assembly Time—We construction time. This simulation is for a contrast robotic assembly with CubeSat single step in a series of steps that are needed assembly requiring humans-in-the loop for for the full assembly of a satellite. In later assembly. CubeSats are usually assembled by a iterations, task planning to sequence the team of people and not robots. Hence, there is assembly steps are added and obtain similar little baseline data available to assess how long results. Using the AL5D 4-DOF robot arm kit, it might take to assemble a CubeSat using which resulted in servo burnouts after less than robots. 100 hours of tests, we modeled six degrees of freedom in a second search for low-cost robots. Table 4. Assembly time of various CubeSats Using six motors instead of two motors completed by human teams resulted in higher power calculations with Satellite Assembly Data Source about the same assembly time. A comparison time by of power and assembly time is provided in teams Table 3. NASA Several Email Table 3. Simulation results for 2-DOF and 6- MarCo months by a correspondence DOF robot performing the same task with the CubeSat large team with JPL same motor parameters Robot Initial Energy Peak Interorbital 2 days by a Email Assembly Used Power IOS team of 2 correspondence Time Used CubeSat people with 2.0 manufacturer 2-DOF 91 seconds 220.2 13.0 Watt- Watts One Conference seconds Small spacecraft per Paper [34] Satellite day 6-DOF 51 seconds 451.1 33.5 Watt- Watts MakerSat-1 5 minutes in Conference seconds International Paper [35] Space Station by 1 We discover that the arm with 6-DOF uses more power while performing the same task at the same speed with greater accuracy. These We begin by estimating how long it takes a tasks include grasping a part from a shelf and human team to assemble a 1U CubeSat as a bringing it to the assembly area and snapping final integration step. Note that this is the final two parts together, using some assumptions on step after the common components and payload force and alignment required for assembling subsystems have been designed, manufactured, LEGO-like parts together. This leads to the and are ready for integration. selection of a 6-DOF robot arm for this work. From estimates obtained in Table 4, we focus Table 2 lists the resulting robot arm options. on MakerSat-1, a 1U CubeSat, which is the The LewanSoul xArm robot arms offered a closest to our concept of using pre-developed low-cost option with reliable results (more than subcomponents to rapidly assembly satellites 170 hours of tests before burnouts) and less (with or without variant payloads) with power and thermal considerations. minimal human intervention in space using lightweight robotic arms. MakerSat-1 was

Ezinne Uzo-Okoro 7 34th Annual Small Satellite Conference designed with similar intentions for rapid input control onset times along the three axes. assembly. The first version of MakerSat-1 was The trade study evaluates five sensors (see released from the International Space Station Table 5) using weighted assessments for the and was able to collect ionizing radiation maximum payload the sensor can grasp, degree particle count in-orbit and experiment on of freedom offered, and weight of the sensor. polymer degradation while operating in space Given that a Level 1 requirement includes the for at least nine months [36]. (A video movement of a maximum payload of 2 kg, the demonstration of assembly under five minutes sensors were rated with the highest weights is available [37]). Thus, we use five minutes as going to the sensor to meet the 2 kg maximum our starting point for CubeSat assembly. To payload requirement. Having six or more make our simulation similar to the MakerSat-1 degrees of freedom offered on a COTS robot assembly, we need at least 2 robot arms: one arm also meets the topological requirements for arm to hold the partly assembled satellite while grasping components. Lastly, since the robot using the other arm inserts and clicks together kit is required to weigh less than 3 kg, sensors parts gathered from a shelf. We allow for that weighed the least were given a higher further model refinement of robot arm rating. When tallied, the weighted assessment functions as the grippers and different motors outcome favors brushless motors and force- in the robotic arm need to be accurately torque (FT) sensors, which are used at the end- modeled. Most importantly, we use five effector (gripper). minutes as a metric for the on-orbit satellite assembly of a 1U CubeSat. Table 5. Sensor Study Outcome, where WA is the Weight Assessment of each parameter

ROBOTIC ASSEMBLY OF A 1U Brushless motors are the preferred motors for CUBESAT space operations [39]. Most robotic The efficient use of sensors on dexterous robot applications require a multi-axis or six-axis FT arms is critical towards achieving a high- sensor to give feedback to the robot about the performance sensor/agility combination, end-effectors, which can be controlled along particularly for space applications [38]. High six axes (three translations and three rotations) performance metrics include a 95% success [40]. To measure the effort in all six axes, the rate on indicators of task completion times, FT sensor usually combines information from distance traveled, inverse motion, maximum a minimum of six unitary measuring elements velocities, amount of multi-axis control, and such as strain gauges. Using the geometry of

Ezinne Uzo-Okoro 8 34th Annual Small Satellite Conference the measuring elements, the force and torque The robot arms The objective of the are computed along the axes and used in the shall sense, grasp, research is to assemble a loop. FT sensors can also be and assemble functional 1U CubeSat leveraged for sensitive tasks including spiral CubeSat using both robot arms and linear search, rotational insertion, and path components recording [41][42]. Requirements—For in-space robotic assembly Six degree-of- The selected topology to be feasible, we propose that the robot arms freedom (DOF) provides sufficient meet the Level 1 requirements in Table 6. All arms with a maneuverability - parts are expected to be examined for resilience kinematic forward/backward, in a space-relevant environment, by running a configuration of up/down, left/right (in thermal vacuum test of all parts. Parts which yaw-pitch-pitch- three perpendicular axes) cannot be space qualified will be swapped out pitch-yaw-roll combined with rotation or sealed, where necessary. Sensors and about three perpendicular generic servo motors are used for the axes with 95% accuracy movement and rotation of the joints. The - to satisfy sense and following include requirements for the sensors grasp requirements, and servo motors: including partial single- ● The system shall include one six-axis fault tolerance, of 1U wrist force-torque sensor that measures the components wrench (three forces and three torques) at the end-effector The robot arm Brushless motors present ● Each of the four joint torque sensors joints shall be a feasible in-space option shall include redundant strain gauge bridges driven by without wear and tear that measure the output torque of each of the brushless motors, associated with Foreign joints, attached to the output of each of the first with a 30:1 gear Object Debris (FOD) four joints of the arm ratio and 256- ● The end-effector shall include link count magneto- strain gauges that measure bending and twist resistant encoders strains for each of the links ● Each servo motor shall have one motor Each robot arm The mass of the end- current sensor that measures the motor current shall weigh a effector and inertia of each servo motor of the arm maximum of 3 kg required to move a 1U ● Each motor shall be controlled using a CubeSat motor controller board. Each robot arm The mass of a 1U Table 6. Level 1 requirements for this work shall move a CubeSat, which is the maximum mass final assembled object, Requirements Rationale of 2 kg weighs a maximum of 2 kg The robot arms For the “buy and fly” shall perform COTS arms to be Each robot arm The robot arms are CubeSat feasible in space, the shall have a expected to be enclosed assembly arms must be used in a minimum arm with components within 1 functions in a locker with a thermal length of 1 m and m of reach. The spacecraft management system for a maximum spacecraft locker thermal control length of 2 m

Ezinne Uzo-Okoro 9 34th Annual Small Satellite Conference accommodates 2.5 m in dynamics equation of length motion for the system.

The robot arms Inverse Kinematics is The robot arms The camera provides the shall use Inverse used to initialize a shall sense target pose of the 1U CubeSat Kinematics rotating angle for each position using a components to the algorithms to servo. Forward camera software to steer the sense and reach Kinematics is used to capture trajectory and to components compute the current determine when the target position. (It should component is within the be noted that Forward robot arm’s capture Kinematics is used to envelope build a position relationship with the Approach base-attached servo and Sensing and grasping parts by robot arms have the end-effector, also been conducted in space since the 1970s known as the impactive [43][44][45] to aid astronauts with assembly or gripper.) repair tasks [46]. The recent successful ground demonstration of NASA’s Dragonfly mission The robot arms After comparing the by Space Systems Loral (SSL) [47] highlights shall use Velocity current target position the feasibility of assembly without humans-in- Kinematics for and goal position to the-loop with a custom-built robot arm. To target position output an error, Velocity grasp components, several COTS robot arms error correction Kinematics is used to with impactive grippers [12][32] are assessed. calculate the updated Flow of Inputs and Outputs—The block rotating angles. Velocity diagram in (Figure 4) depicts the data Kinematics is employed connections from each servo motor to the as a gradient to minimize controller board. The following block diagram error (to a threshold) and depicts the flow of inputs and outputs from the output rotating angles for robot arm kit. The FT Sensor is trained to each servo receive software commands from a computer, The robot arms The first tests to assemble which are passed to the Servo Controller board. The camera provides the pose of the 1U shall be mounted a CubeSat will be focused on a static on assembly of a CubeSat components to the software to steer platform for functional CubeSat. the capture trajectory and to determine when initial laboratory Further tests for space the component is within the robot arm’s tests qualification will assume capture envelope. The Servo Controller sends a a dynamic to the servo motors using amplifiers. environment The servo motors execute the command on the arm joint (shoulder, elbow or wrist) or sensor The robot arms Reaction wheels will be head rotation. The encoded action is sent back shall be mounted used to control the to the Servo Controller board, through the on a platform for orientation of the base of serial port, to the computer. space-qualified the robot arms using a applications

Ezinne Uzo-Okoro 10 34th Annual Small Satellite Conference Figure 4. Robot Assembly Block Diagram either a waste of power to reclaim lost fasteners depicting input and output flow from the or an excess of fasteners to be stored in the system. (All servo motors serve the same locker. function and possess the same characteristics.) Two alternative methods of attachment were considered for the structural design, as seen in The computer interprets the action and sends Figure 5. The first design utilizes latches, or additional commands to the FT Sensor. While small outcroppings, in the top and bottoms of brushless direct current (DC) motors will be the rails that can slide into the base and top of used for the space-qualified test, the the CubeSat but cannot be pulled back out LewanSoul pre-packaged kit-provided servo without first applying pressure. This design motors are used for the initial laboratory was ultimately rejected due to lack of space; the prototype. size of the latch required to secure the rails in place was infeasible. The second mechanical Mechanical Workmanship attachment involves snaps. The panels would Several iterations of structural designs were be placed into a base and held by buttons that necessary to align with the capabilities of the fit into holes in the panels. The rails would then robotic arm and the lack of a human-in-the- snap into knobs on the outside of the structure. loop. There were two primary criteria This design was used for the robotic assembly. determined for the structural design. First, all There were several instances when the low-cost pieces had to be large enough for the robot arm robot could not provide enough torque to place grippers to hold. Second, the design could not parts into the knobs, so the knobs had to be be held together with mechanical fasteners, shaved down by 35%. Additionally, the panels such as springs or screws. This was for two were often not able to be precisely placed into reasons. The first is the limitations of the position, so several tests were required to grippers; screws are both too small and require improve precision and a camera, originally too fine precision to install with the robot arms. used to improve lighting, aided precision. The second is the low gravity environment. The We used additive manufacturing with a Fused limitations on the speed and precision of the Filament Fabrication (FFF) printer for the robot arm would prevent it from recovering a design iterations for this work as it is effective fastener if it was improperly placed and for laboratory prototyping purposes. We released in the low gravity environment. This anticipate the final design will be 3D-printed would cause both time delays in assembly and using a Selective Laser Sintering (SLS) printer,

Ezinne Uzo-Okoro 11 34th Annual Small Satellite Conference as SLS printers have better outgassing We approximate the assembly workspace and properties than FFF printers. 3D-printing use a discrete model to capture the reachable provides us with multiple advantages to space of the robot arms’ capabilities. machining. First, it emphasizes the low-cost and rapid production goals of this mission, as Implementation 3D-printing is both faster and cheaper than The LewanSoul robots are selected because it machined parts. Second, it allows for fine detail is a low-cost COTS option. A Raspberry Pi and features that would be challenging to be camera is set atop a 1.5 ft tall post with an machined. Arduino attached behind it. Red prototype boards in front of the two robot arms. The process begins with the Raspberry Pi camera capturing an image of the platform. OpenCV object detection software libraries are used in a Python software program to identify the color- coded boards, calculating the center of the boards for grasp accuracy (by converting pixels to meters). After an image capture of the field, the pixel position of the boards’ center is calculated, resulting in two sets of (x, y) points Option 1 with rails and latches in meters and pixels. The maximum range for the LewanSoul robot arms is +0.15 m to -0.15 m in the y-direction and 0 m to 0.3 m in the x- direction, which determines the placement of each arm and board stacks. Using Inverse Kinematics [26], the location values are detected and converted into a set of six angles. Given that there are six servos on each arm, the Option 2 without side rails Raspberry Pi would send those angles to the Figure 5. Two current best structural options Arduino for control of the arm through control of the six servos via the USB serial port. The arm proceeds to perform movements to grasp each board at target locations and begin assembly using specified location values. The process is repeated until a CubeSat has been assembled - see assembly sequence in Figure 7.

Figure 6. A rendition of the Camera Mount used on a 1.5 ft post to aid robotic assembly

After a feasible structure is selected, we create and use a representation of the assembly workspace (inside a spacecraft locker) to determine feasibility for robotic task execution.

Ezinne Uzo-Okoro 12 34th Annual Small Satellite Conference Figure 7: Sequence of Robot Arm Assembly range is between 0 and 240 degrees and the of 1U CubeSat in under 8 minutes. minimum increment, or accuracy, for each servo is 13.8 degrees. Using each servo’s Robotic Algorithm unique ID number, their rotating duration and To ensure the robot arms reach the target rotating position are controlled. In the Arduino boards with a single calculation, open-loop code, we pre-defined several functions that control, which is faster than closed-loop move to the vertical initial position, move to control, is used. Control operations can be target location based on input arguments, and either closed-loop or open-loop. The key move to bin location - difference is feedback. An open-loop control move_to_initial(), move_to() and system performs based on the input, and the move_to_bin(), respectively. Serial Port is output has no effect on the control action. used to make Raspberry Pi 4 (RPi4) Closed-loop control is best used when the communicate with the Arduino. Six values are measurements are feasible, and the process has sent for each target location, one angle for each a predictable response to an input control. It servo. The Arduino has only one serial port and enables the process to be set on certain points needs to communicate with both the RPi4 and within a given accuracy and automates the six servos, so an additional hardware serial correction to process disturbances. Yet with port was set up. A protocol requiring the RPi4 open-loop control, outputs rarely change and and Arduino to communicate and confirm process disturbances are not the norm. messages was added to ensure all six values Therefore, we select open-loop control as the were sent and for use as an error detection better choice because no quantitative mechanism. Once complete, the camera would measurement is possible, as with an capture a new image of the boards to be inaccessible or erratic process, and low-cost is processed. a priority. Coverage Planning Functions—For task The Software Serial Port on Arduino is used to planning in 2D workspaces, we determine the control the six different servos, restricting the viewpoints for the entire target surface using a rotating limits for each servo first. The rotation Ezinne Uzo-Okoro 13 34th Annual Small Satellite Conference randomized sampling. This randomized steps in the loop, we update the target points sampling as defined in the Traveling Salesman Tcoverage, all by adding the points in Problem (TSP) [13] enables three functions. Tcoverage and adding (pmax, t) to the The functions are the selection of components, list of solutions. Upon completion of the assembly of components in required time (<50 while loop, we find the degree of coverage. seconds), and connect the components using the Genetic Algorithm in Section 3. We assume Observations that path planning between specific goals The LewanSoul arms assembled structures and dominates the runtime cost compared to the six prototype boards in under eight minutes. computation of approximate solutions to the The robot arms were subjected to 170 hours of TSP. It uses a lower bound estimation of the tests: all servo motors and rotation angles were path length between goals to calculate tested to determine stability, accuracy, and candidate TSP solutions and uses the complete feasibility of operation. We automated path planner for edges in candidate solutions. repetitive tests of each servo motor for over 120 The robotic arm control algorithm [48] finds an hours, while the software for the satellite appropriate solution to two key problems: path assembly was being programmed. There were planning and robot arm placement. This is initially errors in assembly as the robot arms accomplished by using a divide and conquer kept missing the precise assembly area, strategy and optimization heuristic planning structure spaces for board placement, and the approaches to the reachability and the coverage correct angle for side boards. Therefore, while problem. The algorithm shows how we sample the boards were grasped within the first week points from the target surface and use the points of programming, we learned after five weeks of TStarget to estimate the progress of the errors in placement to slow the speed of the arm coverage planning. We store all the points in movement by a factor of two as the robot arm the solutions in the set Tcoverage, allT approached the satellite assembly area. For and align each pose (from the global set) pA, instance, if the board was picked up by the with reachable target points from the robot arm moving in 1480 ms, we also move predefined map of reachability. The main loop each servo motor (robot arm joint) in 1480 ms continues after this phase until R is empty or all as the board approaches its final destination. target points are covered. Next we find the pose When the board arrived at the assembly area, pmax, which includes the largest subset of the the board was lowered carefully into its rest of the target points. We also use the intended position in 740 ms. Despite this coverage planner to find a trajectory t in as slowdown, the robotic assembly of each many points as possible in R(pmax), which component took approximately 22.25 seconds. are stored in Tcoverage. Constraints like It took the same amount of time to grasp stability requirements are taken into account by mechanical structures as it did to grasp boards. the coverage planner. Tcoverage and pmax Additional issues arose during the assembly are removed from R by updating every process, such as loose grippers. The grippers entry(p',T')R to remove the covered became loose after over 100 hours of use and were not able to pick up the boards, which were points (p', T'\Tcoverage). Entries sliding off the gripper pads. The grippers were with no reachable points are emptied during subsequently tightened. On occasion, electrical each timestep, which is 10 ms. Given the multi- tape was used on the gripper pads to retrain the step required, we use the Python time() gripper into a gripping position. function, to measure time and create a function to configure the clock and evaluate the microcontroller at 100 Hz. And for the last

Ezinne Uzo-Okoro 14 34th Annual Small Satellite Conference The standard libraries used are pybullet, which includes calculateinversekinematics(), pybullet_data, math, time, datetime, and numpy. Summary Overall, the robots have shown the capacity to assemble a 1U lab prototype CubeSat in under eight minutes. However, power considerations require improved motors for ISS demonstration as servo motors burnout due to degradation Figure 8. Robot Arm Power Consumption and after less than 200 hours of use. The end- repeatability of movements are predictable effector (gripper) accuracy diminishes with while the accuracy of the robot arms decreases time; therefore, exploration of precision after 120 iterations (surgical) robots for flight is a required next step. Two COTS robot arms and servo motors The camera lighting control was coded into the have shown reliability concerns due to Python program; however, PiCam lighting mechanical and degradation issues on the control was used to ensure adequate lighting at ground; therefore, conducting a future trade all times. Sometimes the LED did not work; study on low-cost offerings for reliable motors tightening the bolts of the LED and camera, and arms is key to moving forward. then restarting it solved the problem. Standardization of electromechanical CubeSat Ultimately, errors were resolved, and the entire components for on-orbit assembly requires 1U CubeSat was assembled with no humans- magnets and snaps for low-cost end-effectors. in-the-loop in seven minutes and 39 seconds. The potential for decreasing the lead time for Using the Inverse Kinematics (IK) approach CubeSat integration and assembly and savings made for less intensive programming; in cost and schedule serve as justification to however, using a robot arm as part of a larger continue to refine and implement this work. It system required a learning curve in robot is clear that ultimately, the function rests with and robotics programming. The the robot arms. Available to support Python code resulted in several hundreds of spacecraft development will foster faster lines of code, which was human-intensive to scientific research and discoveries and would create. There was a decline in the robot arm reduce schedule and cost (by an estimated 95% accuracy requirement after 120 iterations. 50%) associated with building a small satellite. We observed the robots become physically We find that large custom-built robots are not shaky and technically imprecise. For instance, the only for in-space robotic assembly. although the robot arms were programmed with There is utility for precision robot companies the correct coordinates, it kept missing the to support aerospace robotic applications. As structure by 2 cm when installing a board. We small satellites and constellation missions added an error detection to the code and continue to evolve, demand for precise and adjusted the distance for assembly in the rapid CubeSat assembly with no humans-in- assembly area to match 92-98% of the specified the-loop will grow. coordinates. All programming and CAD work was completed on an Apple Mac laptop with access to and use of standard Python libraries.

Ezinne Uzo-Okoro 15 34th Annual Small Satellite Conference CONCLUSION AND FUTURE WORK precision surgical robot arms as the latter is On-orbit robotic assembly missions typically likely to be costly and may negate the low-cost involve humans-in-the-loop and use large goal of the research. An optimization model, custom-built robotic arms designed to service which simulates next-generation design and existing modules. The concept of on-orbit performance, to ensure energy optimization per robotic assembly of modularized CubeSat CubeSat assembly will be conducted. We also components supports use cases, such as rapidly intend to conduct environmental testing of the placing failed nodes within a constellation of robot arms and assess the thermal and power satellites and monitoring damaged assets in budgets for lifetime expectation and self- Low Earth Orbit. This work describes the maintenance. In addition, we use a spacecraft potential and approach to on-orbit robot locker with thermal management control to assembly of small satellites using low-cost reduce the risk of thermal concerns. Steps to robot arms. We show the feasibility of the improve the torque of the robot arms will be robotic assembly of a 1U CubeSat and optimize included in the space qualification for robotic assembly time. We demonstrate the requirements. Three activities must be laboratory prototype assembly of a 1U CubeSat conducted for a flight model. These activities and analyze the systems engineering process are for the on-orbit assembly of small satellites. 1. The modularization of sensor payloads The ground-based lab prototype has shown that 2. The design and test of the locker, robotic arm assembly of modularized shelving and storage units for component components could be proven as a viable option modules including robotic arm accessibility for a new class of CubeSats. The assembly 3. The build and test of FlatSat component process used two dexterous COTS robot arms modules. to assemble modular CubeSat boards fastened Future Work with magnets into a small satellite. The As CubeSat subsystems continue to mature, the assembly steps for a 1U satellite, using open- project will evaluate relevant components and loop control and a Python software program, payloads for robotic assembly testing and required approximately five minutes to analysis. Future work will be focused on three complete. objectives: Flight Hardware Selection 1. The introduction of a new CubeSat Observations and lessons learned from structural standard. The standard uses modular feasibility studies, analyses, and the robot arm and reconfigurable electromechanical demonstration have informed several flight components, which includes providing considerations and highlighted the need for propulsion capability, should the CubeSat need several future work efforts, such as to change orbits. investigating improved subsystems. For 2. The demonstration of space-qualified instance, considering precise (surgical) robot robotic assembly of a 1U CubeSat. A key step arms in the same form factor as the LewanSoul for space qualification involves the calculation robot arms to overcome accuracy and precision of the link budget. The link budget is a issues and exploring durable motors for the theoretical calculation of the end-to-end flight demonstration, with low risk for burnout. performance of the communications link. We will train these new robot arms to sense, 3. The tailoring of the systems grasp, and assemble CubeSat flight engineering process to robotic small satellite components. We also need to conduct a trade assembly with no humans-in-the-loop. study on low-cost COTS robot arms versus

Ezinne Uzo-Okoro 16 34th Annual Small Satellite Conference Acknowledgments (CSCS)/Rescue Flight Resource Book.” We thank Chad Frost and Marcus Murbach National Aeronautics and Space (NASA ), Dr. Neil Administration (2005): 89. Gershenfeld (MIT Center for Bits and Atoms), 9. Kawasaki, Kazuyoshi. "Overview of Dr. Daniel Hastings (MIT AeroAstro JEM-EF on ISS." Proceedings of the RIKEN Department) and Joseph Parrish (DARPA) for Symposium. Saitama. 2008. valuable insights to refine portions of the 10. Steimle, Per C., et al. "Commercial concept from a back-of-the-envelope idea to a Approach to Research Outside the potential relevant mission concept. International Space Station-A Small Size Precursor Service For Future In-Orbit References Testing." AIAA SPACE 2014 Conference and 1. Doggrell, Les. Operationally Exposition. 2014. responsive space: a vision for the future of 11. Steimle, Christian, and Uwe Pape. military space. AIR UNIV MAXWELL AFB "ISS External Payload Platform-a new AL AIRPOWER JOURNAL, 2006. for research in the space 2. Northrop Grumman. "Companies environment." 40th COSPAR Scientific demonstrate groundbreaking satellite life- Assembly. Vol. 40. extension service." [Online]. Available: 12. , Yongjun, et al. "Design and https://news.northropgrumman.com/news/rele optimization of a novel six-axis force/torque ases/northrop-grumman-successfully- sensor for space robot." Measurement 65 completes-historic-first-docking-of-mission- (2015): 135-148. extension-vehicle-with-intelsat-901-satellite 13. Applegate, David L., et al. The 3. Flores-Abad, Angel, et al. "A review traveling salesman problem: a computational of space robotics for on-orbit study. Princeton university press, 2006. servicing." Progress in Aerospace Sciences 68 14. Sinclair, Doug, and Jonathan Dyer. (2014): 1-26. "Radiation effects and COTS parts in 4. Kerzhner, Aleksandr A., et al. SmallSats." (2013). "Architecting cellularized space systems using 15. Piskorz, D. A. N. I. E. L. L. E., and K. model-based design exploration." AIAA Jones. "On-Orbit Assembly of Space Assets: SPACE 2013 Conference and Exposition. A Path to Affordable and Adaptable Space 2013. Infrastructure." The Aerospace Corporation 5. Hill, Lisa, et al. "The Market for (2018). Satellite Cellularization: A historical view of 16. Katz, Daniel S., and Raphael R. Some. the impact of the satlet morphology on the "NASA advances robotic ." space industry." AIAA SPACE 2013 Computer 36.1 (2003): 52-61. Conference and Exposition. 2013. 17. Putz, Peter. "Space robotics in Europe: 6. Barnhart, David, et al. "Changing A survey." Robotics and Autonomous satellite morphology through cellularization." Systems 23.1-2 (1998): 3-16. AIAA SPACE 2012 Conference & 18. Weisbin, Charles R., and Guillermo Exposition. 2012. Rodriguez. "NASA robotics research for 7. Jaeger, Talbot, and Walter Mirczak. planetary surface exploration." IEEE Robotics "Satlets-the building blocks of future & Automation Magazine 7.4 (2000): 25-34. satellites-and which mold do you use?" AIAA 19. Whelan, David A., et al. "Darpa orbital SPACE 2013 Conference and Exposition. express program: effecting a revolution in 2013 space-based systems." Small Payloads in 8. Ceccacci, Anthony, Dye, Paul. “Contingency Shuttle Crew Support

Ezinne Uzo-Okoro 17 34th Annual Small Satellite Conference Space. Vol. 4136. International Society for 30. Reed, Benjamin B., et al. "The restore- Optics and Photonics, 2000. L servicing mission." AIAA SPACE 2016. 20. Barnhart, David, et al. "Phoenix 2016. 5478. program status-2013." AIAA SPACE 2013 31. Lymer, John, et al. "Commercial conference and exposition. 2013. application of in-space assembly." AIAA 21. Smrekar, Sue, and B. Banerdt. "The SPACE 2016. 2016. 5236. InSight mission to Mars." The 8th Mars 32. Patane, Simon, John Schomer, and Conference. Vol. 18. 2014. Michael Snyder. "Design Reference Missions 22. Sallaberger, Christian, Space Plan for Archinaut: A Roadmap for In-Space Task Force, and . Robotic Manufacturing and Assembly." 2018 "Canadian space robotic activities." Acta AIAA SPACE and Astronautics Forum and astronautica 41.4-10 (1997): 239-246. Exposition. 2018. 23. Laryssa, Patten, et al. "International 33. James, Stephen, et al. "Sim-to-real via space station robotics: a comparative study of sim-to-sim: Data-efficient robotic grasping via ERA, JEMRMS and MSS." 7th ESA randomized-to-canonical adaptation Workshop on Advanced Space Technologies networks." Proceedings of the IEEE for Robotics and Automation. 2002. Conference on Computer Vision and Pattern 24. Hirzinger, Gerd, et al. "Sensor-based Recognition. 2019. space robotics-ROTEX and its telerobotic 34. Gilmore, Cheser, et al. "Flexible, High features." IEEE Transactions on robotics and Speed, Small Satellite Production." (2019). automation 9.5 (1993): 649-663. 35. Grim, Braden, et al. "MakerSat: A 25. Hirzinger, G., et al. "Robotics and CubeSat Designed for In-Space 3D Print and mechatronics in aerospace." 7th International Assembly." (2016). 30th Annual Conference Workshop on Advanced Motion Control. on Small Satellites. Proceedings (Cat. No. 02TH8623). IEEE, 36. Earth Observatory Portal Directory. 2002. “MakerSat”. [Online]. Available: 26. Rodgers, Lennon, Simon Nolet, and https://directory.eoportal.org/web/eoportal/sat David W. Miller. "Development of the ellite-missions/m/makersat miniature video docking sensor." Modeling, 37. NNU. “NNU’s MakerSat-1 CubeSat Simulation, and Verification of Space-based Assembly.” [Online]. Available: Systems III. Vol. 6221. International Society https://youtu.be/shLPETczsF4 for Optics and Photonics, 2006. 38. Hirzinger, Gerd, et al. "On a new 27. Bualat, Maria, et al. "Astrobee: generation of torque controlled light-weight Developing a free-flying robot for the robots." Proceedings 2001 ICRA. IEEE international space station." AIAA SPACE International Conference on Robotics and 2015 Conference and Exposition. 2015. Automation (Cat. No. 01CH37164). Vol. 4. 28. Parrish, J. "Robotic Servicing of IEEE, 2001. Geosynchronous Satellites (RSGS)." Defense 39. Murugesan, S. "An overview of Advanced Research Projects Agency electric motors for space applications." IEEE (DARPA). [Online]. Available: https://www. transactions on industrial electronics and darpa. mil/program/robotic-servicing-of- control instrumentation 4 (1981): 260-265. geosynchronous-satellites. 40. Li, Y. F., and X. B. Chen. "On the 29. B.E. Kelm, et al. FREND: Pushing the dynamic behavior of a force/torque sensor for Envelope of Space Robotics. robots." IEEE Transactions on and Satellite Technology. 2008 NRL Review. Instrumentation and Measurement 47.1 (1998): 304-308.

Ezinne Uzo-Okoro 18 34th Annual Small Satellite Conference 41. Tsujimura, Takeshi, and Tetsuro Yabuta. "Object detection by tactile sensing method employing force/torque information." IEEE Transactions on robotics and Automation 5.4 (1989): 444-450. 42. Liu, Guangjun, et al. "A base force/torque sensor approach to robot manipulator inertial parameter estimation." Proceedings. 1998 IEEE International Conference on Robotics and Automation (Cat. No. 98CH36146). Vol. 4. IEEE, 1998. 43. Watson, Judith J., Timothy J. Collins, and Harold G. Bush. "A history of astronaut construction of large space structures at NASA ." Proceedings, IEEE Aerospace Conference. Vol. 7. IEEE, 2002. 44. Doggett, William. "Robotic assembly of truss structures for space systems and future research plans." Proceedings, IEEE Aerospace Conference. Vol. 7. IEEE, 2002. 45. Bruner, Wesley, Carlos Enriquez, and Sreekumar Thampi. "Mechanism analysis and verification approach for ISS truss assembly." 37th Aerospace Mechanisms Symp. 2004. 46. Hastings, Daniel E., and Carole Joppin. "On-orbit upgrade and repair: The hubble example." Journal of spacecraft and 43.3 (2006): 614-625. 47. NASA, Tech Demonstration. "NASA’s Dragonfly Project Demonstrates Robotic Satellite Assembly Critical to Future Space Infrastructure Development." (2017). 48. Paus, Fabian, et al. "A combined approach for robot placement and coverage path planning for mobile manipulation." 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2017.

Ezinne Uzo-Okoro 19 34th Annual Small Satellite Conference