Moonbots: Modular Lunar Robotic Systems

• Moonbots are modular lunar robotic systems featuring heterogeneous modules for mobility, manipulation, power, and payload integration under strict lunar constraints.
• They utilize advanced reconfiguration strategies such as graph search and IK-based alignment to adapt to varying tasks and challenging off-world environments.
• The systems integrate distributed software architectures and hybrid control paradigms to ensure reliable operation and scalability in lunar exploration and construction.

Moonbots are lunar robotic systems whose defining themes are modularity, reconfiguration, and task adaptation under off-world constraints. In the most specific and technically developed usage, MoonBot is a modular and on-demand reconfigurable robot engineered for moon base construction, with heterogeneous modules for mobility, manipulation, power, communications, and payload integration under stringent lunar payload mass limits and varying environmental conditions (Uno et al., 26 Dec 2025). Closely related literature extends the concept into distributed software architecture and deployment for the MoonBots platform (Neppel et al., 3 Nov 2025), and into 4-DOF limb modules “which we call Moonbots” that assemble into multiple robot morphologies for lunar exploration and construction (Diaz et al., 8 Jan 2026). The term also appears in broader lunar robotics work for heterogeneous rover fleets, optimized hopping robots, and autonomous excavation systems (Kilic et al., 2021, Kalita et al., 2019).

1. Terminological scope and research usage

The literature does not use “Moonbots” as a single canonical hardware designation. Instead, the label covers several distinct but related lunar robotics concepts, ranging from modular construction robots to rover fleets and hopping systems. This suggests that “Moonbots” functions as a family label for lunar robotic systems rather than a single standardized embodiment.

Usage	Description	Source
MoonBot	Modular and on-demand reconfigurable robot toward moon base construction	(Uno et al., 26 Dec 2025)
MoonBots platform	Distributed heterogeneous modularity, software architecture, and deployment	(Neppel et al., 3 Nov 2025)
Moonbots limbs	4-DOF robot limbs connecting into nine functional configurations	(Diaz et al., 8 Jan 2026)
“Moonbots” solution	Heterogeneous fleet of three four-wheeled rovers for NASA SRC2	(Kilic et al., 2021)
SphereX Moonbot	Hopping robot optimized by AMDCO for extreme lunar exploration	(Kalita et al., 2019)
“Moonbot” networks / lunar excavation robots (“Moonbots”)	Pit-bot cave explorers and off-world excavation teams	(Thangavelautham et al., 2017, Thangavelautham et al., 2020)

A common misconception is that Moonbots are defined solely by self-reconfigurable hardware. The cited work shows a broader scope: physical reconfiguration is central in the construction-oriented platform, but software, communication, deployment, autonomy, and multi-robot coordination are also treated as core aspects of Moonbots research (Neppel et al., 3 Nov 2025).

2. Modular morphology and reconfiguration models

The construction-oriented MoonBot architecture is explicitly heterogeneous. It comprises four module families: Mobility (“Wheel”) modules, Manipulation (“Limb” + “Hand/Gripper”) modules, Power/comms (“Body”) modules, and Payload modules such as an extendable tower and inflatable HIDAS cells. Each 480 mm-diameter Wheel module carries its own BLDC motor, gearbox, battery line, and two grapple fixtures for chainable locomotion. The manipulation subsystem is a 1.55 m long, 7-DOF articulated link, with each joint driven by a 50 W or 80 W BLDC plus harmonic drive, and with a 1-DOF parallel-jaw gripper at each end. Body modules are zero-DOF hubs housing a control board, battery, IMU, Wi‑Fi transceiver, and up to four grapple fixtures. Payload modules attach via the same grapple fixture used by Wheel and Body modules (Uno et al., 26 Dec 2025).

MoonBot assemblies are represented as an undirected graph $G(V,E)$, where $V$ is the set of modules and $E$ is the set of mechanical connections, with global configuration defined by an adjacency matrix $A\in\{0,1\}^{|V|\times|V|}$. Forward kinematics for a limb chain is written as

$$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$

with

$$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$

so that the pose of an end-effector or grapple point is

$p_n=T_{0n}\,p_{n,\mathrm{local}}.$

On-demand reconfiguration is described as a graph-search problem, using BFS or A* on the module graph to select a target morphology, followed by coupling-pair identification via $A$, IK-based alignment, connector closure, and recomputation of each limb’s kinematic tree (Uno et al., 26 Dec 2025).

A parallel Moonbots hardware line develops identical 4-DOF limb modules and wheel modules that self-assemble into manipulators, walkers, and vehicles. These 4-DOF limbs use a Roll–Pitch–Pitch–Roll chain with $d_1=0.05\ \mathrm{m}$, $L_2=0.18\ \mathrm{m}$, $V$0, a gripper offset of approximately $V$1, joint limits of $V$2 for roll joints and $V$3 to $V$4 for pitch joints, and a maximum reach of approximately $V$5. From these modules, nine functional configurations are reported: 4DOF-limb, 8DOF-limb, vehicle, dragon, minimal, quadruped, cargo, cargo-minimal, and bike (Diaz et al., 8 Jan 2026).

3. Connector mechanics, mass budgeting, and lunar environmental adaptation

Connector design is a central technical issue because modularity increases the number of mechanical interfaces and therefore the number of potential failure points. The MoonBot parallel-jaw gripper uses POM sliding and a dust-proof design, with pinch force $V$6 and contact area $V$7. The corresponding connector shear-strength estimate is

$V$8

and a safety factor $V$9–$E$0 is targeted. Alternative interfaces include screw-type and diaphragm-type connectors. For the screw connector, the torque required to engage is

$E$1

while the diaphragm connector has claw engagement force $E$2 per claw and total $E$3 (Uno et al., 26 Dec 2025).

Mass and payload constraints are treated explicitly. For the construction-oriented system, module masses on Earth are $E$4 for a Limb, $E$5 for a Wheel, and $E$6 for a Body, with capacity scaling by $E$7 under lunar gravity $E$8. For an assembly of $E$9 modules,

$A\in\{0,1\}^{|V|\times|V|}$0

Volume packing efficiency for a launch-lock palette is expressed as

$A\in\{0,1\}^{|V|\times|V|}$1

The stated trade-off is direct: adding modules increases functional versatility, manipulability, and wheel count, but also increases mass and the number of connector failure points (Uno et al., 26 Dec 2025).

Environmental adaptation is addressed through materials, sealing, coatings, and thermal analysis. Reported materials include SLA-printed EPX82 resin, milled duralumin limbs, and POM sliding parts, with anti-static coatings to mitigate tribocharging. Dust protection uses snap-fit lids with magnets, minimal clearances, and labyrinth seals at joints. Housing thermal stress under $A\in\{0,1\}^{|V|\times|V|}$2 swings is estimated by

$A\in\{0,1\}^{|V|\times|V|}$3

with $A\in\{0,1\}^{|V|\times|V|}$4 and $A\in\{0,1\}^{|V|\times|V|}$5, giving $A\in\{0,1\}^{|V|\times|V|}$6, described as well below yield. Electronics are selected for high TID tolerance, while future flight units are to include shielding analysis (Uno et al., 26 Dec 2025).

The 4-DOF Moonbots limb paper reports a more compact per-module budget: actuator including Harmonic Drive at approximately $A\in\{0,1\}^{|V|\times|V|}$7, links plus gripper at approximately $A\in\{0,1\}^{|V|\times|V|}$8, and control box with LattePanda 3, batteries, and converters at approximately $A\in\{0,1\}^{|V|\times|V|}$9, for a total of approximately $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$0. Its environmental constraints are stated as vacuum, wide thermal range from $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$1 to $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$2, lunar regolith dust, radiation exposure, and low gravity requiring re-tuned control gains (Diaz et al., 8 Jan 2026).

4. Power, control, and distributed software architecture

MoonBot power architecture is split across module-local resources. Each Limb has two battery lines: $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$3 for the on-board computer and sensors, and $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$4 for actuators. With nominal capacity $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$5 per line, the energy is approximately $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$6 per battery. The reported power model is

$$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$7

where each joint term is

$$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$8

Typical cruise power is approximately $$T_{0n}=T_1(\theta_1)\,T_2(\theta_2)\,\cdots\,T_n(\theta_n),$$9 per limb. Control is described as hybrid distributed-centralized: distributed low-level ROS1/CANopen joint nodes run on each module computer, specifically LATTEPANDA Alpha, while a centralized high-level planner and Motion Stack at Levels 4–5 run on a mission control PC. Communication is via a Wi‑Fi mesh, and v1 prototypes do not pass wired power or data through mechanical connectors. A significant operational point is that all tests to date are human-in-the-loop teleoperation; supervised autonomy and onboard vision-based docking are explicitly part of the development roadmap (Uno et al., 26 Dec 2025).

The MoonBots software architecture generalizes modularity beyond mechanics to software, communication, and orchestration. Each module runs components structured into seven sub-blocks: Core, Injection, Override, API, Interface, Executor, and Comm. Heterogeneity is managed by swapping Injection and Override classes at launch, so that, for example, H-line V1 and V2 modules differ only by parameters passed into their Override blocks. The communication model is purely data-centric: sensors, actuators, and planners read or write resources via ROS 2 topics layered over Zenoh. DDS is described as incurring $$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$0 startup overhead because every node discovers every other peer, whereas Zenoh inserts lightweight routers in a hybrid P2P/brokered mesh to reduce discovery and recovery times. Deployment is automated by a four-stage Python orchestrator covering update, build, assemble, and launch, with total deployment time

$$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$1

At the center of this stack is the open-source Motion Stack, whose named components are JointManager, KinematicsManager, TrajectoryPlanner, Controller, and a Python API (Neppel et al., 3 Nov 2025).

Dynamic reconfiguration in the MoonBots platform is handled by a module-local state machine, INIT → WAIT_FOR_CONFIG → CONFIGURED → RUNNING → RECONFIGURE → INIT, with modules discovering neighbors by subscribing to a shared configuration topic and recomputing kinematic graphs when a reconfigure message arrives. Reliability is written as

$$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$2

which formalizes the claim that redundant limbs or wheels increase system reliability without code changes (Neppel et al., 3 Nov 2025).

The 4-DOF Moonbots limb line adopts a different but related control stack. Communication is based on CAN bus daisy-chained among ODrive motor drivers, a USB–CAN bridge to a LattePanda SBC, and an 802.11ac wireless link to ground control. Reported control modes are current mode with an 8 kHz FOC inner loop, velocity mode with a 4 kHz PID-velocity loop, and position mode with a 1 kHz PID-position loop plus trapezoidal trajectory generation. Joint-level control uses

$$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$3

and Cartesian IK uses damped least squares,

$$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$4

Its limb-to-limb handshake procedure requires pre-alignment via IK and connector closure only when $$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$5 and $$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$6 (Diaz et al., 8 Jan 2026).

5. Demonstrations, performance metrics, and engineering lessons

Field demonstrations for MoonBot were organized around self-assembly, locomotion, civil engineering, infrastructure deployment, and assistive operations with inflatable modules. On-palette self-assembly of a 7-DOF limb built from two 3-DOF modules plus one 1-DOF module achieved average alignment error $$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$7 and $$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$8. A minimal autonomous pick-and-place sequence using one Limb and one Wheel reported pose error along $$T_i(\theta_i)= \begin{bmatrix} R_i(\theta_i) & d_i \ 0 & 1 \end{bmatrix},$$9 up to $p_n=T_{0n}\,p_{n,\mathrm{local}}.$0 and $p_n=T_{0n}\,p_{n,\mathrm{local}}.$1, attributed to calibration drift. Locomotion tests on $p_n=T_{0n}\,p_{n,\mathrm{local}}.$2 deep silica sand showed that a minimal configuration could climb slopes up to $p_n=T_{0n}\,p_{n,\mathrm{local}}.$3 with in-situ tail-limb press and stalled beyond, while a vehicle with one Limb and two Wheels exceeded $p_n=T_{0n}\,p_{n,\mathrm{local}}.$4, used skid-steer control, and traveled at approximately $p_n=T_{0n}\,p_{n,\mathrm{local}}.$5. A multicycle formed from three Minimals in parallel demonstrated enhanced redundancy, supported more than $p_n=T_{0n}\,p_{n,\mathrm{local}}.$6 load under lunar scaling, and had zero immobilization events in tests (Uno et al., 26 Dec 2025).

Task	Reported result	Value
Self-assembly	Average alignment error	$p_n=T_{0n}\,p_{n,\mathrm{local}}.$7, $p_n=T_{0n}\,p_{n,\mathrm{local}}.$8
Vehicle locomotion	Slope capability and speed	$p_n=T_{0n}\,p_{n,\mathrm{local}}.$9, $A$0
Rock clearing	Dragon picks rocks; Minimal drags sled	$A$1 rocks, $A$2
Raking leveling	Force and flatness	$A$3, $A$4 over $A$5
Tower transport	Drag distance and duration	$A$6 tower over $A$7 in $A$8
Solar array deployment	Upright orientation accuracy	$A$9 with respect to vertical
HIDAS assistance	Stopper placement repeatability	$d_1=0.05\ \mathrm{m}$0

Civil-engineering tasks included rock clearing, where the Dragon configuration picked approximately $d_1=0.05\ \mathrm{m}$1 rocks and loaded a sled, and a Minimal dragged the sled at $d_1=0.05\ \mathrm{m}$2. In raking and leveling, the reported rake force was approximately $d_1=0.05\ \mathrm{m}$3 and surface flatness reached $d_1=0.05\ \mathrm{m}$4 over a $d_1=0.05\ \mathrm{m}$5 patch. For infrastructural hardware transport, a Dragon dragged a $d_1=0.05\ \mathrm{m}$6 extendible tower over $d_1=0.05\ \mathrm{m}$7 in approximately $d_1=0.05\ \mathrm{m}$8, with terrain traction force $d_1=0.05\ \mathrm{m}$9 measured by a load cell on the sled hitch. For panel-type solar arrays, a Minimal delivered the panel and a Dragon oriented it upright to within $L_2=0.18\ \mathrm{m}$0 of vertical. During inflatable HIDAS assistance, external force $L_2=0.18\ \mathrm{m}$1 produced $L_2=0.18\ \mathrm{m}$2 rise, confirming seal integrity, and stopper insertion achieved $L_2=0.18\ \mathrm{m}$3 positional repeatability. Hardware lessons included failed-limb swap in less than $L_2=0.18\ \mathrm{m}$4, dust ingress into harmonic drives, lower stiffness for the genderless diaphragm connector, and the planned addition of AprilTags plus electrical power/data pass-through in the gripper interface. Software lessons included the robustness of a clamped-integral remote controller against communication dropouts, the value of homogeneous low-level code, and the major benefit of a third-person external camera for operator accuracy (Uno et al., 26 Dec 2025).

Software field deployment results complement the hardware metrics. The MoonBots platform was validated over nine weeks at JAXA’s sand field, five days in DLR’s LUNA simulator, and six days at Osaka 2025 Expo. On seven robots, DDS versus Zenoh performance was reported as follows: startup latency $L_2=0.18\ \mathrm{m}$5–$L_2=0.18\ \mathrm{m}$6 versus $L_2=0.18\ \mathrm{m}$7, startup bandwidth $L_2=0.18\ \mathrm{m}$8 versus $L_2=0.18\ \mathrm{m}$9, stutter duration $V$00 versus $V$01, average latency $V$02 versus $V$03, and max stable robots $V$04 versus $V$05. Deployment timings were reported as $V$06–$V$07 to add a new robot family, $V$08 to swap an entire assembly, $V$09 for full software install, $V$10 to start a remote robot, and $V$11 to onboard a new developer through the Python API (Neppel et al., 3 Nov 2025).

The modular limbs paper adds component-level control characterization. Static load tests on a single actuator reported average current and maximum position drift of $V$12 and less than $V$13 rev at $V$14, $V$15 and less than $V$16 rev at $V$17, $V$18 and less than $V$19 rev at $V$20, and $V$21 with less than $V$22 rev at $V$23. Position step response showed rise time of approximately $V$24, settling time of approximately $V$25, and overshoot below $V$26 in trapezoidal mode. In multi-joint tests, synchronous steps to all four joints produced no loss of sync, a Cartesian step of $V$27 along $V$28 had maximum tracking error of approximately $V$29 and settling below $V$30, and gripper joints maintained connection under $V$31 pull tests (Diaz et al., 8 Jan 2026).

6. Broader lunar robotics context and future directions

Outside the modular construction lineage, “Moonbots” also denotes heterogeneous rover fleets. In the NASA Space Robotics Challenge 2 qualification round, the “Moonbots” solution consisted of three four-wheeled rovers—Scout, Excavator, and Hauler—with a common drive base and task-specific hardware. The autonomy stack used nested SMACH state machines in ROS, a two-stage EKF for fusing visual odometry, wheel odometry, and IMU, a global planner based on base_global_planner, and a local planner based on Dynamic Window Approach. Reported results included localization drift greater than $V$32 without homing and less than $V$33 with $V$34–$V$35 homing updates, Scout sensing approximately $V$36–$V$37 volatiles per run and scoring $V$38–$V$39, average collection of $V$40–$V$41 volatiles per $V$42, mean absolute CubeSat localization error of approximately $V$43 in $V$44, Scout sub-$V$45 reporting accuracy after each homing event, collection of $V$46–$V$47 of available volatiles before timeout, and CubeSat detection plus alignment robust in $V$48 of qualification-round seeds (Kilic et al., 2021).

Other Moonbot formulations emphasize ballistic or propulsive mobility for extreme terrain. The optimized SphereX Moonbot is a hopping robot designed through Automated Multidisciplinary Design and Control Optimization with $V$49 design variables and four objectives spanning mass, size, payload fraction, power draw, and residual power fraction. For a surface mission, the average optimized robot mass is approximately $V$50 with radius approximately $V$51 and available payload approximately $V$52; in a pit scenario the mass rises to approximately $V$53. Propulsion is a miniaturized RP‑1/H$V$54O$V$55 bipropellant thruster, with $V$56, approximately $V$57 or approximately $V$58 per average hop, roughly $V$59 hops from a $V$60 Li-ion battery and roughly $V$61 hops from a $V$62 fuel-cell pack, and reported mass reductions of $V$63–$V$64, power-budget savings of $V$65–$V$66, and approximately $V$67 increased payload fraction relative to a manually designed baseline (Kalita et al., 2019). A related pit-bot concept describes networks of $V$68, $V$69 diameter ball robots carrying $V$70 payloads, with $V$71, range approximately $V$72, and aggregate flight time approximately $V$73 on the Moon, supported by stereo camera, laser rangefinder, and cooperative triangulation for mapping cave interiors (Thangavelautham et al., 2017).

Moonbots research also overlaps with excavation, site preparation, and cooperative load transport. For off-world open-pit mining, an engineer’s reference explicitly frames lunar excavation robots as “Moonbots” and uses the Artificial Neural Tissue architecture with population size $V$74, evaluation horizon $V$75 timesteps, $V$76 randomized trials, and team sizes $V$77. Throughput peaks at $V$78, giving an optimal density of approximately $V$79 robots per workload block, while phototactic beacons increase single-robot fitness from $V$80 to $V$81 and improve four-robot performance under tight time constraints from $V$82 to $V$83; bucket-brigade incidence rises from $V$84 at $V$85 to $V$86 at $V$87 (Thangavelautham et al., 2020). Related lunar logistics work on teams of climbing robots analyzes $V$88 tethered spherical climbers, each of mass approximately $V$89 and diameter approximately $V$90, hauling payloads of $V$91–$V$92 safely on $V$93–$V$94 slopes under lunar gravity, with tether dynamics, microspine adhesion, and LBKPIECE path planning over steep crater terrain (Kalita et al., 2018).

Future directions for the modular MoonBot lineage are explicit and operational. The reported roadmap includes thermal-vacuum, vibration, and radiation qualification on a flight model; vision-based docking and force-tactile hybrid controllers for fully autonomous reconfiguration; scaling to fleets of more than $V$95 modules; and coordinated construction of a small lunar-hut analogue (Uno et al., 26 Dec 2025). The 4-DOF Moonbots line adds autonomous reconfiguration via vision and force sensing, dust- and radiation-hardened connectors and sealing, an integrated power/data bus across modules, reinforcement learning for adaptive gait and manipulation strategies, and miniaturized IMU and LiDAR for SLAM and balance control (Diaz et al., 8 Jan 2026). Taken together, these directions place Moonbots at the intersection of modular robotics, distributed systems, off-world construction, and multi-robot lunar operations.