Lunar Robotic Construction System Using Raw Regolith: Design Conceptualization

Abstract

This paper outlines the inception, conceptualization and primary morphological selection of a robotic system that employs raw lunar regolith for constructing protective berms and shelters on the Moon’s surface. The lunar regolith is considered the most readily available material for in situ resource utilization on the Moon. The lunar environment is characterized, and the operational task is defined, informing the development of high-level system requirements and a functional analysis through the glass-box method. The key morphological areas are identified, and candidate concepts are evaluated using the Analytic Hierarchy Process (AHP). The evaluation process employs a new approach to aggregating expert data through the ZM_II method to establish priorities of the design criteria, which eliminates the need for pairwise comparisons in data collection. Each criterion is associated with a specific and quantifiable metric, which is then used to evaluate the morphologies during the AHP. The selected morphologies are determined as: a vibrating hopper for intake (normalized decision value of 27.5% out of 5 candidate solutions), a roller system for container deployment and filling (26.2% out of 7), a magnetic RCU interface (22.6% out of 7), and a 4-DoF manipulator to place the RCUs in the environment (23.6% out of 5). The final morphology is selected by combining the decision values across the primary morphological areas into a unified decision metric. This is followed by the preliminary selection of the system’s surrounding architecture. The design conceptualization is performed within a real-life operational scenario, namely, to create a blast berm for the landing pad using the lunar regolith provided by an existing excavator. The next phase of the work will include the system’s detailed design, as well as investigations on the requirements for a variety of construction tasks on the lunar surface.

1. Introduction

Space-faring nations are working towards a sustainable return to the Moon over the next decade. The International Space Exploration Coordination Group (ISECG), a consortium of 27 space agencies, has published a Global Exploration Roadmap (GER), citing human and robotic lunar surface missions as a stepping stone to exploring deep space destinations, such as Mars [1,2]. The Moon can act as a key launch point to other celestial bodies, since it conveniently mitigates the effects of the Earth’s atmosphere and gravity. The global focus on the establishment of a sustainable presence on the lunar surface has brought to light the need for new construction technologies, whose aim is no longer purely scientific, but to manipulate in situ resources reliably and at scale. Many space agencies, including the National Aeronautics and Space Administration (NASA) [3], the European Space Agency (ESA) [4,5,6], the China National Space Administration (CNSA), Russia’s Roscosmos [7], and the Indian Space Research Organization (ISRO) [8], have indicated their intentions to establish permanent structures on the lunar surface in the coming years, with special attention being paid to the exploitation of polar water ice [9]. A particularly crucial prerequisite to such infrastructure is the creation of berms, shelters, and other protective means for the equipment, machines and structures that will be initially deployed. A comprehensive survey on lunar construction is provided in [10].

Studies have begun to investigate a variety of methods and materials for the creation of such lunar surface structures, leveraging in situ resource utilization (ISRU) to minimize up-mass and exploit the abundance of lunar regolith on the Moon. Therefore, there is a need for robotic construction systems that can safely and efficiently manipulate and integrate the simplest and most readily available materials on the Moon, namely, the raw lunar regolith. One recent study quantitatively evaluates confinement formation, or regolith bagging, as the best method for building lunar surface structures [11]. Confinement formation methods boast strengths in terms of in situ material usage ratio, size, resistance to temperature variation, and formation time. Of particular importance, these methods require minimal regolith pre-processing, as compared to other candidates such as additive manufacturing and regolith cements. This would increase the reliability of the construction method, especially for initial missions, where the properties and composition of the regolith are yet to be fully determined. Although the work proposes protecting inhabited structures using these regolith bags, another work proposes the construction of superadobe structures [12]. A recent study [13] also proposes the use of regolith bags for blast walls, robotic system (or electronics) shelters, and greenhouse structures, the latter two being based on the superadobe configuration.

This paper begins with some relevant background (Section 2) on lunar regolith excavators (Section 2.1), lunar construction methods and materials (Section 2.2), and robotic systems for lunar construction (Section 2.3). Then, an overview of Regolith Containment Units (RCUs) is presented (Section 3), analogous to terrestrial “sandbags” and inspired by the aforementioned confinement formation methods. These RCUs will be filled and manipulated by a Lunar Robotic Construction System (LRCS) to create lunar surface structures, such as berms and shelters for surface equipment and machines, as illustrated in Figure 1. The conceptualization of the LRCS is next presented (Section 4), outlining the environmental considerations (Section 4.1), the operational scenario (Section 4.2), a functional analysis using the glass-box method (Section 4.3), and a set of high-level system requirements (Section 4.4). Once this framework is established, an assessment process is undertaken for the four primary LRCS functional modules using the Analytic Hierarchy Process (AHP, Section 5). This includes an expert survey of design assessment criteria priorities (Section 5.1), using the ZM_II ranking aggregation method to integrate the results into the AHP in a novel way (Section 5.2). The assessment is conducted (Section 6) for each of the four primary LRCS functional modules (Section 6.1–Section 6.4), using specific metrics to assess the candidate solutions with respect to the design criteria. Once the final morphology, i.e., physical configuration, is evaluated (Section 6.5), the surrounding architecture for the LRCS is also preliminarily selected (Section 7). Some concluding remarks are made at the end (Section 8).

This paper performs a systematic design conceptualization of a lunar robotic construction system, which reconciles the AHP with simple expert rankings and the evaluation of candidate solutions through specific and quantifiable metrics. The outcome is a novel system that utilizes raw regolith to create RCUs and then manipulates them into the lunar environment to construct surface structures. The paper, thus, contributes to the advancement of both lunar robotic construction technologies as well as engineering design research.

2. Background

A number of robotic systems have been proposed to perform a variety of construction tasks on the lunar surface (see [10] for a survey.) Yet, none of them address tasks such as filling containers with raw regolith and manipulating them, although such works can be used as inspiration for the LRCS conceptualization. Generally, the design problem can be divided into regolith acquisition (excavation), construction method (and materials), and RCU manipulation.

2.1. Lunar Regolith Excavators

The regolith excavation task has been under development since the early 2000s, with the most promising concepts reaching flight-readiness. Lunar regolith excavation systems are broadly classified as continuous or discrete, based on whether or not their cutting surfaces remain in constant contact with the soil [14]. Discrete systems, such as front-end loaders and dozers, interrupt contact to clear or dump material. Continuous systems, such as bucket wheels and drums, maintain constant engagement and are often more efficient in low-gravity conditions. These systems are further distinguished as complete (with integrated mobility) or partial (stationary) platforms. NASA’s Kennedy Space Center (KSC) has led recent developments in continuous-complete systems with the Regolith Advanced Surface Systems Operations Robot (RASSOR) series [15]. These dual bucket-drum excavators counteract lunar gravity by using counter-rotating drums, and they have progressed from the teleoperated RASSOR [16] to RASSOR 2.0 [17] that features semi-autonomous control and improved mass efficiency, excavating up to 2.7 metric tons of regolith per day with just 4 W/kg of power consumption. The latest iteration, ISRU Pilot Excavator (IPEx) [18], shown in Figure 2, reduces system size by 30% and continues development toward TRL 5 with enhanced bucket drum [19] and wheel designs [20].

Some other excavation concepts have also been investigated. The Lunar Excavation and Size Separation System (LES³), integrated with the 40–60 kg LUVMI-X rover [21] (co-funded by the ESA (Paris, France) and private investors in Luxembourg), is a compact regolith excavation prototype weighing under 2 kg. It excavates 100 g per cycle using a low-force cylindrical scoop, optimized for shallow surface layers to avoid compaction effects. Regolith is classified into two size fractions via dry sieving and vibration, leveraging vibration to also reduce excavation forces. A larger-scale concept combines a bucket chain excavator with ballistic conveyance, launching sealed regolith containers to remote sites [22]. While Discrete Element Modeling (DEM) of the concept supports feasibility in lunar conditions, the absence of supporting infrastructure raises deployment concerns. A bucket-ladder system has also been developed, which aims to reach TRL 5–6, building on prior results showing >1000 kg/h excavation rates at 0.01 W/kg/h in Earth conditions [23]. However, the mechanical complexity of bucket-ladder systems and the lack of lunar-environment testing remain as major challenges to practical implementation. A simpler system, the CASPER rover, employs an Archimedes screw-based propulsion and excavation system [24], achieving a 30 kg/h excavation rate with only 3.4 kg mass and 30 W power consumption. Testing has been limited to silica sand, which differs significantly from lunar regolith, highlighting the need for higher-fidelity validations. Similar works include auger and flexible screw systems for regolith collection, conveyance, and drilling tasks [25,26], sharing core principles with CASPER but varying in mechanical design and intended function.

2.2. Lunar Construction Methods and Materials

The methods and materials for lunar construction have been widely investigated, including regolith binding, additive manufacturing, and raw material methods. The use of conventional regolith-based concrete on the Moon is challenged primarily by its water requirement, which demands either the costly transport of water from Earth or the development of in situ resource utilization (ISRU) capabilities. Additionally, lunar environmental conditions, notably extreme temperature fluctuations and vacuum, complicate traditional water-based curing processes [27]. To address these concerns, waterless concrete alternatives have been considered, resulting in the inception of sulfur concrete [28], as well as polymer concrete [29] and geopolymer concrete [30]. In general, the Moon’s reduced gravity (1/6 g) could scale down the structural demands on such regolith binders proportionately. For instance, terrestrial standards requiring 25–30 MPa compressive strength could correspond to 4–5 MPa for lunar structures [31,32]. However, the sensitivity to environmental factors, especially during curing, remains a major barrier to the usage of regolith binders.

Additive manufacturing (AM) has gained traction as a construction approach thanks to its capacity to use minimally processed regolith. Some AM techniques employ binders, such as D-shape printing [33]. However, some of the most appealing AM methods eliminate the need for binders entirely, minimizing the up-mass. Various sintering methods can be employed, including direct [34], laser [35,36,37,38], microwave [39], and solar sintering [40,41]. The latter of these, solar sintering, is particularly attractive due to the potential abundance of solar power in highly illuminated regions near the lunar poles. This power availability helps mitigate the extremely high energy demands of sintering methods, limiting scalability for large structures on the Moon [42].

An alternative approach is regolith confinement, or bagging, where regolith is enclosed in durable fabric containers to form structural units [11]. These RCUs provide a low-cost and low-processing construction alternative for simple structures, leveraging the bulk properties of the lunar regolith for the mitigation of thermal variations, radiation, and projectiles. These RCUs are the focus of this study (see Section 3) and can be used to form some key parts of the initial lunar surface infrastructure.

2.3. Robotic Systems for Lunar Construction

To implement a variety of construction methods on the Moon’s surface safely and reliably, lunar robotic systems are required, which can be broken down into a few categories, namely, cranes, mobile manipulators, and robot teams. Lunar cranes have evolved from early concepts, such as LEVPU [43] and SPIDER [44], to NASA’s Lunar Surface Manipulation System (LSMS) [45,46,47]. The LSMS variants support payloads from 35 kg to over 1000 kg, with key developments such as autonomous control, modularity, and integration with mobility platforms like Chariot [48,49,50,51]. The LSMS L35, selected for the LANDO project, will demonstrate autonomous payload offloading using Astrobotic’s Peregrine lander. The ESA’s Moon Village project also explores heavy-lift systems, including a gantry crane concept built on the ATHLETE platform [52,53]. This modular system supports up to 15,000 kg payloads with a 6600 kg mass, and includes robotic arms with interchangeable end-effectors for gripping, excavation, and construction. NASA’s COLDArm [54], a 4-Degree-of-Freedom (DoF) manipulator created by the Jet Propulsion Laboratory (La Cañada Flintridge, CA, USA), is designed for cryogenic temperatures, offers 65 kg tip load capacity without heaters, optimizing it for volatile-sensitive operations in extreme environments.

Mobile manipulators are another useful type of lunar construction morphology. Examples include Helelani from the Pacific International Space Center for Exploration Systems (PISCES), used for Vertical Takeoff Vertical Landing (VTVL) pad preparation [55]; China’s Super Mason (CSM), employing an ABB 6-DoF arm for structural assembly [56]; and the Lunar Exploration Rover System (LERS), a 6-DoF arm on a mobile base aimed at infrastructure deployment [57]. The ATHLETE platform remains a cornerstone of lunar mobility and manipulation, offering limb-level autonomy and versatility for tasks ranging from payload transport to drilling and additive manufacturing [58]. Individual limbs handle payloads from 500 kg to over 11,000 kg. The development of lunar mobile manipulators has also been pursued in the private sector. The four-wheeled rover from GITAI demonstrates two manipulators: the 9-DoF S1 [59] and the modular Inchworm, capable of interfacing with landers, habitats, and rovers [60]. The system has been field-tested both on the ISS and in lunar analog environments, with particular emphasis on lunar dust mitigation [61].

Robot teams for lunar construction, such as Germany’s MultiRob 3D [62], are also emerging, exploring coordinated excavation, transport, and additive manufacturing using lunar regolith simulants [63]. Meanwhile, ESA’s PRO-ACT project [64] leads in multi-agent autonomy, featuring robots such as IBIS, MANTIS, and the new Veles, capable of complex terrain traversal, tool exchange, and collaborative construction [65,66]. A companion gantry system supports heavy-lift tasks within a 1.7 m³ workspace. Several efforts are also exploring collaborative manipulation, such as tower-climbing robots for infrastructure assembly [67] and dual-arm systems for autonomous solar panel deployment [68].

Although none of these robotic systems can fully address the necessary LRCS functionalities, they can be used as inspiration for parts of the system, such as the manipulation of the RCUs.

3. Regolith Containment Unit (RCU)

Most lunar habitation concepts require the use of in situ regolith to provide shielding against radiation, micrometeoroids, and extreme temperatures [69]. This necessitates not only efficient excavation and transport systems, but also effective strategies for shaping and containing regolith as a structural material. Among proposed solutions, Regolith Containment Units (RCUs), analogous to terrestrial sandbags, have garnered interest due to their high in situ material utilization, fast deployment potential, and lower thermal sensitivity, though they typically exhibit slightly reduced compressive strength relative to solid construction methods [11].

Some NASA-led studies evaluate various container materials and filling mechanisms for the RCUs. One concept proposes the use of fiberglass composite weave containers of 0.0283 m³ volume, filled via a conveyor and funnel system, theoretically capable of producing 120 filled RCUs per hour over an 8 h shift [70]. A similar study proposes Kevlar 149 containers of 0.487 m³, with a scoop-based system achieving 10 filled RCUs per hour under comparable operational assumptions [71]. More recent work explores smart RCUs equipped with microprocessors, RF transponders, RFID tags, and piezoelectric power sources to enable real-time monitoring of placement and structural integrity [12]. Such RCUs are positioned as a low-entry-barrier solution for lunar infrastructure, requiring minimal processing of raw regolith compared to technologies such as additive manufacturing. Some proposed structural concepts include berms, superadobe walls, and shielding enclosures for sensitive systems like greenhouses or habitats [13].

A notable NASA initiative aims to construct a garage-like lunar structure using RCUs filled with regolith [72]. The project evaluates numerous container fabrics, including Vectran, Zylon, Twaron, Nextel, Nomex, and Gore PTFE, with Vectran emerging as the top candidate. Multiple filling approaches are considered, including shovels, funnels, tubes, and air-based systems, but the most effective is found to be a hopper with a helical flexible screw conveyor. The final structures are physically assembled at Marshall Space Flight Center (MSFC). Follow-up work analyzes structural stability, finding that top-connected RCUs yield more stable configurations than center-connected RCUs (see Figure 3) [73]. The structural performance of RCUs is also examined. One study tests aramid fabric-based RCUs in configurations such as columns, arches, and protective cladding, finding them viable for integration with prefabricated or inflatable habitats. The units achieve a 6.9 MPa compressive strength and retained 90% strength after vacuum aging, demonstrating suitability for load-bearing applications [74].

Other studies explore automated filling methods using existing or proposed lunar excavation technologies. One concept considers the Honeybee Robotics’ pneumatic excavator, outlining operational configurations such as site-level filling, centralized silos, and final-location filling [75]. However, these systems do not go beyond the conceptual or proof-of-concept stage. A separate work proposes a wall-scaling RCU bagger system for outpost burial, in which regolith is conveyed, the bag formed, and the filled unit deposited directly atop a stacking structure [76]. While this approach is tested at one-tenth scale using sand and remains the only analogous system to the LRCS, it lacks the dexterity needed for more complex structural assemblies.

4. System Conceptualization

The conceptualization of the LRCS can be broken down into three stages (Figure 4), namely, Task Definition, System Conceptualization, and Design Selection. The Task Definition stage includes characterization of the lunar environment (Section 4.1) as well as a breakdown of the expected system operation (Section 4.2), including any operational trade-offs. The information is then used in the System Conceptualization stage to conduct a functional analysis (Section 4.3), using the glass-box method [77], and define a set of system-level requirements iteratively. The Design Selection stage then uses the functional analysis and system requirements to make a systematic assessment of various design alternatives, using the Analytic Hierarchy Process (AHP) [78], to select the best candidates according to a set of criteria. The criteria are prioritized through a survey from the field experts, using the ZM_II method for ranking aggregation [79,80,81]. In this paper, four key functional modules are highlighted in the assessment process. Once the morphologies for these functional modules are determined, the morphologies for the remaining (standard) functional modules will be selected, e.g., power generation.

4.1. Environmental Characterization

The Moon is subject to a gravitational acceleration of only 1.63 m/s² at this surface, virtually no atmosphere, extreme temperature variations, radiation, and micrometeorite impacts [82]. The temperature at the Apollo landing sites varies by 282 °C, from $-171$ °C to 111 °C [83], while it can drop below $-213$ °C in Permanently Shadowed Regions (PSRs) near the poles [84]. This variation is drastically reduced with depth. At the Apollo landing sites, where the equilibrium temperature of lunar regolith is in the range of $-15$ °C to $-20$ °C, the variation is nearly 10 °C at 20 cm depth, and 1 °C at 40 cm depth. This would make the lunar regolith a good thermal insulator [85]. In the absence of a magnetic field, the Moon is constantly exposed to galactic cosmic rays (GCRs, ∼0.3 Sv/year) and solar particle events (SPEs, up to 0.6 Sv/h), far exceeding safe human exposure limits, and thus requiring shielding for long-duration surface systems. Regolith-based shielding of 1 m is commonly proposed to mitigate radiation effects under nominal conditions [86]. Seismic activities on the Moon are limited. The Apollo missions collected a total of approximately 12,000 “moonquake” readings, with most measuring 1–2 and a maximum of 3 on the Richter scale [87]. Micrometeorites have an average mass of ${10}^{-10}$ to ${10}^{-8}$ kg, and an average speed of 72 km/s [83]. In general, structures can be protected from micrometeorites by lunar regolith and structural elements like Whipple shields [88].

The lunar regolith is a well-graded (consistent particle and petrological distribution with depth) but locally poorly sorted layer of the lunar surface, with an average rock-free depth of 10 m in the maria, and 12 m in the highlands [89]. Formed by billions of years of meteoritic bombardment, the regolith is a fine, adhesive, sticky, grainy, and highly abrasive material, composed of lithic, breccia, mineral, and glass fragments, as well as agglutinates [90]. Lunar regolith particles are extremely fine, with an average size of 70 μm and an average elongation of 1.35, making them ovular. The typical lunar regolith particle has nearly 8 times the surface area of an equivalent sphere, contributing to its high interlocking behaviour [91]. The lunar regolith consists of a loosely packed upper layer, which rapidly compacts with depth past 15 cm. The bulk density stands at 1.3 g/cm³ at the surface, and asymptotically approaches 1.92 g/cm³ at lower depth. The shear strength of lunar soil increases rapidly after 15 cm, and again after 30 cm, which can make excavation particularly challenging [92]. Electrostatic charging under UV radiation and Van de Waals forces further complicate operations, as regolith particles can cling to tools and equipment [82]. A summary of common geotechnical properties for the lunar regolith can be found in

Table 1. Common lunar regolith geotechnical properties.

Property	Range	Units	Source
Bulk Density ($\rho$)	1.3–1.92	g cm−3	[82]
Porosity (n)	40–65	%	[92]
Specific Gravity (G)	$3.1$	-	[82]
Friction Angle ($\varphi$)	30–50	°	[82]
Cohesion (c)	0.1–3.4	kPa	[92]

, and a few recent reviews of lunar regolith and its properties can be found in [90,91,92].

4.2. Operation

The LRCS will need to perform a series of tasks to accomplish the overall goal, which can be broken down as shown in Figure 5. Of particular importance, the LRCS will need to be capable of interfacing with NASA’s ISRU Pilot Excavator (IPEx) to intake lunar regolith. Without loss of generality and to be specific with the conceptualization process, this paper focuses on the operation for constructing a blast berm for landing pads. A major challenge to the goal of repeatable lunar landings is the interaction of rocket engine exhaust plumes with the lunar regolith, which can create blast ejecta and cratering. The ejecta can reach velocities over 2000 m/s and travel large distances in a vacuum, potentially entering orbit and cis-lunar space. It is estimated that the Apollo lunar lander displaced 2600 kg of regolith during landing, leading to erosion and cratering [93]. The ejecta creates high velocity particles that can damage surface structures and remain suspended in orbit, and the cratering results in uneven terrain, both of which could hinder future lunar landings [94,95]. Simulations show that larger particles of the ejecta would be traveling with supersonic speeds at less than 3° from the touchdown, and smaller particles would be ejected with slower velocities at close to 17° [96,97]. While current rocket and localization technologies may require landing pads of up to 100 m diameter, the landing area is projected to be reduced to 25 m as landing accuracy improves [96,98,99]. As such, protecting surface structures against the most harmful supersonic particles ejected at 3° from a 100m diameter landing pad, in the worst case, will be considered as the operational task, as shown in Figure 6. The RCUs can form a blast berm structure, which needs to be $50\times tan(3°)\approx 2.6$ m tall. Considering RCUs of size 0.43 m × 0.25 m × 0.10 m (see Section 4.4.2) stacked in a pyramidal arrangement,

$$\underset{\# \mathrm{stacks}}{\underbrace{⌊{\displaystyle \frac{\pi \times 100\mathrm{m}}{0.43\mathrm{m}}}⌋}}\times \underset{\mathrm{for}\mathrm{one}\mathrm{pyramid}\mathrm{stack}}{\underbrace{{\displaystyle \frac{2.6\mathrm{m}}{0.1\mathrm{m}}}\times {\displaystyle \frac{1+\frac{2.6}{0.1}}{2}}}}\approx 256,230$$

(1)

bags are required to build the berm.

Another important consideration is the system’s modes of operation, which can be broadly broken down as follows:

Operational Trade-Offs

One key operational consideration is the required frequency or speed of operation. This is highly dependent on the rate at which lunar regolith will be brought to the system by IPEx. Some properties of IPEx are summarized in

Table 2. IPEx properties (Courtesy: NASA KSC).

Excavation System	Counter-Rotating Bucket Drums
Maximum Traversable Slope	30°
Maximum Obstacle Height	7 cm
Excavator Dry Mass	35 kg
Regolith Capacity	30 kg
Load-dump Cycle Time	13.5 min
Length × Width × Height *	76 cm × 62.5 cm × 68.2 cm

* Stowed configuration—bucket drums up (see Figure 2).

. Of particular importance is the load-dump cycle time, i.e., the time taken to excavate a full load of regolith and dump it at the LRCS location, which indicates that 30 kg of lunar regolith can be brought to the LRCS every 13.5 min. Assuming that a single RCU can be filled using one IPEx load, and that just one IPEx system is deployed, the LRCS needs to intake the lunar regolith and fill one RCU in under 13.5 min.

It is also important to consider the possibility of decoupling the filling and manipulation of the RCUs. In other words, the LRCS could be one stand-alone system (mobile or otherwise), or it could be configured as two separate subsystems, one completing the RCU filling task and the other being a mobile manipulator that places the RCU in the environment. A key factor for making such a decision is the environment where the LRCS is likely to be deployed. The vast majority of proposed habitat locations on the lunar surface involve the lunar North and South poles. In 2022, NASA identified 13 candidate landing regions within 6 degrees of latitude of the lunar South Pole for the Artemis missions [100]. The missions will aim to establish a sustainable human presence on the Moon with the Artemis Base Camp [3]. NASA cites the presence of in situ resources in PSRs, particularly water ice, as well as large periods of continuous sunlight, as some of the primary considerations for these candidates [100]. The broader international community also considers establishing a presence primarily at the lunar South Pole, due to largely overlapping factors such as illumination and power profiles, presence of volatiles in PSRs (e.g., water ice), and terrain [5,101,102,103,104,105].

The illumination conditions at the lunar poles are highly impacted by the local terrain, owing to the low inclination angle of the Moon’s rotational axis at 1.54° [106]. This inclination creates two phenomena of interest at the lunar poles. The first phenomenon is Peaks of Eternal Light or Peaks of Light (POL), where the terrain elevation is higher than the horizon, allowing for permanent sunlight and solar power generation. The second phenomenon involves PSRs, where there are permanent shadows in the craters, which can trap essential resources and volatiles such as water ice. An ideal location would include a region of near-permanent sunlight for power generation, next to a PSR for resource extraction, and relatively flat terrain for habitat construction and landing [107,108]. Some works have also considered the implementation of solar towers, ranging from 2 m in height up to 2000 m, to induce permanent sunlight at the solar array locations, achieving constant or near-constant illumination in some cases [109,110]. Small-scale solar towers are already being designed, such as the Lunar Array, Mast, and Power System (LAMPS), which can extend up to 15 m height [111]. Considering these factors, as well as the size of the anticipated blast berm, a decoupled system would be ideal to leverage the possibility of near-constant solar power at the South Pole, while still being able to manipulate the RCUs into the environment. This means that the lunar regolith intake and the RCU filling, along with the core functionalities of the LRCS, e.g., power generation (solar cells), communications, etc., would be undertaken by a collective assembly with limited or no mobility. Meanwhile, the RCU manipulation task would be undertaken by a mobile manipulator, with its power obtained from either an Energy Storage System (ESS) or a Radioisotope Thermoelectric Generator (RTG). Another advantage of the decoupled system is that the most massive elements of the system would not need to be mobile, which helps to reduce the overall power consumption and the risk of damage to these components. Instead, only the components necessary to transport and manipulate the RCUs would require a high degree of mobility.

4.3. Functional Analysis

The key functions that a system is required to perform to accomplish the design objectives can be identified through the functional analysis, which consists of two primary processes, namely, functional decomposition and functional allocation [112]. Functional decomposition involves hierarchically breaking down the top-level function of the system (or goal of the design) into unique sub-functions. Functional allocation will then involve the mapping of sub-functions to subsystems (or modules). The two processes can continue iteratively, as depicted in Figure 7 (for two levels). Based on the principles of taxonomy design [113], the ending condition for each level would be to not have any further addition, merging or splitting of functions and subsystems (or modules). Further, subsystems and those functions within each subsystem should be uniquely identified.

In this paper, both functional decomposition and allocation processes are performed through the glass-box method [77], which treats the system initially as a “black box” with certain defined inputs and outputs, and then incrementally reveals the box by following the flow of inputs throughout the system.

Figure 8 shows the first level of functional analysis for the LRCS. The primary subsystems include RCU Filling, RCU Transfer, RCU Manipulation, Thermal Management, Power Management, Communications, Onboard Computing, and Mobility. In addition, a subsystem is included for a third-party regolith analysis and processing tool to support potential science objectives, such as those in the Artemis Program plan [3]. The beneficiation of lunar regolith is a necessary function during regolith intake, involving the filtering of large rocks and particles that would constitute undesirable feedstock for the filling of the RCUs. The filtered materials will be ejected from the system as the primary form of waste, along with the heat. The Mobility subsystem will ensure a large workspace, as well as proper deployment of the system from its launch vehicle. Since the LRCS is a decoupled system, the subsystems are divided primarily between two machines, the RCU Bagger and the Mobile Manipulator, both sharing some common subsystems. The level 2 functional analysis of the filling and manipulation subsystems is shown in Figure 9 and Figure 10, respectively.

A number of functional modules from Figure 8 are selected as the primary modules for design selection: intake lunar regolith, deploy empty RCU, interface with and secure RCU, and transport RCU to desired pose. The other functional modules will be selected to support the final solution (morphology) for these primary modules.

4.4. System Requirements

Table 3 displays some of the key requirements for the LRCS. The functional requirements are directly derived from the functional analysis presented in Figure 8. The constraints on the stowed volume are based on the properties of the Apollo Lunar Roving Vehicle (LRV) as an upper bound [114]. The LRV remains one of the largest robotic systems sent to the Moon, and is a good estimate for what we know we should be able to deliver to the lunar surface. An estimate is necessary, as most landers currently being developed for the lunar surface do not yet specify volumetric constraints.

4.4.1. Environmental Requirements

The average daily radiation experienced on the lunar surface has been measured at 1369 μSv/day, including both solar wind and galactic cosmic rays [115]. Assuming an expected lifetime of one year, this comes to $1369\mathrm{\mu }\mathrm{Sv}/\mathrm{day}\times 365\mathrm{days}=499,685\mathrm{\mu }\mathrm{Sv}$ or 0.5 Sv. The SLS-SPEC-159 Cross-Program Design Specification for Natural Environments by MSFC [116] specifies the expected temperature ranges at an 85-degree lunar latitude. These values are rounded up for the maximum and down for the minimum to find the operable temperature range of 60 K to 190 K (based on mean temperature) and survivable temperature range of 40 K to 230 K (based on 1 Sigma temperature). The LRCS will also need to withstand an ambient pressure of $3\times {10}^{-15}$ atm (vacuum conditions) [83], a maximum solar irradiance of 1425.7 W/m² [117], and exposure to lunar regolith dust [118].

Table 3. System requirements.

ID	Requirement: The System Shall…
Functional Requirements
SYS-F-01	intake lunar regolith from an external lunar regolith excavation and processing system.
SYS-F-02	beneficiate lunar regolith.
SYS-F-03	transfer excess regolith to a third-party Regolith and Analysis Processing Subsystem.
SYS-F-04	fill the RCUs with lunar regolith.
SYS-F-05	seal the RCUs.
SYS-F-06	internally transfer the RCUs from the filling configuration to the manipulation configuration.
SYS-F-07	be decoupled into a mobile manipulation system and a centralized RCU filling system.
SYS-F-08	interface with and secure the RCUs for manipulation.
SYS-F-09	manipulate the RCUs into a desired pose within the environment.
SYS-F-10	be able to move its mobile manipulator base and localize itself in the lunar environment.
SYS-F-11	have onboard computing capabilities.
SYS-F-12	collect and store system health and history data.
SYS-F-13	be able to communicate with lunar surface and orbital communication systems.
SYS-F-14	generate, store, and manage sufficient power for operation during the operational lifetime.
SYS-F-15	be able to maintain its thermal state within operable ranges.
SYS-F-16	be capable of operating in the lunar environment.
SYS-F-17	be deployable from a landing vehicle to the lunar surface.
Performance Requirements
SYS-P-01	support a sustained operational (average) power draw of 390 W.
SYS-P-02	support a peak operational power draw of 730 W.
SYS-P-03	be capable of filling 1 RCU in under 13.5 min.
SYS-P-04	be capable of manipulating 1 RCU in under 13.5 min.
SYS-P-05	have an operational lifetime of 6 Earth months (approximate lunar South Pole daytime).
SYS-P-06	have a total battery charge of at least 86 A h [119].
SYS-P-07	be capable of reaching a manipulation workspace volume of 2 m × 2.6 m × 2 m (depth, height, width).
SYS-P-08	be capable of reaching a mobility workspace area of $\pi \times {\left(50\mathrm{m}\right)}^2=2500{\mathrm{m}}^2$.
SYS-P-09	be capable of intaking at least 60 kg of lunar regolith (double IPEx capacity).
Environmental Requirements
SYS-E-01	have an operational temperature range of 60 K to 190 K [116].
SYS-E-02	have a survivable temperature range of 40 K to 230 K [116].
SYS-E-03	maintain the temperature of electronics between 233 K and 358 K.
SYS-E-04	be capable of withstanding a total radiation dose of at least 0.5 Sv.
SYS-E-05	operate while exposed to a maximum solar irradiance of 1425.7 W/m2 [117].
SYS-E-06	operate while exposed to lunar regolith and lunar regolith dust [118].
SYS-E-07	operate in an ambient pressure of $3\times {10}^{-15}$ atm [83].
Constraints
SYS-C-01	have a maximum mass of 810 kg.
SYS-C-02	have a maximum stowed volume of 6 m3.
SYS-C-03	use non-rigid RCUs with dimensions of 0.43 m × 0.25 m × 0.10 m.
SYS-C-04	use RCUs with a mass of 22 kg.
SYS-C-05	have a regolith intake interface of dimension 0.625 m × 0.312 m to accommodate IPEx.

Table 3. System requirements.

ID	Requirement: The System Shall…
Functional Requirements
SYS-F-01	intake lunar regolith from an external lunar regolith excavation and processing system.
SYS-F-02	beneficiate lunar regolith.
SYS-F-03	transfer excess regolith to a third-party Regolith and Analysis Processing Subsystem.
SYS-F-04	fill the RCUs with lunar regolith.
SYS-F-05	seal the RCUs.
SYS-F-06	internally transfer the RCUs from the filling configuration to the manipulation configuration.
SYS-F-07	be decoupled into a mobile manipulation system and a centralized RCU filling system.
SYS-F-08	interface with and secure the RCUs for manipulation.
SYS-F-09	manipulate the RCUs into a desired pose within the environment.
SYS-F-10	be able to move its mobile manipulator base and localize itself in the lunar environment.
SYS-F-11	have onboard computing capabilities.
SYS-F-12	collect and store system health and history data.
SYS-F-13	be able to communicate with lunar surface and orbital communication systems.
SYS-F-14	generate, store, and manage sufficient power for operation during the operational lifetime.
SYS-F-15	be able to maintain its thermal state within operable ranges.
SYS-F-16	be capable of operating in the lunar environment.
SYS-F-17	be deployable from a landing vehicle to the lunar surface.
Performance Requirements
SYS-P-01	support a sustained operational (average) power draw of 390 W.
SYS-P-02	support a peak operational power draw of 730 W.
SYS-P-03	be capable of filling 1 RCU in under 13.5 min.
SYS-P-04	be capable of manipulating 1 RCU in under 13.5 min.
SYS-P-05	have an operational lifetime of 6 Earth months (approximate lunar South Pole daytime).
SYS-P-06	have a total battery charge of at least 86 A h [119].
SYS-P-07	be capable of reaching a manipulation workspace volume of 2 m × 2.6 m × 2 m (depth, height, width).
SYS-P-08	be capable of reaching a mobility workspace area of $\pi \times {\left(50\mathrm{m}\right)}^2=2500{\mathrm{m}}^2$.
SYS-P-09	be capable of intaking at least 60 kg of lunar regolith (double IPEx capacity).
Environmental Requirements
SYS-E-01	have an operational temperature range of 60 K to 190 K [116].
SYS-E-02	have a survivable temperature range of 40 K to 230 K [116].
SYS-E-03	maintain the temperature of electronics between 233 K and 358 K.
SYS-E-04	be capable of withstanding a total radiation dose of at least 0.5 Sv.
SYS-E-05	operate while exposed to a maximum solar irradiance of 1425.7 W/m2 [117].
SYS-E-06	operate while exposed to lunar regolith and lunar regolith dust [118].
SYS-E-07	operate in an ambient pressure of $3\times {10}^{-15}$ atm [83].
Constraints
SYS-C-01	have a maximum mass of 810 kg.
SYS-C-02	have a maximum stowed volume of 6 m3.
SYS-C-03	use non-rigid RCUs with dimensions of 0.43 m × 0.25 m × 0.10 m.
SYS-C-04	use RCUs with a mass of 22 kg.
SYS-C-05	have a regolith intake interface of dimension 0.625 m × 0.312 m to accommodate IPEx.

4.4.2. RCU Specification

According to the U.S. Army Corps of Engineers [120], the most commonly used sandbags for terrestrial missions measure 35.56 cm × 60.96 cm on a flat surface before filling. The corresponding sandbag unit filled to two-thirds full capacity (for desired engineering properties) measures 10.16 cm × 25.4 cm × 43.18 cm, or a volume V = 11,143.20 cm³. Considering the highest possible density for lunar regolith (worst case) of $\rho$ = 1.92 g/cm³ [92], the estimated mass M for the RCU can be obtained as follows:

$$M=\rho \times V\approx 22\mathrm{kg}$$

(2)

Considering that IPEx holds a maximum load of 30 kg (see Table 2), this represents approximately a 26.7% margin between a single anticipated IPEx load and the required regolith to form one RCU.

4.4.3. Power Budget

Based on the goals of the Artemis Program [3], the most likely deployment location for the LRCS would be the Moon’s South Pole, where many regions experience near-constant illumination [121]. This means that surface systems could leverage solar power for near-constant power generation. The power generation equation for a solar array can be calculated as follows:

$$P=(A·cos(\theta \left)\right)·\varphi ·\eta$$

(3)

where A is the area of the solar array, $\theta$ is the incidence angle of the Sun, $\varphi$ is the solar irradiance, and $\eta$ is the efficiency of the solar panel. At the lunar South Pole, the elevation angle of the Sun can be up to approximately 1.5 degrees, so we consider $\theta =1.5°/$$2=0.75°\approx 1°$ [122]. The average solar panel efficiency for silicon solar cells is 15% [110]. More efficient solar cells with efficiencies up to 32% can be supplied by Spectrolab [123] and Azur Space [124]. However, these solar panels are relatively small (both less than 85 cm²). The solar irradiance at the surface of the Moon varies from $\varphi =1308.9$ W/m² to $\varphi =1425.7$ W/m² [117]. Considering this variation in $\varphi$ and an estimated solar panel surface area between 3 to 8 m², the estimate of power generation would follow the pattern shown in Figure 11.

The range of power consumption for the modes of operation (see Section 4.2) is estimated using the data available for several reference designs, including Carnegie Mellon’s lunar rover [125] and Rover for Operations support, Cargo, and Investigations (ROCI) [126], as well as the Martian rover MarsFast [127]. The power for the Manipulation subsystem is estimated by comparison to the maximal power consumption during nominal operation of an industrial robotic arm [128], giving an estimate of 150 W. The power for the RCU Filling and Manipulation subsystems is estimated by assuming the usage of one motor each with maximum and minimum power demands similar to those of the ROCI motors [126]. NASA Ames defines a set of power margins that are adopted in this paper [129]. The estimated minimum and maximum power budget for the subsystems and their margins are presented in

Table 4. Minimum/maximum power budget estimate.

Subsystem	Source (Min/Max)	RCU Filling	RCU Transfer	RCU Manipulation	Idle	Traversal	Power Margin (%)
Structural	N/A	0	0	0	0	0	30%
Thermal	MarsFast/MarsFast	6/15	6/15	6/15	6/15	6/15	30%
PowerManagement	MarsFast/Carnegie Mellon	4/35	4/35	4/35	4/35	4/35	30%
Communications	ROCI/Carnegie Mellon	8.5/90	8.5/90	8.5/90	8.5/90	8.5/90	30%
OnboardComputing	MarsFast/Carnegie Mellon	14/60	14/60	14/60	14/60	14/60	30%
Filling	ROCI/ROCI	35.4/90	0/0	0/0	0/0	0/0	30%
Transfer	ROCI/ROCI	0/0	35.4/90	0/0	0/0	0/0	30%
Manipulation	Estimate	0/0	0/0	150/300	0/0	0/0	30%
Mobility	MarsFast/ROCI	0/0	0/0	0/0	0/0	41/361.6	30%
Total		67.9/290	67.9/290	182.5/500	32.5/200	73.5/561.6
Total w Margin		88.27/377	88.27/377	237.25/650	42.25/260	95.55/730.08

.

4.4.4. Mass Budget

The maximum mass budget for the LRCS subsystems is estimated through reference designs. The lunar systems considered are MoonNext [130], a Carnegie Mellon lunar rover [125], and ROCI [126], as well as the Martian rovers BURRO [131] and MarsFast [127]. Also, the American Institute of Aeronautics and Astronautics (AIAA) presents a set of mass margins and mass growth allowances (MGA) in [132]. The estimated mass budget is presented in

Table 5. Mass budget estimate.

Subsystem	Source	Maturity	MGA (%)	Mass (kg)	Mass + MGA (kg)
Structural	BURRO	E1	18%	155	182.90
Thermal	Carnegie Mellon	E1	30%	10.3	13.39
Power Management	Carnegie Mellon	E1	20%	118.4	142.08
Communications	Carnegie Mellon	E1	20%	18	21.60
Onboard Computing	Carnegie Mellon	E1	25%	6.5	8.13
Filling	Estimate	E1	20%	5	6.00
Transfer	Estimate	E1	20%	10	12.00
Manipulation	BURRO	E1	20%	150	180.00
Mobility	BURRO	E1	20%	180	216.00
RCU	–	E1	20%	22	26.4
Total				675.20	808.50

, indicating that the upper bound for the mass of our system should be approximately 810 kg (including MGA). The maturity level E1 refers to an Estimated state of the design.

5. Design Assessment: Methodology

The potential morphologies for major functional modules are assessed using the Analytic Hierarchy Process (AHP) method [78]. The method is a value-based decision-making process, where the value of a candidate solution with respect to each criterion and the weight of the criteria are obtained based on pairwise comparisons. Specifically, it involves quantitative representation and evaluation of the candidates for each module based on their relative preference (RP) with respect to each design criterion, as well as the relative importance (RI) of the criteria. The first step is to quantify the design criteria that are to be used for the evaluation, which are identified by the proposed design requirements. The pairwise RP evaluation of morphologies is then performed with respect to each criterion, as well as the pairwise RI evaluation of the criteria. All comparisons are conducted on an ordinal 1-to-9 scale (see

Table 6. AHP ordinal scale.

Value	Description of RP/RI
1	Equally preferred/important
3	Moderately preferred/important
5	Strongly preferred/important
7	Very strongly preferred/important
9	Extremely preferred/important
2, 4, 6, 8	Intermediate values

).

The pairwise RI of P criteria is entered into a $P\times P$ matrix S, and each element $s_{ij}$ is then column-normalized

$$m_{ij}={\displaystyle \frac{s_{ij}}{{\sum }_{i=1}^P{s}_{ij}}},$$

(4)

and each row is averaged:

$$t_i={\displaystyle \frac{1}{P}}\sum _{j=1}^P{m}_{ij}.$$

(5)

The value $t_i$ ($i=1,2,\dots ,P$) is the RI value for criterion i. Similarly, for each criterion, the pairwise RP of N morphologies is entered into an $N\times N$ matrix R, and each element $r_{kl}$ is then column-normalized

$$n_{kl}^i={\displaystyle \frac{r_{kl}^i}{{\sum }_{k=1}^N{r}_{kl}}},$$

(6)

and each row is averaged:

$$f_k^i={\displaystyle \frac{1}{N}}\sum _{l=1}^N{n}_{kl}.$$

(7)

The value $f_k^i$ ($i=1,2,\dots ,P$; $k=1,2,\dots ,N$) is the RP value for morphology k with respect to criterion i.

The RP and RI values are then used to produce the overall decision value $D_k$ of each candidate morphology k:

$$D_k=t_1\times f_k^1+t_2\times f_k^2+\cdots +t_P\times f_k^P,(k=1,2,\dots ,N).$$

(8)

To ensure coherent and rational decision-making throughout the process, the logical consistency of the pairwise rankings (the importance score between every two elements) made at each step is checked by a quantitative consistency index. An ideally consistent pairwise ranking of three elements a, b and c should follow the transitivity relationship, given that an element is equally preferred/important to itself:

$$h_{ac}=h_{ab}\times h_{bc},$$

(9)

where $h_{ac}$ is the relative ranking of element a to c. The consistency of a ranking matrix can then be measured by a Consistency Index (CI) computed as follows:

$$\begin{array}{c}\hfill CI={\displaystyle \frac{{\lambda }_{max}-q}{q-1}},\end{array}$$

(10)

where ${\lambda }_{max}$ is the largest eigenvalue of the ranking matrix, and q is the number of morphologies N or criteria P. It can be shown that for an ideally consistent ranking matrix, the consistency index $CI$ is equal to zero. However, in practical cases, a ranking matrix can hardly end up ideally consistent, and the ranking can still be acceptable if $CI$ is reasonably smaller than that of a matrix of the same dimension with all entries randomly selected from the 1-to-9 range. Such a random matrix can numerically be constructed repeatedly, and it can be shown that for each dimension q, the consistency index ${\left(CI\right)}_R$ of the matrix converges to a value shown in

Table 7. ${\left(CI\right)}_R$ values for Consistency Ratio [78,134].

q	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
${\left(CI\right)}_R$	0	0.58	0.9	1.12	1.24	1.32	1.41	1.45	1.49	1.51	1.48	1.56	1.57	1.59	1.605	1.61	1.615	1.62	1.625

. Heuristically, a ranking matrix can be considered acceptable if

$$\begin{array}{c}\hfill CR={\displaystyle \frac{CI}{{\left(CI\right)}_R}}\le 0.10,\end{array}$$

(11)

where $CR$ is called the Consistency Ratio (CR) [133].

Examples of the AHP can be found in [135,136]. In this work, the AHP is conducted in a sequential fashion, following the flow of regolith through the system and evaluating at each stage.

5.1. Expert Criteria Prioritization and Ranking Aggregation

To help establish the RI of the criteria, a survey of experts is first conducted to obtain their ranking of the criteria. Then, the expert rankings are aggregated to achieve a set of output importance grades to inform the assignment of criteria priorities [137]. Consider M experts who express their judgments on the ranking of P criteria. These judgments are to be expressed as a set of strict and complete subjective rankings. Strict ranking does not permit ties or indifference relationships, and complete ranking implies that experts have ranked every criterion.

This paper utilizes a ranking aggregation technique, called ZM_II, to produce the set of aggregated criteria rankings [79,80]. The ZM_II technique is selected since it achieves relatively high aggregated consensus (Equations (26) and (27)), and it outputs a ratio scaling of the criteria, which can be used to better inform the RI values than simply ordering the criteria, as well as producing an ideally consistent RI matrix (double check) [81,138]. Further, the ZM_II method does not require pairwise comparisons, reducing the degree of subjectivity of the process and the required cognitive load on experts to collect data. The technique is an improvement to Thurstone’s Law of Comparative Judgment (LCJ) [139,140] by introducing anchor elements $C_t$ (M-element) and $C_b$ (Z-element), which represent the most and least important (dummy) criteria, respectively. This means that no criterion should be positioned higher than $C_t$ or lower than $C_b$. Consequently, the total number of criteria in the $Z{M}_{II}$ ranking process is considered as $(P+2)$.

The ZM_II method is a random-distribution-based technique, which is formulated by considering a random value $c_i$ for the ranking position of each criterion i. The position of each criterion is postulated to be distributed normally, such that $c_i$ ~$\mathcal{N}({\mu }_i,{\sigma }_i^2)$, where ${\mu }_i$ and ${\sigma }_i^2$ are the (unknown) mean value and variance, respectively, in terms of the ranking of the criterion. Then, assuming that the variance (and covariance) is the same for all $c_i$’s (and $c_j$’s), it can be asserted that [81]

$$p_{ij}=\mathcal{P}[(c_i-c_j)>0]=1-\Phi [-({\mu }_i-{\mu }_j)],$$

(12)

where $p_{ij}$ is the probability that the value of $c_i$ is higher than $c_j$, with $\Phi$ being the cumulative distribution function (CDF) of the right-tailed standard normal distribution $\mathcal{N}({\mu }_i,{\sigma }_i^2)$, where ${\mu }_i=0$ and ${\sigma }_i^2=1$. The value of $p_{ij}$ can be estimated by its mean ${\overline{p}}_{ij}={\displaystyle \frac{k_{ij}}{m_{ij}}}$, where $k_{ij}$ is the total comparison score across all experts between $C_i$ and $C_j$, and $m_{ij}$ is the number of experts who ranked criteria i and j. For a given comparison by an expert, a score of 1 is assigned for strict preference ($c_i>c_j$) and a score of 0.5 for indifference ($c_i=c_j$). Further, in the case of complete ranking, $m_{ij}$ equals the total number of experts M.

By re-arranging Equation (12), we express each pairwise comparison as a linear constraint between the unknown mean rankings ${\mu }_i$ and ${\mu }_j$, such that

$${\mu }_i-{\mu }_j=-z_{ij},\mathrm{where}{z}_{ij}={\Phi }^{-1}[1-{\overline{p}}_{ij}],$$

(13)

forms the skew-symmetric matrix $\mathbf{Z}\in {\mathbb{R}}^{(P+2)\times (P+2)}$. The model breaks down if all experts express the same judgments (CDF approaches $\pm \infty$), which is avoided by the simple approximation ${\Phi }^{-1}[1-{\overline{p}}_{ij}]=-1.995$ for ${\overline{p}}_{ij}\ge 0.977$, or ${\Phi }^{-1}[1-{\overline{p}}_{ij}]=1.995$ for ${\overline{p}}_{ij}\le 0.023$. For each pair of criteria, Equation (13) corresponds to a single expert-derived constraint between them, and the complete system is constructed from all comparisons:

$$\mathcal{C}=\left\{\right(i,j)\mid 1\le i<j\le (P+2\left)\right\},$$

where $(P+2)$ is the total number of criteria, including the two dummy criteria that act as anchor elements. These dummy criteria serve to constrain the range of the ratio scale output between 0 ($C_b$) and 100 ($C_t$). The total number of comparisons is $\left|\mathcal{C}\right|=Q$, and each comparison yields one row of an overdetermined set of linear equations:

$$\tilde{\mathbf{A}}\mathbf{X}=\tilde{\mathbf{B}},\tilde{\mathbf{A}}\in {\mathbb{R}}^{Q\times (P+2)},\tilde{\mathbf{B}}\in {\mathbb{R}}^{Q\times 1},$$

(14)

where $\mathbf{X}={[{\mu }_b,{\mu }_t,{\mu }_1,{\mu }_2,\dots ,{\mu }_P]}^T\in {\mathbb{R}}^{(P+2)\times 1}$ is the column vector of unknown mean rankings, and ${\mu }_t$ and ${\mu }_b$ are the obtained mean values for the dummy criteria $C_t$ and $C_b$, respectively. Each row of $\tilde{\mathbf{A}}$ encodes a constraint for some pair $(i,j)\in \mathcal{C}$, with $+1$ at position i, $-1$ at position j, and all other entries zero. From Equation (13), the corresponding entry in $\tilde{\mathbf{B}}$ is $-z_{ij}$. To produce a final square system of the form

$$\mathbf{A}\mathbf{X}=\mathbf{B},$$

(15)

the constraint pairs in $\tilde{\mathbf{A}}$ are collapsed into a Laplacian matrix $\mathbf{A}\in {\mathbb{R}}^{(P+2)\times (P+2)}$, such that each row i of $\mathbf{A}$ aggregates all the constraints for criterion i. In the case of strict and complete rankings, this results in a symmetric matrix $\mathbf{A}$ with $-1$ in all off-diagonal entries, and $(P+1)$ in all diagonal entries. The column matrix $\mathbf{B}\in {\mathbb{R}}^{(P+2)\times 1}$ is the resulting sum of the constraints, effectively meaning that $b_j={\sum }_{i=1}^{(P+2)}{z}_{ij}$. The system (15) can be solved in a closed form to obtain an estimate of the mean of the criteria rankings [81]:

$$\mathbf{X}={\left({\mathbf{A}}^T\mathbf{A}\right)}^{-1}{\mathbf{A}}^T\mathbf{B}.$$

(16)

The solution (16) assumes that the uncertainties for elements of $\mathbf{X}$ are identical and they do not affect each other, which is not true in real-life rankings. To address these shortcomings, the solution can be modified by applying the Generalized Least Squares (GLS) method [141]:

$$\mathbf{X}={\left({\mathbf{A}}^T\mathbf{W}\mathbf{A}\right)}^{-1}{\mathbf{A}}^T\mathbf{W}\mathbf{B}.$$

(17)

The matrix $\mathbf{W}$ is constructed as follows [79]:

$$\mathbf{W}={\left[{\mathbf{J}}_{\mathbf{p}}{\mathbf{\Sigma }}_{\mathbf{p}}{\mathbf{J}}_{\mathbf{p}}^T\right]}^{-1},$$

(18)

where the Jacobian matrix is approximated by

$${\mathbf{J}}_{\mathbf{p}}={\displaystyle \frac{\partial \left[{\Phi }^{-1}(1-{\overline{p}}_{ij})\right]}{\partial {\overline{p}}_{ij}}}\approx \left\{\begin{array}{cc}\left|{\displaystyle \frac{2\sqrt{2}{\overline{p}}_{ij}-\sqrt{2}}{{\overline{p}}_{ij}(1-{\overline{p}}_{ij})\sqrt{-2\sqrt{2\pi }[ln\left(4\right)+ln({\overline{p}}_{ij}-{\overline{p}}_{ij}^2)]}}}\right|,\hfill & \mathrm{if}{\overline{p}}_{ij}\ne 0.5\hfill \\ 2.506628,\hfill & \mathrm{if}{\overline{p}}_{ij}=0.5\hfill \end{array}\right.$$

(19)

and the covariance matrix ${\mathbf{\Sigma }}_{\mathbf{p}}$ is a diagonal matrix (assuming that all $p_{ij}$ estimates are statistically independent), whose diagonal elements are the variance of $p_{ij}$:

$${\sigma }_{p_{ij}}^2={\displaystyle \frac{{\overline{p}}_{ij}(1-{\overline{p}}_{ij})}{m_{ij}}}.$$

(20)

The system of Equation (17) can now be solved for $\mathbf{X}$. To scale the solutions into a $[0,K]$ range, the following transformation is applied:

$${\mathbf{X}}^{\prime }={\displaystyle \frac{K}{{\mu }_t-{\mu }_b}}{[0,({\mu }_t-{\mu }_b),({\mu }_1-{\mu }_b),({\mu }_2-{\mu }_b),\dots ,({\mu }_P-{\mu }_b)]}^T.$$

(21)

The calculated mean values ($\mathbf{X}$ or ${\mathbf{X}}^{\prime }$) for the criteria can be used to recover an aggregated ranking by simply ordering them from the highest mean value (top rank) to the lowest (bottom rank).

The uncertainty in the estimation of mean values of the criteria ranking can be quantified by using the Multivariate Law of Propagation of Uncertainty (MLPU) [141]:

$${\Sigma }_{\mathbf{X}}={\left({\mathbf{A}}^T\mathbf{W}\mathbf{A}\right)}^{-1},$$

(22)

and using the transformation relation yields

$${\Sigma }_{{\mathbf{X}}^{\prime }}={\mathbf{J}}_{{\mathbf{X}}^{\prime }}{\Sigma }_{\mathbf{X}}{\mathbf{J}}_{{\mathbf{X}}^{\prime }}^T.$$

(23)

The Jacobian in Equation (23) can be constructed as follows [79]:

$${\mathbf{J}}_{{\mathbf{X}}^{\prime }}=\left[\begin{array}{ccccc}{\displaystyle \frac{\partial x_b^{\prime }}{\partial x_b}}& {\displaystyle \frac{\partial x_b^{\prime }}{\partial x_t}}& {\displaystyle \frac{\partial x_b^{\prime }}{\partial x_1}}& \dots & {\displaystyle \frac{\partial x_b^{\prime }}{\partial x_P}}\\ {\displaystyle \frac{\partial x_t^{\prime }}{\partial x_b}}& {\displaystyle \frac{\partial x_t^{\prime }}{\partial x_t^{\prime }}}& {\displaystyle \frac{\partial x_t^{\prime }}{\partial x_1}}& \dots & {\displaystyle \frac{\partial x_t^{\prime }}{\partial x_P}}\\ {\displaystyle \frac{\partial x_1^{\prime }}{\partial x_b}}& {\displaystyle \frac{\partial x_1^{\prime }}{\partial x_t}}& {\displaystyle \frac{\partial x_1^{\prime }}{\partial x_1}}& \dots & {\displaystyle \frac{\partial x_1^{\prime }}{\partial x_P}}\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ {\displaystyle \frac{\partial x_P^{\prime }}{\partial x_b}}& {\displaystyle \frac{\partial x_P^{\prime }}{\partial x_t}}& {\displaystyle \frac{\partial x_P^{\prime }}{\partial x_1}}& \dots & {\displaystyle \frac{\partial x_P^{\prime }}{\partial x_P}}\end{array}\right]={\displaystyle \frac{K}{{\mu }_t-{\mu }_b}}\left[\begin{array}{ccccc}0& 0& 0& \dots & 0\\ 0& 0& 0& \dots & 0\\ -{\displaystyle \frac{{\mu }_t-{\mu }_1}{{\mu }_t-{\mu }_b}}& -{\displaystyle \frac{{\mu }_1-{\mu }_b}{{\mu }_t-{\mu }_b}}& 1& \dots & 0\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ -{\displaystyle \frac{{\mu }_t-{\mu }_P}{{\mu }_t-{\mu }_b}}& -{\displaystyle \frac{{\mu }_P-{\mu }_b}{{\mu }_t-{\mu }_b}}& 0& \dots & 1\end{array}\right]$$

(24)

A 95% confidence interval can then be calculated assuming a normal distribution for $p_{ij}$ and $x_i^{\prime }$:

$$\mathrm{confidence}\mathrm{interval}=x_i^{\prime }\pm 2{\sigma }_i^{\prime },i=1,2,\dots ,P$$

(25)

where ${\sigma }_i^{\prime }=\sqrt{{\Sigma }_{{\mathbf{X}}^{\prime }(i,i)}}$.

A variety of methods can be employed to examine the degree of consensus, i.e., the agreement between the aggregated ranking and those provided by the experts. This paper uses a global p-indicator that quantifies the degree of consensus by calculating a ratio of the number of comparison agreements between individual expert rankings and the aggregated ranking over the total number of pairwise comparisons across all experts [138]:

$$p={\displaystyle \frac{{\sum }_{k=1}^M{a}_k}{M\times \left(\genfrac{}{}{0pt}{}{P}{2}\right)}}={\displaystyle \frac{1}{M}}\sum _{k=1}^M\left[{\displaystyle \frac{a_k}{\left(\genfrac{}{}{0pt}{}{P}{2}\right)}}\right],$$

(26)

where $a_k$ is the sum of the agreements for $\left({\displaystyle \genfrac{}{}{0pt}{}{P}{2}}\right)={\displaystyle \frac{P(P-1)}{2}}$ pairwise comparisons between the k-th expert’s ranking and the aggregated ranking. Consequently, this indicator varies between 0 for full disagreement and 1 for full agreement between the expert rankings and the aggregated ranking.

Further, it is possible to quantify the degree of inter-expert agreement using Kendall’s coefficient of concordance for strict and complete rankings [80,142]:

$$w={\displaystyle \frac{12\left({\sum }_{i=1}^P{R}_i^2\right)-3M^{2}P{(P+1)}^2}{M^{2}P(P^2-1)}},$$

(27)

where $R_i$ is the sum of the rank positions (e.g., 1, 2, …, or P) of the i-th criterion across all experts. This indicator also varies between 0 for full disagreement and 1 for full agreement between the expert rankings.

5.2. Criteria Prioritization and Evaluation

Four core functional modules from the functional analysis in Figure 8 are selected for expert evaluation and further analysis, namely, intake lunar regolith (FM-1), deploy empty RCU (FM-2), interface with and secure RCU (FM-3), and transport RCU to desired pose (FM-4). A total of 47 engineers working in aerospace were surveyed on the criteria for each of the four selected functional modules. The pool consists primarily of engineers from MDA Space, NASA, and the University of Toronto affiliates. The surveyed engineers all have experience working with space and/or lunar robotic systems. The expert pool breakdown by specialization can be seen in Figure 12, and by years of experience in Figure 13. There is a fair distribution in specializations, with most respondents specializing in GNC (Guidance, Navigation and Control) or Mechanical engineering. Meanwhile, the number of respondents has a reverse relationship with the experience duration of experts, which is to be expected as it is more difficult to find and get a hold of the most experienced engineers, i.e., over 10–15 years experience, who may be in more prominent positions.

The collective rankings of the selected criteria for each of the four selected modules are shown in

Table 8. Collective rankings produced by ZM_II aggregation method.

Rank	FM-1	FM-2	FM-3	FM-4
1	Reliability: $66.8\pm 34.3$	Reliability: $67.0\pm 36.1$	Reliability: $70.1\pm 32.5$	Reliability: $67.6\pm 30.7$
2	Longevity: $59.9\pm 31.6$	Longevity: $63.5\pm 34.6$	Longevity: $64.2\pm 30.4$	Longevity: $62.8\pm 29.0$
3	Exposure to dust: $54.4\pm 29.3$	Complexity: $50.9\pm 29.2$	Exposure to dust: $57.4\pm 27.9$	Exposure to dust: $58.9\pm 27.5$
4	Complexity: $52.6\pm 28.5$	Exposure to dust: $50.1\pm 28.9$	Complexity: $55.0\pm 26.9$	Complexity: $51.5\pm 24.6$
5	Mass: $50.4\pm 27.5$	Power: $48.1\pm 27.7$	Power: $50.5\pm 24.8$	Workspace—module volume ratio: $49.3\pm 23.6$
6	Dust production: $46.8\pm 25.9$	Mass: $47.3\pm 27.5$	Dust production: $49.2\pm 24.5$	Power: $48.9\pm 23.4$
7	Power: $46.3\pm 25.5$	Storage volume: $40.4\pm 24.2$	Mass: $48.0\pm 24.0$	Manipulation accuracy: $47.1\pm 22.7$
8	Regolith compaction: $36.4\pm 21.0$	Transport speed: $32.7\pm 20.2$	Attachment force: $47.0\pm 23.4$	Dust production: $46.5\pm 22.5$
9	Delivery flow rate: $36.2\pm 20.7$	-	Rigid attachment: $46.7\pm 23.2$	Mass: $45.5\pm 22.0$
10	-	-	Attachment points: $32.6\pm 17.0$	Structural rigidity: $41.4\pm 20.3$
11	-	-	Attachment speed: $29.3\pm 15.5$	Dexterity: $40.8\pm 20.1$
12	-	-	-	Manipulation resolution: $39.6\pm 19.5$
w	0.257	0.311	0.332	0.204
p	0.689	0.713	0.730	0.661

. It can be observed that criteria such as reliability, longevity, exposure to dust, and complexity are consistently ranked in positions 1–4. This can be attributed to the fact that producing fail-safe systems is a priority in the aerospace industry (from which experts were pooled), and that these criteria appear relatively similar without additional information being presented to the survey respondents. The remaining criteria experience a fair amount of variability in their ranking positions, although it can be noted that module-specific criteria, e.g., “delivery flow rate” for FM-1, are generally ranked near the bottom. No notable differences were observed in the expert rankings based on the varying experience levels and specializations, indicating that the rankings represent the general priorities of the expert pool.

Although the w and p indicators cannot be directly compared, one can generally observe that there was a low inter-expert concordance (below 35% for w) before applying the aggregation technique. After applying the ZM_II aggregation technique, the p-indicator lies in the range from 65% to 75%, indicating a relatively good performance for the aggregation technique. The variance in the initial expert responses (represented by w) inherently limits how much the p-indicator can tend towards 100% agreement with the collective ranking.

The aggregated rankings are used to determine the RI of the criteria for the AHP, as explained in the subsequent sections. Since the scaled aggregated mean values of expert rankings ${\mathbf{X}}^{\prime }$, derived from the ZM_II method, are ratio scale representations of the expert preferences, by normalizing them, they can effectively replace the normalized RI values $t_i$ in the AHP (see Equation (5)):

$$\mathbf{T}={[{\displaystyle \frac{x_1^{\prime }}{{\sum }_{i=1}^P{x}_i^{\prime }}},\dots ,{\displaystyle \frac{x_P^{\prime }}{{\sum }_{i=1}^P{x}_i^{\prime }}}]}^T={\displaystyle \frac{K}{({\mu }_t-{\mu }_b){\sum }_{i=1}^P{x}_i^{\prime }}}{[({\mu }_1-{\mu }_b),\dots ,({\mu }_P-{\mu }_b)]}^T$$

(28)

where $\mathbf{T}\in {\mathbb{R}}^{P\times 1}$ is a column vector containing all the normalized RI values for the P criteria (ignoring dummy criteria). The normalization maintains the relative ratios between the criteria, preserving the ratio scale. The new RI values are used directly when calculating the decision values in Equation (8).

It is worth noting that the ZM_II method offers a way to reconcile non-strict (tie-permitting) and incomplete rankings with the AHP, allowing for some information loss due to the lack of pairwise comparisons. Typically, it would not be possible to have incomplete pairwise comparisons for the AHP, but the ZM_II method allows for a ratio scale output using a variety of input formats. This can make the data collection process easier, since it avoids the need for experts to conduct the tedious number of pairwise comparisons conventionally required for the AHP.

To produce the RP matrices of solutions with regard to the criteria, each criterion is quantified by a metric obtained from physical characteristics of the design problem. Given some metric evaluation ${\overline{r}}_k$ for a candidate solution k ($k=1,\dots ,N$), the metric evaluations are adjusted to a 1-to-9 scale:

$$Q\left({\overline{r}}_k\right)=1+{\displaystyle \frac{{\overline{r}}_k-{\overline{r}}_{min}}{{\overline{r}}_{max}-{\overline{r}}_{min}}}\times (9-1).$$

(29)

Then, the AHP ratio for an RP matrix $\mathbf{R}$ is formulated as

$$r_{kl}\approx {\displaystyle \frac{Q\left({\overline{r}}_k\right)}{Q\left({\overline{r}}_l\right)}},k,l=1,\dots ,N$$

(30)

The entries $r_{kl}$ are rounded to the nearest AHP 1–9 scale value (see Table 6). This would form the RP matrix for a given criterion, adjusting all metric comparisons to the AHP 1–9 scale.

6. Design Assessment: Application

The formulations developed in Section 5 are implemented in this section for the evaluation of the candidate solutions for the four primary functional modules, namely, intake lunar regolith, deploy empty RCU, interface with and secure RCU, and transport RCU to desired pose.

6.1. Intake Lunar Regolith

The RCU Filling subsystem consists of five functional modules, namely, regolith intake, beneficiation, filling, sealing, and deploying the empty RCU (see Figure 8). A primary morphological assessment of the RCU Filling subsystem is with regard to the intake mechanism, which directly manipulates the lunar regolith toward the filling module. The criteria for evaluation of the intake module, along with the normalized RI of the criteria (calculated as described in Section 5.2), are shown in

Table 9. Evaluation criteria for the intake module (# marks number).

ID	Criteria	Description	Metric	Goal	RI (${\mathit{t}}_{\mathit{i}}$)
C-I-1	Exposure to dust	The number of interfaces between moving parts, which are exposed to the regolith.	# of interfaces	Min	0.121
C-I-2	Dust production	The potential for dust production due to agitation, sharp movements, dumping of the regolith, etc.	ordinal scale (Likert 1–5)	Min	0.104
C-I-3	Longevity	The anticipated failure rate of the module based on motor load and transmission [143].	failures per million operating hours	Min	0.133
C-I-4	Reliability	The number of independent points of single failure of the module.	# of points	Min	0.148
C-I-5	Power	The power required to move 30 kg of regolith into the subsystem; calculated as a function of the work required to move the regolith per the anticipated required time according to the delivery flow rate.	Watts	Min	0.103
C-I-6	Mass	The anticipated mass of the module; calculated by taking the estimated surface area of the candidate solution and assuming an average thickness of aluminum of 0.01 m for the structure.	kilograms	Min	0.112
C-I-7	Complexity	The module complexity is measured as the total number of moving parts. This includes motors, gears, vibrating surfaces, and other parts moving in the workspace.	# of moving parts	Min	0.117
C-I-8	Delivery flow rate	The estimated delivery flow rate of the regolith into the filling module.	kilograms per minute	Max	0.0805
C-I-9	Regolith compaction	The anticipated degree of regolith compaction caused by the module, which occurs when inter-particle compression forces are increased.	ordinal scale (Likert 1–5)	Min	0.0809

The term $t_i$ is explained in Section 5.

.

The candidate solutions for the intake module are depicted and described in

Table 10. Intake mechanism candidate solutions.

	Concept	Description
Auger and Hopper		A hopper which collects the lunar regolith from IPEx, using gravity to feed downward towards a horizontal auger which forces the regolith into the system.
Vibrating Intake Chute		Leverages the angular vibration of a surface to supplement gravity in feeding the lunar regolith down a slope. The vibration and large surface area of the feeding surface allows for regolith to remain well distributed.
Vibratory Bowl Feeder		Similar to the Vibrating Intake Chute, this concept allows for regolith to be dumped into its centre and leverages vibration to propagate regolith upwards along a feeding path around its cylindrical walls. This method is commonly used for dispensing in terrestrial factory settings.
Front Loader		A traditional loading method, often used in excavation systems, where regolith is dumped into a stationary container, which is then rotated upwards until the regolith falls into the system.
Vibrating Funnel		A funnel which collects regolith from the excavator and propagates it towards a smaller exit leading into the system. This concept also involves a couple of sparse grating layers, combined with vibration, which aim to prevent compaction and promote the effective flow of regolith.

. These candidates focus on keeping the regolith consistently flowing to prevent the compaction and high-interlocking behaviour of the lunar regolith. A number of candidates have been inspired by the study in [144] regarding lunar dust mitigation and lunar regolith conveyance (and intake) technologies. Of particular interest is the success of hopper and auger systems, as well as vibrating surfaces and bowl feeders. Another study specifically designed an RCU creation and stacking system, using a funnel and propeller shaft for sand (regolith analogue) intake [76]. Other studies have successfully implemented auger [145,146] and vibration [146,147,148] conveyance of lunar regolith simulants. Furthermore, terrestrial sandbaggers are also considered, which largely follow a hopper and auger or funnel morphology. Examples include the Sandbagger [149] and the PanPac Engineering end-to-end palletization system [150]. Some works have also investigated funnel systems for the conveyance of lunar regolith simulant, successfully demonstrating their feasibility [151,152]. Finally, front loaders are commonly used mechanisms for moving soil in terrestrial applications, and have been tested for lunar excavation purposes [14].

The calculations and formulations for some of the more involved metrics are detailed in the following subsections.

6.1.1. CR-F-3: Longevity

The longevity of the candidate solutions is considered as a subset of reliability in terms of anticipated failure rate. There exist many handbooks detailing methods for the theoretical calculation of failure rates for electro-mechanical systems, largely based on empirical studies of components and their failures. Notably, the MIL-217F handbook [153] by the American Department of Defense is a cornerstone often used in these estimations. However, the last release of the handbook was in 1991 (Notice 2), and some studies have examined its obsolescence [154]. Some newer handbooks, such as the NSWC-11 handbook [143] by the American Naval Surface Warfare Center, have since been released. Other recent handbooks also exist, such as the Toshiba TDSC Reliability Handbook [155] and the 217 Plus Handbook [156] (a privately developed update to MIL-217F), but these largely focus on the reliability of electrical and semiconductor components. The newest and ongoing attempt to create a reliability standard is FIDES [157], with the latest edition being released in 2022. However, FIDES requires a registration process, and its methodology has been criticized by works such as [158]. The NSWC-11 handbook was found to be the most recent, relevant and reliable source for the failure rate estimation of our electro-mechanical candidate solutions.

In general, the NWSC-11 handbook defines the failure of components by establishing the base failure rate and multiplying this base rate by related factors. These failure rates can be summed up to approximate the failure rate of a larger system or complex components. Within the context of the design, we consider the reliability of the gears (transmission) and the motor, which is assumed to be a brushless DC motor (common for space applications). The failure rate of gears is defined as

$${\lambda }_G={\lambda }_{G,B}·C_{GS}·C_{GP}·C_{GA}·C_{GL}·C_{GT}·C_{GF},$$

(31)

where ${\lambda }_G$ is the failure rate of the gear under specific conditions, ${\lambda }_{G,B}$ is the base failure rate of the gear, and the C terms relate to the operating conditions (see the handbook for full details). All failure rates are considered in terms of failures/million operating hours. The failure rate of a motor is defined as

$${\lambda }_M=({\lambda }_{M,B}·C_{SF})+{\lambda }_{WI}+{\lambda }_{BS}+{\lambda }_{ST}+{\lambda }_{AS}+{\lambda }_{BE}+{\lambda }_{GR}+{\lambda }_C,$$

(32)

where ${\lambda }_M$ is the total failure rate of the motor system, ${\lambda }_{M,B}$ is the base failure rate of the motor, $C_{SF}$ is the motor load service factor, and the other $\lambda$ terms relate to the additional components of the motor system (see the handbook for full details). Considering the current stage of design, many factors to perform the full estimation from the handbook are yet to be determined, so we consider a simplified model for the candidate designs

$${\lambda }_C=n_{motors}·{\lambda }_{M,B}·C_{SF}+n_{gears}·{\lambda }_{G,B},$$

(33)

where ${\lambda }_C$ is the estimated failure rate of the candidate design, $n_{motors}$ is the total number of motors, and $n_{gears}$ is the total number of gears.

All candidate designs for the intake mechanism are considered to have only one brushless DC motor with a base failure rate of 1.75 failures/million operating hours. However, the service factor on these motors varies. The Auger and Hopper System is considered to have a uniform load due to continuous and consistent operation ($C_{SF}=1.00$). The Front Loader is considered to have a medium impact due to its frequently reversible motor operation and moderate shock during loading and unloading ($C_{SF}=2.00$). The Vibrating Intake Chute, Vibratory Bowl Feeder, and Vibrating Funnel are considered to have a heavy load due to heavy vibration ($C_{SF}=3.00$). Only the Auger and Hopper System, as well as the Front Loader, are estimated to necessitate a transmission due to lower operating speeds. Both of these candidates are estimated to have two gears as a minimum, each with a base failure rate of 1 failure/million operating hours. These formulations allow for the estimation of the failure rates for the candidates as follows:

$${\lambda }_{\mathrm{Auger}/\mathrm{Hopper}}=1·1.75·1.00+2·1=3.75$$

(34)

$${\lambda }_{\mathrm{Chute}}=1·1.75·3.00+0·1=5.25$$

(35)

$${\lambda }_{\mathrm{Bowl}\mathrm{Feeder}}=1·1.75·3.00+0·1=5.25$$

(36)

$${\lambda }_{\mathrm{Front}\mathrm{Loader}}=1·1.75·2.00+2·1=5.50$$

(37)

$${\lambda }_{\mathrm{Funnel}}=1·1.75·3.00+0·1=5.25.$$

(38)

6.1.2. CR-F-5: Power

Although it is challenging to accurately estimate the power consumption of the candidate designs without specific knowledge of the motors, the estimation is done by considering the geometry of the candidate designs and the Work necessary to move one IPEx load (30 kg) to the output point. The Work done is then divided by the anticipated time to deliver 30 kg of regolith according to the estimated delivery flow rate (see Section 6.1.4). The Work can generally be calculated as

$$W=d·(F_g+F_f),$$

(39)

where d is the average distance traveled by the lunar regolith, $F_g$ is the force due to gravity, and $F_f$ is the force due to friction. The Power can then be found as

$$P={\displaystyle \frac{W}{(30\mathrm{kg}/D)·60\frac{\mathrm{s}}{\min }}},$$

(40)

where D is the estimated delivery rate of the candidate in kg/min.

The coefficient of kinetic friction between aluminum (the assumed material of the structure) and the lunar regolith has been empirically found to be 0.8882 [159]. The acceleration due to gravity on the Moon is 1.625 m/s². Using these values, the anticipated power of candidate solutions is estimated, starting with the Auger and Hopper System

$$\begin{array}{cc}\hfill W_{\mathrm{Auger}/\mathrm{Hopper}}& =0.30\mathrm{m}·(0+0.8882·1.625{\displaystyle \frac{\mathrm{m}}{{\mathrm{s}}^2}}·30\mathrm{kg}·cos(0°))=13.00\mathrm{J}\hfill \end{array}$$

(41)

$$\begin{array}{cc}\hfill P_{\mathrm{Auger}/\mathrm{Hopper}}& ={\displaystyle \frac{13.00\mathrm{J}}{(30\mathrm{kg}/6.32\frac{\mathrm{kg}}{\min })·60\frac{\mathrm{s}}{\min }}}=0.046\mathrm{W}.\hfill \end{array}$$

(42)

For the Vibrating Chute,

$$\begin{array}{cc}\hfill W_{\mathrm{Chute}}& =0.42\mathrm{m}·(0+0.8882·1.625{\displaystyle \frac{\mathrm{m}}{{\mathrm{s}}^2}}·30\mathrm{kg}·cos(16.70°))=17.42\mathrm{J}\hfill \end{array}$$

(43)

$$\begin{array}{cc}\hfill P_{\mathrm{Chute}}& ={\displaystyle \frac{17.42\mathrm{J}}{(30\mathrm{kg}/6.99\frac{\mathrm{kg}}{\min })·60\frac{\mathrm{s}}{\min }}}=0.068\mathrm{W}.\hfill \end{array}$$

(44)

For the Vibratory Bowl Feeder,

$$\begin{array}{cc}\hfill W_{\mathrm{Bowl}\mathrm{Feeder}}& =10.12\mathrm{m}·(1.625{\displaystyle \frac{\mathrm{m}}{{\mathrm{s}}^2}}·30\mathrm{kg}+0.8882·1.625{\displaystyle \frac{\mathrm{m}}{{\mathrm{s}}^2}}·30\mathrm{kg}·cos(1.36°))=931.44\mathrm{J}\hfill \end{array}$$

(45)

$$\begin{array}{cc}\hfill P_{\mathrm{Bowl}\mathrm{Feeder}}& ={\displaystyle \frac{931.44\mathrm{J}}{(30\mathrm{kg}/1.68\frac{\mathrm{kg}}{\min })·60\frac{\mathrm{s}}{\min }}}=0.87\mathrm{W}.\hfill \end{array}$$

(46)

For the Front Loader,

$$\begin{array}{cc}\hfill W_{\mathrm{Front}\mathrm{Loader}}& =0.47\mathrm{m}·(1.625{\displaystyle \frac{\mathrm{m}}{{\mathrm{s}}^2}}·30\mathrm{kg}+0)=22.91\mathrm{J}\hfill \end{array}$$

(47)

$$\begin{array}{cc}\hfill P_{\mathrm{Bowl}\mathrm{Feeder}}& ={\displaystyle \frac{22.91\mathrm{J}}{(30\mathrm{kg}/60\frac{\mathrm{kg}}{\min })·60\frac{\mathrm{s}}{\min }}}=0.76\mathrm{W}.\hfill \end{array}$$

(48)

For the Vibrating Funnel,

$$\begin{array}{cc}\hfill W_{\mathrm{Funnel}}& =0.088\mathrm{m}·(0+0.8882·1.625{\displaystyle \frac{\mathrm{m}}{{\mathrm{s}}^2}}·30\mathrm{kg}·cos(55.22°))=2.17\mathrm{J}\hfill \end{array}$$

(49)

$$\begin{array}{cc}\hfill P_{\mathrm{Funnel}}& ={\displaystyle \frac{2.17\mathrm{J}}{(30\mathrm{kg}/160\frac{\mathrm{kg}}{\min })·60\frac{\mathrm{s}}{\min }}}=0.19\mathrm{W}.\hfill \end{array}$$

(50)

6.1.3. CR-F-6: Mass

The mass of each candidate is estimated by finding the surface area and assuming a generic 0.01 m aluminum thickness across the entire surface. The density of aluminum (2710 kg/m³) is then used to calculate the estimated mass. For each candidate,

$$m_{\mathrm{Auger}/\mathrm{Hopper}}=1.45{\mathrm{m}}^2·0.01\mathrm{m}·2710{\displaystyle \frac{\mathrm{kg}}{{\mathrm{m}}^3}}=43.09\mathrm{kg}$$

(51)

$$m_{\mathrm{Chute}}=1.44{\mathrm{m}}^2·0.01\mathrm{m}·2710{\displaystyle \frac{\mathrm{kg}}{{\mathrm{m}}^3}}=39.08\mathrm{kg}$$

(52)

$$m_{\mathrm{Bowl}\mathrm{Feeder}}=2.67{\mathrm{m}}^2·0.01\mathrm{m}·2710{\displaystyle \frac{\mathrm{kg}}{{\mathrm{m}}^3}}=72.36\mathrm{kg}$$

(53)

$$m_{\mathrm{Front}\mathrm{Loader}}=0.63{\mathrm{m}}^2·0.01\mathrm{m}·2710{\displaystyle \frac{\mathrm{kg}}{{\mathrm{m}}^3}}=17.07\mathrm{kg}$$

(54)

$$m_{\mathrm{Funnel}}=0.74{\mathrm{m}}^2·0.01\mathrm{m}·2710{\displaystyle \frac{\mathrm{kg}}{{\mathrm{m}}^3}}=23.04\mathrm{kg}.$$

(55)

6.1.4. CR-F-8: Delivery Flow Rate

The delivery flow rate (represented as $\dot{m}$) of the candidates is considered as the flow of regolith at the interface between the intake and the bagging mechanisms. The process for calculating this value varies depending on the nature of the mechanism, e.g., vibration or auger. Starting with the Auger and Hopper System, one study finds that an auger with a 15.24 cm radius is able to convey lunar regolith simulant at an average rate of 12.9426 kg/min in vacuum conditions [146]. This result is used to linearly approximate the anticipated output of the Auger and Hopper System, which uses an auger with a 10cm radius. The approximation can be done by considering the surface area of the augers, since this represents the contact surface with the regolith:

$${\dot{m}}_{\mathrm{Auger}/\mathrm{Hopper}}={\displaystyle \frac{\pi ·r_{\mathrm{Auger}/\mathrm{Hopper}}^2}{\pi ·r_{\mathrm{experiment}}^2}}·{\dot{m}}_{\mathrm{experiment}}={\displaystyle \frac{{\left(0.1\mathrm{m}\right)}^2}{{\left(0.1524\mathrm{m}\right)}^2}}·12.9426{\displaystyle \frac{\mathrm{kg}}{\min }}=8.49{\displaystyle \frac{\mathrm{kg}}{\min }}.$$

(56)

Next, the flow rate for the Vibrating Chute and the Vibratory Bowl Feeder can be linearly approximated using the same study [146], which implements a 91.44 cm × 33.02 cm × 13.97 cm vibratory conveyor for lunar regolith simulant, achieving a mass flow rate of 12.5916 kg/min in vacuum conditions. In this case, the output of the candidates can be approximated by scaling the outlet width

$${\dot{m}}_{\mathrm{Chute}}={\displaystyle \frac{w_{\mathrm{Chute}}}{w_{\mathrm{experiment}}}}·{\dot{m}}_{\mathrm{experiment}}={\displaystyle \frac{0.4\mathrm{m}}{0.3302\mathrm{m}}}·12.5916{\displaystyle \frac{\mathrm{kg}}{\min }}=15.25{\displaystyle \frac{\mathrm{kg}}{\min }}.$$

(57)

However, the Vibrating Chute also needs to account for the additional flow due to gravity caused by the incline. By working out the kinematics, it is found that a factor of ${\displaystyle \frac{1}{cos\left(\alpha \right)}}$, where $\alpha$ is the slope, can account for the additional horizontal flow of a particle down the ramp. Consequently,

$${\dot{m}}_{\mathrm{Chute}}={\displaystyle \frac{1}{cos(16.7°)}}·15.25{\displaystyle \frac{\mathrm{kg}}{\min }}=15.92{\displaystyle \frac{\mathrm{kg}}{\min }}.$$

(58)

Similarly, for the Vibratory Bowl Feeder,

$${\dot{m}}_{\mathrm{Bowl}\mathrm{Feeder}}={\displaystyle \frac{w_{\mathrm{Bowl}\mathrm{Feeder}}}{w_{\mathrm{experiment}}}}·{\dot{m}}_{\mathrm{experiment}}={\displaystyle \frac{0.1\mathrm{m}}{0.3302\mathrm{m}}}·12.5916{\displaystyle \frac{\mathrm{kg}}{\min }}=3.81{\displaystyle \frac{\mathrm{kg}}{\min }}.$$

(59)

In this case, the small upwards slope of 1.36° does not affect the flow rate locally due to the small angle approximation.

The Front Loader is different from the other intake methods, since it is a discrete device. As such, the delivery flow rate for the Front Loader can be approximated to be the entire load at once, giving 60 kg/min.

Finally, the Vibrating Funnel delivery flow rate can be approximated by scaling the outlet surface area of a hopper experiment conducted in the lunar gravity and vacuum conditions [152]

$${\dot{m}}_{\mathrm{Funnel}}={\displaystyle \frac{l_{\mathrm{Funnel}}\times w_{\mathrm{Funnel}}}{l_{\mathrm{experiment}}\times w_{\mathrm{experiment}}}}·{\dot{m}}_{\mathrm{experiment}}={\displaystyle \frac{0.3\mathrm{m}\times 0.2\mathrm{m}}{0.015\mathrm{m}\times 0.018\mathrm{m}}}·0.72{\displaystyle \frac{\mathrm{kg}}{\min }}=160{\displaystyle \frac{\mathrm{kg}}{\min }}.$$

(60)

6.1.5. Intake Lunar Regolith Module Evaluation

Each intake module candidate solution is designed to be compatible with system requirements, particularly with respect to the interface with the IPEx excavation system. As such, the opening to each candidate solution is capable of capturing regolith from one of the two IPEx bucket drums, which measures 0.625 m × 0.312 m. Further, each candidate solution is designed to intake 60 kg of lunar regolith, which is double the capacity of IPEx. Assuming an average lunar regolith density of 1.58 g/cm³ [82], this represents a storage volume of 60,000 g per $1.58$ g/cm³ $\approx 38,000$ cm³ (≈0.038 m³) for each candidate solution. The evaluation of each candidate solution according to the metrics defined by the criteria is shown in

Table 11. Metric evaluation values for intake module (# marks number).

ID	Metric	Auger and Hopper	Vibrating Intake Chute	Vibratory Bowl Feeder	Front Loader	Vibrating Funnel
C-I-1	# of interfaces	0	0	0	1	0
C-I-2	ordinal scale (Likert 1–5)	1	2	2	3	2
C-I-3	failures per million operating hours	3.75	5.25	5.25	5.5	5.25
C-I-4	# of points	2	1	1	2	1
C-I-5	Watts	0.046	0.068	0.87	0.76	0.19
C-I-6	kilograms	43.09	39.08	72.36	17.07	23.04
C-I-7	# of moving parts	4	2	2	4	2
C-I-8	kilograms per minute	6.32	6.99	1.68	60	160
C-I-9	ordinal scale (Likert 1–5)	3	1	1	4	2

.

The quantitative estimations of the criteria are directly used to produce the RP pairwise rankings for all candidates relative to each criterion, as detailed in Section 5.2. An example of the RP pairwise rankings for C-I-6 (mass) is shown in

Table 12. RP matrix for intake module for C-I-6 (Mass).

Candidate	Auger and Hopper	Vibrating Intake Chute	Vibratory Bowl Feeder	Front Loader	Vibrating Funnel	RP (${\mathit{f}}_{\mathit{k}}^{\mathit{i}}$)
Auger and Hopper	1	1	5	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	0.165
Vibrating Intake Chute	1	1	6	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	1	0.200
Vibratory Bowl Feeder	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$5$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$6$}\right.$	1	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$7$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$8$}\right.$	0.0379
Front Loader	2	2	7	1	1	0.324
Vibrating Funnel	2	1	8	1	1	0.278
CR	0.01 ($\ll 0.1$)

The term $f_k^i$ is explained in Section 5.

, performed similarly for each of the intake module criteria.

A summary of the RI of criteria, the RP of the candidate solutions, and the final decision values is shown in

Table 13. AHP decision matrix for intake module.

		RP ($f_k^i$)
ID	RI ($t_i$)	Auger and Hopper	Vibrating Intake Chute	Vibratory Bowl Feeder	Front Loader	Vibrating Funnel
C-I-1	0.121	0.243	0.243	0.243	0.0270	0.243
C-I-2	0.104	0.379	0.194	0.194	0.0396	0.194
C-I-3	0.133	0.540	0.132	0.132	0.0644	0.132
C-I-4	0.148	0.0354	0.310	0.310	0.0345	0.310
C-I-5	0.103	0.300	0.300	0.0349	0.0732	0.293
C-I-6	0.112	0.165	0.199	0.0338	0.324	0.278
C-I-7	0.117	0.0345	0.310	0.310	0.0345	0.310
C-I-8	0.0805	0.0751	0.0751	0.0676	0.246	0.536
C-I-9	0.0809	0.138	0.283	0.283	0.0351	0.262
Decision Values (Dk)		0.217	0.232	0.185	0.0916	0.275

The terms $t_i$, $f_k^i$, and $D_k$ are explained in Section 5.

, indicating a preference for the Vibrating Funnel, followed by the Auger and Hopper System and the Vibrating Intake Chute.

6.2. Deploy Empty RCU

The RCU deployment functional module is also selected for morphological assessment within the RCU Filling subsystem. The evaluation criteria for the module are shown in

Table 14. Evaluation criteria for RCU deployment module (# marks number).

ID	Criteria	Description	Metric	Goal	RI (${\mathit{t}}_{\mathit{i}}$)
C-D-1	Exposure to dust	The number of interfaces between moving parts, which are exposed to the regolith.	# of interfaces	Min	0.125
C-D-2	Longevity	The anticipated failure rate of the module based on motor load, transmission, and other mechanism failure rates, such as conveyor belts, chain-link conveyors, and springs [143].	failures per million operating hours	Min	0.159
C-D-3	Reliability	The number of independent points of single failure of the module.	# of points	Min	0.168
C-D-4	Power	The power required to move the RCU along the track to the final deployment location. Approximated by the total number of required motors.	# motors	Min	0.120
C-D-5	Mass	The anticipated mass of the module; calculated by taking the estimated surface area of the candidate solution and assuming an average thickness of aluminum of 0.01 m for the structure.	kilograms	Min	0.118
C-D-6	Complexity	The module complexity is measured as the total number of moving parts. This includes motors, gears, vibrating surfaces, and other parts moving in the workspace.	# of moving parts	Min	0.127
C-D-7	Transport speed	How quickly the RCUs can be moved from the stored to the deployed configuration.	Nominal scale (very slow to very fast)	Max	0.0816
C-D-8	Storage volume	The volume of RCUs that can be stored by the module.	meters cubed	Max	0.101

The term $t_i$ is explained in Section 5.

, along with the normalized RI of criteria, calculated as described in Section 5.2. The concept of an automated RCU bagger is studied in [76], with the final design using a series of rollers and hooks to store and open RCUs. Although there has been little further formal research on different methods of RCU storage and deployment, a number of industrial bagging systems can be used as inspiration [160,161,162,163,164,165,166]. Generally, baggers can perform horizontal or vertical filling. In the case of the RCU deployment module, the filling should be performed vertically to allow gravity to naturally guide the granular lunar regolith into the bag. Further, the candidates should use a variation of Form–Fill–Seal (FFS) bagging, which is commonly used for granular materials such as feedstock or fertilizer. This would involve the complete opening of the bag mouth prior to filling. However, it is likely that the full “forming” element is unnecessary, since the granular regolith can help force the rest of the non-rigid RCU to open as it is filled.

The RCU deployment functional module can be broken down into two primary tasks: RCU storage and RCU deployment. The RCUs could be stored either in a rigid container or as a part of a roll of RCUs. The RCU deployment task involves the transport mechanism to place the RCU in the filling configuration (open mouth). This could be done using a track (chain-link, wire, or conveyor) or with a series of rollers if the RCU is stored as a roll. Considering these options, the storage and deployment concepts are listed in

Table 15. Candidate solutions for RCU deployment module.

Storage Concept	Deployment Concept	Concept ID *
Piston and Rigid Container	Chain-link	PR-CL
	Wire and Pulley	PR-WP
	Conveyor Belt	PR-CB
Spring-loaded Rigid Container	Chain-link	SR-CL
	Wire and Pulley	SR-WP
	Conveyor Belt	SR-CB
Roll	Rollers	R-R

* IDs used for conciseness due to a large number of concepts.

and shown in Figure 14.

RCU Deployment Module Evaluation

Several key observations can be made from the metric evaluation presented in

Table 16. Metric evaluation values for RCU deployment module (# marks number).

ID	Metric	PR-CL	PR-WP	PR-CB	SR-CL	SR-WP	SR-CB	R-R
C-D-1	# of interfaces	3	3	3	3	3	3	3
C-D-2	failures per million operating hours	25.43	140.93	140.93	116	231.5	231.5	9
C-D-3	# of points	4	4	4	3	3	3	3
C-D-4	# motors	3	3	3	2	2	2	2
C-D-5	kilograms	39	39	45	24	24	30	44
C-D-6	# of moving parts	5	5	5	5	5	5	4
C-D-7	Nominal scale (very slow to very fast)	Medium	Medium	Slow	Fast	Fast	Medium	Very Fast
C-D-8	meters cubed	0.0788	0.0788	0.0788	0.0338	0.0338	0.0338	0.126

. Failure rate estimations account for motors and additional mechanical components such as conveyors and springs, referencing data from the NWSC-11 handbook [143]. Among all configurations, the WP and CB designs exhibit the highest estimated failure rates. The SR variants demonstrate the lowest overall mass; however, this is primarily attributed to their reduced RCU volume capacity. In terms of deployment speed, PR configurations are generally slower, due to piston loading times, whereas the R-R configuration enables continuous feeding, making it the fastest. The R-R option also provides the highest RCU storage volume, even when constrained to similar external dimensions as rigid container configurations. The quantitative estimations of the criteria are directly used to produce the RP pairwise rankings for all candidates relative to each criterion, as detailed in Section 5.2. An example of the RP pairwise rankings for C-D-5 (mass) is shown in

Table 17. RP matrix for RCU deployment module with respect to C-D-5 (Mass).

Candidate	PR-CL	PR-WP	PR-CB	SR-CL	SR-WP	SR-CB	R-R	RP (${\mathit{f}}_{\mathit{k}}^{\mathit{i}}$)
PR-CL	1	1	3	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	2	0.0913
PR-WP	1	1	3	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	2	0.0913
PR-CB	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	1	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$9$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$9$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$7$}\right.$	1	0.0316
SR-CL	3	3	9	1	1	1	7	0.264
SR-WP	3	3	9	1	1	1	7	0.264
SR-CB	2	2	7	1	1	1	5	0.218
R-R	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	1	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$7$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$7$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$5$}\right.$	1	0.0401
CR	0.005 ($\ll 0.1$)

The term $f_k^i$ is explained in Section 5.

.

A summary of the RI of criteria, the RP of the candidate solutions, and the final decision values is shown in

Table 18. AHP decision matrix for RCU deployment module.

		RP ($f_k^i$)
ID	RI ($t_i$)	PR-CL	PR-WP	PR-CB	SR-CL	SR-WP	SR-CB	R-R
C-D-1	0.125	0.143	0.143	0.143	0.143	0.143	0.143	0.143
C-D-2	0.159	0.263	0.131	0.131	0.140	0.0313	0.0313	0.272
C-D-3	0.168	0.0256	0.0256	0.0256	0.231	0.231	0.231	0.231
C-D-4	0.120	0.0256	0.0256	0.0256	0.231	0.231	0.231	0.231
C-D-5	0.118	0.0913	0.0913	0.0316	0.264	0.264	0.218	0.0401
C-D-6	0.127	0.0667	0.0667	0.0667	0.0667	0.0667	0.0667	0.6
C-D-7	0.0816	0.109	0.109	0.0294	0.210	0.210	0.109	0.223
C-D-8	0.101	0.181	0.181	0.181	0.0368	0.0368	0.0368	0.346
Decision Values (Dk)		0.114	0.0926	0.0790	0.167	0.150	0.136	0.262

The terms $t_i$, $f_k^i$, and $D_k$ are explained in Section 5.

, indicating a strong preference for the R-R deployment configuration, followed by the SR-CL configuration. The R-R configuration suffers from a higher mass requirement, due to the assumption that an aluminum shell protects the RCU bag roll, but performs well in all other criteria.

6.3. Interface with and Secure RCU

The RCU manipulation subsystem consists of seven functional modules (see Figure 8). A primary morphological assessment of the RCU Manipulation subsystem is with regard to the mechanism for interfacing and securing the RCU. The evaluation criteria for the interface module are shown in

Table 19. Evaluation criteria for the RCU interface module (# marks number).

ID	Criteria	Description	Metric	Goal	RI (${\mathit{t}}_{\mathit{i}}$)
C-S-1	Exposure to dust	The number of interfaces between moving parts, which are exposed to the regolith.	# of interfaces	Min	0.104
C-S-2	Dust production	The potential for dust production due to agitation, sharp movements, etc.	ordinal scale (Likert 1–5)	Min	0.0894
C-S-3	Longevity	The anticipated failure rate of the module based on motor load [143].	failures per million operating hours	Min	0.117
C-S-4	Reliability	The number of independent points of single failure of the module.	# of points	Min	0.127
C-S-5	Power (Active/Passive)	Whether the interface mechanism requires active power to grip the RCU.	Yes/No	No	0.0917
C-S-6	Mass	The anticipated mass of the module based on reference designs.	kilograms	Min	0.0873
C-S-7	Complexity	The module complexity is measured as the total number of moving parts. This includes motors, gears, vibrating surfaces, and other parts moving in the workspace.	# of moving parts	Min	0.100
C-S-8	Attachment force	The force that can be enacted by the interface on the RCU.	Newtons	Max	0.0855
C-S-9	Rigid attachment	Whether the interface rigidly attaches to the RCU. Rigid attachment allows for easier manipulation of the RCU.	Yes/No	Yes	0.0850
C-S-10	Attachment points	How many individual points of attachment the interface has on the RCU.	# of attachment points	Max	0.0594
C-S-11	Attachment speed	How quickly the interface module can move into a secured configuration.	Nominal scale (very slow to very fast)	Max	0.0533

The term $t_i$ is explained in Section 5.

, along with the normalized RI of the criteria, calculated as described in Section 5.2.

Although the RCUs will be non-rigid, the degree of compliance is unknown. However, it is anticipated that due to the high compacting behaviour of the lunar regolith, RCUs will not experience a high degree of deflection (semi-rigid). Some reviews of robotic grippers are conducted in [167,168,169]. A review of automated assembly for non-rigid objects is also presented in [170]. Another domain of interest for lunar surface manipulation is that of compliant grippers (see reviews in [171,172]), particularly with regard to mitigating the lunar dust [118]. The definition of the manipulation task as a pick-and-place operation can also be used to constrain the gripping problem, with similar operations being done in industrial settings [173]. The concepts for the interface module, derived from common gripper configurations, are illustrated in

Table 20. RCU interface mechanism candidate solutions.

	Concept	Examples
Hook		[174,175,176,177]
Gecko		[178,179,180]
Compliant Gripper		[118,181,182,183]
Rigid Gripper		[184,185,186]
Magnetic		[187,188,189]
Coupling		[175,190,191]
Flatbed		[192,193,194,195]

, including some specific examples from literature and industry.

RCU Interface Module Evaluation

Some key observations can be drawn from the evaluation process, summarized in

Table 21. Metric evaluation values for RCU interface module (# marks number).

ID	Metric	Hook	Gecko	Compliant Gripper	Rigid Gripper	Magnetic	Coupling	Flatbed
C-S-1	# of interfaces	1	1	0	2	0	3	0
C-S-2	ordinal scale (Likert 1–5)	2	1	3	3	1	2	5
C-S-3	failures per million operating hours	3.5	3.5	3.5	3.5	0	3.5	0
C-S-4	# of points	2	2	2	2	1	1	1
C-S-5	Yes/No	Yes	Yes	Yes	Yes	Yes	Yes	No
C-S-6	kilograms	0.045	0.05	5	8	1.9	1	3
C-S-7	# of moving parts	2	2	2	3	0	4	1
C-S-8	Newtons	216	6.5	100	1000	300	1000	216
C-S-9	Yes/No	No	No	No	No	Yes	Yes	No
C-S-10	# of attachment points	2	∞	∞	∞	2	2	4
C-S-11	Nominal scale (very slow to very fast)	Very Slow	Very Slow	Fast	Fast	Very Fast	Slow	Slow

. Only the Compliant and Magnetic grippers are immune to jamming caused by dust exposure at moving interfaces. The Flatbed configuration is particularly prone to generating dust due to the scooping motion it employs, which disturbs the surrounding regolith. The longevity estimates only consider the operational lifespan of motors, and the Flatbed is the only design that relies on a non-powered mechanical interface. The attachment force calculations depend on the friction coefficient between Kevlar, used as an RCU material analogue, and aluminum (structural frame material), estimated to be 0.18 [196]. Amongst all options, the Gecko gripper demonstrates the lowest attachment force, reflecting its status as a relatively new and less mature technology. Only the Magnetic and Coupling configurations provide rigid or quasi-rigid attachment interfaces. The Gecko, Compliant and Rigid grippers have effectively infinite attachment points, as they can interface with the RCU at nearly any location; this was cut-off at a value of 10 for comparison. In terms of attachment speed, the Hook and Gecko designs are the slowest, due to their deliberate attachment process, whereas the Magnetic gripper enables instantaneous engagement. The quantitative estimations of the criteria are directly used to produce the RP pairwise rankings for all candidates relative to each criterion, as detailed in Section 5.2. An example of the RP pairwise rankings for C-S-6 (mass) is shown in

Table 22. RP matrix for RCU interface module with respect to C-S-6 (mass).

Candidate	Hook	Gecko	Compliant Gripper	Rigid Gripper	Magnetic	Coupling	Flatbed	RP (${\mathit{f}}_{\mathit{k}}^{\mathit{i}}$)
Hook	1	1	2	9	1	1	1	0.182
Gecko	1	1	2	9	1	1	1	0.182
Compliant Gripper	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	1	4	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	0.0891
Rigid Gripper	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$9$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$9$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$4$}\right.$	1	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$7$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$8$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$6$}\right.$	0.0230
Magnetic	1	1	2	7	1	1	1	0.175
Coupling	1	1	2	8	1	1	1	0.179
Flatbed	1	1	2	6	1	1	1	0.172
CR	0.02 ($\ll 0.1$)

The term $f_k^i$ is explained in Section 5.

.

A summary of the RI of criteria, the RP of the candidate solutions, and the final decision values is shown in

Table 23. AHP decision matrix for interface with and secure RCU module.

		RP ($f_k^i$)
ID	RI ($t_i$)	Hook	Gecko	Compliant Gripper	Rigid Gripper	Magnetic	Coupling	Flatbed
C-S-1	0.104	0.172	0.172	0.182	0.0891	0.182	0.0232	0.182
C-S-2	0.0894	0.161	0.202	0.125	0.125	0.202	0.161	0.0231
C-S-3	0.117	0.0435	0.0435	0.0435	0.0435	0.391	0.0435	0.391
C-S-4	0.127	0.0323	0.0323	0.0323	0.0323	0.290	0.290	0.290
C-S-5	0.0917	0.0667	0.0667	0.0667	0.0667	0.0667	0.0667	0.6
C-S-6	0.0873	0.182	0.182	0.0891	0.0230	0.175	0.179	0.172
C-S-7	0.100	0.151	0.151	0.151	0.0805	0.261	0.0284	0.177
C-S-8	0.0855	0.103	0.0330	0.0567	0.301	0.103	0.301	0.103
C-S-9	0.0850	0.0435	0.0435	0.0435	0.0435	0.391	0.391	0.0435
C-S-10	0.0594	0.0303	0.273	0.273	0.273	0.0303	0.0303	0.0909
C-S-11	0.0533	0.0326	0.0326	0.227	0.227	0.275	0.102	0.102
Decision Values (Dk)		0.0945	0.107	0.105	0.104	0.226	0.150	0.215

The term t_i, $f_k^i$ and D_k are explained in Section 5.

, indicating a preference for the Magnetic interface, followed by the Flatbed. The preference for these can largely be attributed to the simplicity of both designs, with the fewest moving parts and no motors.

6.4. Transport RCU to Desired Pose

Another primary morphological assessment of the RCU Manipulation subsystem is with regard to the functional module for transporting the RCUs to the desired pose. The evaluation criteria for the manipulation of RCUs are shown in

Table 24. Evaluation criteria for the RCU transport module (# marks number).

ID	Criteria	Description	Metric	Goal	RI (${\mathit{t}}_{\mathit{i}}$)
C-T-1	Exposure to dust	The number of interfaces between moving parts, which are exposed to the regolith.	# of interfaces	Min	0.0982
C-T-2	Dust production	The potential for dust production due to agitation, sharp movements, etc.	ordinal scale (Likert 1–5)	Min	0.0775
C-T-3	Longevity	The anticipated failure rate of the module based on motor load [143].	failures per million operating hours	Min	0.105
C-T-4	Reliability	A combination of the mechanical and task reliability. Mechanical reliability refers to the likelihood of single-point failure and task reliability considers the risk of task failure when faced with imperfect setups, such as non-optimally positioned RCUs.	ordinal scale (Likert 1–5)	Max	0.113
C-T-5	Power	The relative power consumption as proxied by the number of gravity-loaded joints.	# of gravity-loaded joints	Min	0.0815
C-T-6	Mass	The anticipated mass of the module; estimated by reference designs.	kilograms	Min	0.0759
C-T-7	Complexity	The module complexity is measured as the total number of moving parts. This includes motors, gears, vibrating surfaces, and other parts moving in the workspace.	# of moving parts	Min	0.0859
C-T-8	Workspace—module volume ratio	A measure of the reachable workspace given a specific module configuration.	unitless	Max	0.0822
C-T-9	Manipulation accuracy	A measure of the error between the desired RCU pose and the actual achieved RCU pose.	ordinal scale (Likert 1–5)	Min	0.0786
C-T-10	Manipulation resolution	The smallest measurable distance or increment that a robot can physically produce.	meters	Min	0.0660
C-T-11	Dexterity	A measure of the degrees of freedom, the degree of articulation, and the ability of the module to enter a variety of configurations within its workspace.	ordinal scale (Likert 1–5)	Max	0.0680
C-T-12	Structural rigidity	The resistance to deflections caused by external forces or torques.	Nominal scale (very stiff to very flexible)	Min	0.0690

The term $t_i$ is explained in Section 5.

, along with the normalized RI of the criteria, calculated as described in Section 5.2. It is important to note that the selected manipulation morphology will be paired with a mobile base, which will also expand its effective workspace volume.

There is a number of manipulation mechanisms that could perform the required functions for the RCU transport module (see Figure 15), which can be divided into planetary space and Earth applications. One planetary robotic manipulator is Skymaker by MDA, an articulated 4-Degree-of-Freedom (DoF) manipulator being developed for the lunar surface [197], similar to NASA’s COLDArm [54]. Another higher DoF manipulator is the Perseverance robotic arm, which is an articulated 5-DoF manipulator, primarily for scientific purposes, that has been deployed to Mars [198,199,200]. A different approach to the RCU transport task can be taken by NASA’s LSMS lunar cranes (see Section 2.3), specifically by the L35 crane that has payload and reach capabilities within a reasonable range for RCU manipulation [201]. There are a few terrestrial alternatives that can provide additional morphologies, particularly when considering the task as a pick-and-place problem. One manipulator configuration optimized for such tasks in industrial settings is the Selective Compliance Assembly Robot Arm (SCARA). The Yamaha YK1200 series [202] is a family of terrestrial SCARA manipulators with payload specifications matching the requirements. A final alternative morphology is an automated forklift system, similar to those produced by Rockwell Automation, Agilox, Hyster, or Toyota [192,193,194,195,203].

RCU Transport Module Evaluation

Several notable insights emerge from the evaluation process, summarized in

Table 25. Metric evaluation values for RCU transport module (# marks number).

ID	Metric	4-DoF	5-DoF	SCARA	Forklift	Crane
C-T-1	# of interfaces	4	5	3	2	4
C-T-2	ordinal scale (Likert 1–5)	2	2	2	4	3
C-T-3	failures per million operating hours	14	17.5	14	7	14
C-T-4	ordinal scale (Likert 1–5)	5	4	3	2	1
C-T-5	# of gravity-loaded joints	3	3	1	1	2
C-T-6	kilograms	40	76	96	20	11
C-T-7	# of moving parts	4	5	4	3	4
C-T-8	unitless	900	294	41	5	900
C-T-9	ordinal scale (Likert 1–5)	2	2	1	5	3
C-T-10	meters	1	1	1	1	1
C-T-11	ordinal scale (Likert 1–5)	4	5	3	1	1
C-T-12	Nominal scale (very stiff to very flexible)	Medium	Medium	Stiff	Stiff	Very Flexible

. The Forklift design is the most susceptible to dust production due to direct ground contact during RCU pickup and drop-off operations. For longevity, only motor components were considered, naturally favouring configurations with fewer motors. The SCARA manipulator, despite having additional degrees of freedom, features only one gravity-loaded joint, making its power consumption equivalent to that of the Forklift. In terms of workspace efficiency, the 4-DoF manipulator and the Crane perform best relative to module volume, whereas the Forklift performs worst, due to its constrained single-axis movement. The Forklift also scores lowest in manipulation accuracy, primarily relying on coarse x–y positioning from its mobile base. The Crane similarly suffers from reduced accuracy due to potential sway in its cable system. Manipulation precision was considered equal for all candidates, as accurate estimation would require detailed specifications of actuation and sensing hardware. Finally, the Crane is deemed the most flexible, due to its non-rigid cable architecture, followed by the 4-DoF and 5-DoF manipulators whose slender limbs enable some structural deflection. The SCARA and Forklift designs are comparatively more rigid. The quantitative estimations of the criteria are directly used to produce the RP pairwise rankings for all candidates relative to each criterion, as detailed in Section 5.2. An example of the RP pairwise rankings for C-T-6 (mass) is shown in

Table 26. RP matrix for transport RCU to desired pose module for C-T-6 (mass).

Candidate	4-DoF	5-DoF	SCARA	Forklift	Crane	RP (${\mathit{f}}_{\mathit{k}}^{\mathit{i}}$)
4-DoF	1	2	6	1	1	0.258
5-DoF	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.$	1	3	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	0.109
SCARA	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$6$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$3$}\right.$	1	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$8$}\right.$	$\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$9$}\right.$	0.0372
Forklift	1	3	8	1	1	0.294
Crane	1	3	9	1	1	0.302
CR	0.05 ($\ll 0.1$)

The term $f_k^i$ is explained in Section 5.

.

A summary of the RI of criteria, the RP of the candidate solutions, and the final decision values is shown in

Table 27. AHP Decision Matrix for Transport RCU to Desired Pose Module.

		RP ($f_k^i$)
ID	RI ($t_i$)	4-DoF	5-DoF	SCARA	Forklift	Crane
C-T-1	0.0982	0.161	0.0419	0.305	0.331	0.161
C-T-2	0.0775	0.276	0.276	0.276	0.0301	0.141
C-T-3	0.105	0.189	0.0462	0.189	0.387	0.189
C-T-4	0.113	0.353	0.270	0.219	0.119	0.0399
C-T-5	0.0815	0.0397	0.0397	0.365	0.365	0.190
C-T-6	0.0759	0.258	0.109	0.0372	0.294	0.302
C-T-7	0.0859	0.194	0.0396	0.194	0.379	0.194
C-T-8	0.0822	0.383	0.143	0.0496	0.0424	0.383
C-T-9	0.0786	0.239	0.239	0.291	0.0345	0.196
C-T-10	0.0660	0.200	0.200	0.200	0.200	0.200
C-T-11	0.0680	0.382	0.302	0.230	0.0433	0.0433
C-T-12	0.0690	0.232	0.232	0.252	0.252	0.0331
Decision Values (Dk)		0.236	0.161	0.218	0.213	0.171

The terms $t_i$, $f_k^i$, and $D_k$ are explained in Section 5.

, indicating a preference for the 4-DoF morphology, closely followed by the SCARA morphology.

6.5. Final LRCS Morphology Selection

Now that the candidate solutions for the key functional modules have been evaluated, it is possible to assemble a high-level morphology for the LRCS. It is worth noting that the best morphology may not necessarily be the combination of the best performing candidates during evaluation, since not all candidates can be physically combined across the various modules, e.g., the flatbed interface and the crane transport module. The decision value ${\left(D_k\right)}_i$ for each functional module i is scaled to account for the spread in the decision value:

$${\left(D_k^{\prime }\right)}_i={\displaystyle \frac{N_i}{N}}{\left(D_k\right)}_i;i=1,\dots ,F;k=1,\dots ,N_i,$$

(61)

where $N_i$ is the total number of candidate solutions for module i, F is the total number of functional modules evaluated ($F=4$ in this case), and N is the total number of candidate solutions across all modules. This effectively transforms ${\left(D_k\right)}_i$ values into a common scale ${\left(D_k^{\prime }\right)}_i$, such that they can be summed up without giving additional preference to modules with fewer candidate solutions. The combination of all functional modules will then form a set of $\lambda ={\prod }_{i=1}^F{N}_i$ solutions. The decision values of such solutions are $D^{\prime }=\{D_1^{\prime },D_2^{\prime },\dots ,D_{\lambda }^{\prime }\}$, where

$$D_j^{\prime }=\sum ({\left(D_{k_j}^{\prime }\right)}_1,{\left(D_{k_j}^{\prime }\right)}_2,\dots ,{\left(D_{k_j}^{\prime }\right)}_F);j=1,\dots ,\lambda .$$

(62)

For the LRCS evaluation studied in this paper, $F=4$ and $\lambda =5\times 7\times 7\times 5=1225$ module combinations, resulting in the top 10 compatible morphologies listed in

Table 28. Morphological table for four primary functional modules.

Intake Lunar Regolith	Deploy Empty RCU	Interface with and Secure RCU	Transport RCU to Desired Pose	${\mathit{D}}_{\mathit{j}}^{\prime }$
Vibrating Funnel	R-R	Magnetic	4-DoF	0.249
Vibrating Funnel	R-R	Flatbed	4-DoF	0.246
Vibrating Funnel	R-R	Magnetic	SCARA	0.245
Vibrating Funnel	R-R	Magnetic	Forklift	0.244
Vibrating Funnel	R-R	Flatbed	SCARA	0.242
Vibrating Funnel	R-R	Flatbed	Forklift	0.241
Vibrating Chute	R-R	Magnetic	4-DoF	0.240
Hopper and Auger	R-R	Magnetic	4-DoF	0.237
Vibrating Chute	R-R	Flatbed	4-DoF	0.237
Vibrating Chute	R-R	Magnetic	SCARA	0.236

. The clear favourites for the Intake Lunar Regolith and Deploy Empty RCU functional modules are the Vibrating Funnel and the R-R morphologies, respectively. The top candidates for the Interface with and Secure RCU functional module are the Magnetic and Flatbed morphologies, with a preference for Magnetic. Top preferences for the Transport RCU to Desired Pose functional module include the 4-DoF and SCARA morphologies. Moving forward, the design process will continue with the top-most combined morphology, including the Vibrating Funnel, R-R deployment, Magnetic interface, and 4-DoF manipulator.

7. Selection of Surrounding Architecture

Since the morphology of the primary functional modules has been established, it is now possible to select technologies for the surrounding architecture. This is done methodically by considering the functional modules listed in Figure 8. These functional modules, the selected candidates, the placement, and the reasoning are broken down in

Table 29. LRCS architecture breakdown.

	Function	Solution	Placement	Reasoning
Power Management	Generate power	Solar power	RB	Solar arrays gain a mass advantage over nuclear reactors at polar sites illuminated for over 75% of a lunar cycle [204].
	Store power	Li-ion batteries	RB + MM	Li-ion batteries have a higher TRL than alternatives like Regenerative Fuel Cells (RFCs), and have a mass advantage for short discharge durations under 10–18 h [205,206,207,208].
	Distribute power	Power bus	RB + MM	A custom Power Management and Distribution (PMAD) architecture will need to be designed according to the system’s needs, which can follow standards, such as the AIAA S-122 [209] or the ECSS-E-ST-20-20C [210].
Onboard Computing	Collect and monitor system health information; Process and communicate telemetry data; Memory management; Parse and process incoming commands; Command and Control	CPUs # 1 and # 2, and Communications Bus	RB + MM	Two Central Processing Units (CPUs) will manage all of the Onboard Computing subsystem functionalities on the RB (CPU # 1) and MM (CPU #2). Additionally, an internal communications architecture will need to be implemented based on the hardware, such as MIL-STD-1553B or SpaceWire [211].
RCU Filling	Intake lunar regolith; Fill RCU with lunar regolith	Vibrating Funnel	RB	Previously selected (Section 6.5)
	Beneficiatelunar regolith	Grizzly bars	RB	Has been found to be effective for removing larger rock sizes when used with RASSOR (IPEx predecessor) [90].
	Deploy Empty RCU	R-Rmorphology	RB	Previously selected (Section 6.5)
	Seal RCU	Thermal sealing	RB	Thermal sealing is a common strategy in industrial bagging processes, and allows for fewer moving parts as compared to other methods like sewing or folding. The sealing bar can be integrated into the opening mechanism and rollers of the Deploy Empty RCU R-R configuration [212,213,214].
RCU Transfer	Receive RCU; Internally transfer RCU; Prepare RCU for manipulation	Pallet conveyor	RB	Allows for the accumulation of multiple RCUs and easy low-power dispensing to the MM [215,216].
RCU Manipulation	Interface with and secure RCU; Release RCU	Magnetic interface	MM	Previously selected (Section 6.5)
	Identify desired RCU pose; Calculate trajectory; Identify obstacles; Verify Pose	Sensors and CPU # 2	MM	The MM CPU will need to process and fuse sensor data to construct a map of its surroundings for path planning and localization. This is essential for both the manipulation and MM trajectory generation. The specific sensors necessary for this task require further investigation.
	Transport RCU to desired pose	4-DoF	MM	Previously selected (Section 6.5)
Commun- ications	Process signals	Transceiver	RB + MM	Common signal processing component
	Send and receive communications signals	K/Ka antenna	RB + MM	It is anticipated that lunar communications terminals, such as the RB, would leverage the K/Ka band when communicating with Earth, as well as with lunar surface rovers, such as the MM [217].
Thermal Management	Monitor thermal levels	Thermistors	RB + MM	Common thermal sensor
	Thermal actuation (passive or active); Radiate excess heat; Intake heat	Thermal surfaces	RB + MM	Further thermal analysis is required to determine if or where active thermal actuation is needed, as well as the placement of passive coatings and thermal conducting structures (e.g., thermal straps).
Mobility	Localize system within environment; Obstacle avoidance	Sensors, and CPUs # 1 and # 2	RB + MM	Same functionality as previously described for the sensors and CPU # 2. The RB will also need to perform sensor fusion for simple localization and self-inspection.
	Navigate and move within environment; Deploy from launch vehicle	Differential 4-Wheel Drive and 6-Wheel Rocker-Bogie	RB + MM	The RB will require limited mobility, such as for deploying from the launch vehicle, which can be achieved by a simple differential 4-wheel drive mobility system. The 6-wheel rocker-bogie suspension system has been implemented on the Mars Science Laboratory, Mars Exploration Rovers, Pathfinder, and Perseverance. The latest iteration includes 4 individually steerable front and back wheels, allowing the vehicle to swerve, make arcing turns, and turn 360 degrees in place. This flexibility in steering would enable the MM to enter any necessary configurations for manipulation [198,200].
Lunar Dust Mitigation	Lunar Dust Mitigation	EDS, Surface Coatings, and Seals	RB + MM	The Electrodynamic Dust Shield (EDS) is an active dust mitigation tool with high TRL, which leverages the electrostatic charging of lunar regolith to keep surfaces, such as solar panels, clear from dust [218,219,220,221,222,223,224]. Passive dust mitigation techniques can also be implemented through surface coatings that minimize the dust attraction, such as the lotus leaf coating [144,225,226] and topology modification [144,227,228], as well as seals to prevent dust ingress to critical interfaces, such as the spring-loaded Teflon seal [144,229,230].

. Since the RCU bagging and manipulation tasks are decoupled (see Section 4.2), the placement locations are broken down into the RCU Bagger (RB) and Mobile Manipulator (MM).

8. Conclusions and Future Work

The work has highlighted the importance of robotic systems for construction on the lunar surface, focusing on the conceptualization of a Lunar Robotic Construction System (LRCS) for filling and manipulating Regolith Containment Units (RCUs). The system is intended to intake raw lunar regolith from an excavator, namely, NASA’s ISRU Pilot Excavator (IPEx), to fill the RCUs and build structures, such as berms. A blast berm for a 100m-diameter landing pad is considered for the operational scenario. The LRCS is conceptualized, beginning with a functional analysis and high-level requirements development, identifying the intake, bagging, interface, and transport functional modules to be the integral morphological elements for the evaluation. The Analytical Hierarchy Process (AHP) is used to evaluate a multitude of candidate morphologies. An expert survey is conducted to prioritize the evaluation criteria, finding that factors such as reliability, longevity and dust exposure consistently rank as the highest priorities. The ZM_II aggregation technique is used to aggregate these findings, being incorporated into the AHP in a novel way that reduces expert cognitive load by eliminating pairwise comparisons. The quantitative approach to the AHP, including both ranking aggregation and metric evaluations, demonstrates the potential for a systematic design process based on quantifiable data. Future work can extend this to other design scenarios or explore further systemization. The resulting morphology includes a vibrating hopper for intake, a roller system for container deployment and filling, a magnetic RCU interface, and a 4-DoF manipulator to place the RCUs in the environment. The system is to be decoupled, separating the manipulation and transport tasks from the rest of the functions. Finally, the preliminary selection of the surrounding system architecture is presented, completing the conceptual design. The next phase of this work will include the detailed design of the system, including modeling and detailed power, mechanical and thermal investigations, as well as system validation through simulations. Ideally, the LRCS can be integrated into developing European lunar construction ecosystems, such as PRO-ACT, to enhance the capabilities and versatility of the robotic construction team. Future work should investigate how these robotic systems can meet the requirements for the variety of construction tasks that need to be performed on the lunar surface in the upcoming decades.