MoonBot: Modular Lunar Robotics

• MoonBot is a modular robotic system designed for lunar construction, featuring heterogeneous hardware modules and on-demand reconfiguration to meet strict payload constraints.
• It integrates specialized modules like limbs, wheels, body, and grippers with a distributed software architecture that supports autonomous assembly, teleoperation, and multi-robot cooperation.
• Field demonstrations validate its adaptive locomotion, robust connector design, and effective integration of excavation tools, advancing lunar in-situ resource utilization and infrastructure development.

MoonBot is a modular and on-demand reconfigurable robotic system for lunar surface operations and moon base construction, engineered to maximize functionality while operating within the stringent mass constraints of lunar payloads and adapting to varying environmental conditions and task requirements. The platform combines heterogeneous functional modularity in hardware with what its software architecture paper terms distributed, heterogeneous modularity in software, communication, and deployment. Across the current literature, MoonBot is presented as a self-assembling and reconfigurable system for civil engineering operations, infrastructural component transportation and deployment, assistive operations with inflatable modules, autonomous alignment during assembly, and compatibility with a modular bucket drum excavator for lunar ISRU (Uno et al., 26 Dec 2025, Neppel et al., 3 Nov 2025).

1. System concept and module taxonomy

MoonBot’s hardware design is organized around specialized modules rather than strictly identical units. The design philosophy is described as heterogeneous functional modularity: modules are specialized, optimizing the trade-off between flexibility, performance, and operational simplicity. The core modules are a Limb module, a Wheel module, a Body module, and a Hand module (gripper) (Uno et al., 26 Dec 2025).

Module	Description	Stated role
Limb module	Symmetrical 7-DOF articulated arm with a gripper at each end	Can serve as an arm or leg
Wheel module	Dual-motor, independently driven base	High-speed locomotion
Body module	Central hub with battery and computer	Grapple fixtures for module interconnection
Hand module	1-DOF parallel jaw	Object grasping and connecting modules

Each module is described as self-contained, with power, compute, and sensors. Modules are transported as compact units and assembled in situ on the lunar surface. The resulting morphologies are task-specific. A Minimal configuration consists of 1 Limb + 1 Wheel; a Vehicle configuration uses 1 Limb + 2 Wheels; a Dragon configuration joins 2 Minimals in series to improve traction and load capacity; and a Multicycle configuration places multiple Minimals in parallel for heavier payloads and redundancy (Uno et al., 26 Dec 2025).

This organization is explicitly tied to lunar mission constraints. The modular architecture allows multiple task profiles to be fulfilled with a single stowed kit of modules, while heterogeneous modules place multi-DOF capability only where useful. The papers therefore distinguish MoonBot from purely clone-modular or single-purpose rover concepts: reconfiguration is intended to support transportation, manipulation, infrastructure deployment, civil preparation, and assistance to deploy inflatable modules within one platform family (Uno et al., 26 Dec 2025).

2. Reconfiguration mechanisms and connector design

MoonBot’s reconfiguration workflow includes on-palette self-assembly, in which modules are fixed to a launch palette and a Limb module, acting as a robotic arm, sequentially assembles itself with Wheel or other Limb modules to form the desired configuration. This process is central to the claim of on-demand reconfiguration, because the robot is not merely re-tasked in software; it is reconstituted physically into different morphologies (Uno et al., 26 Dec 2025).

Three connector types are described. The parallel jaw gripper is used mainly for MoonBot and is essential both for module assembly and general manipulation. The screw-type connector is gendered, 3D printed, and robust for structural joints. The diaphragm-type connector is genderless and allows quick, flexible coupling without strict alignment, although with slightly less connection strength. Connector sensing is implemented with infrared sensors in the gripper and at grapple points for reliable wireless detection of successful connections; the first prototype does not yet include a physical signal or power bus (Uno et al., 26 Dec 2025).

The parallel jaw gripper is reported with a maximum opening of 80 mm and a pinch force of 2.1 tons. Its sliding mechanism uses polyacetal (POM) for low friction and dust exclusion, and a trapezoidal screw converts BLDC motor rotary motion to jaw linear motion. For the screw-type connector, the reported lesson is that critical tolerancing is needed for passive self-alignment during assembly. For the diaphragm-type connector, the reported lesson is improved usability for self-reconfiguration and maintenance, but also a need for improved alignment protocols, with visual servoing and 2D markers identified as a possible remedy (Uno et al., 26 Dec 2025).

The lessons learned are materially important. Parallel jaw grippers and screw connectors were mechanically robust, successful in field tasks, and tolerant to mild misalignments. At the same time, sand and dust ingress were identified as a significant risk, especially for connectors with exposed screws or hatches. Future iterations are described as likely to incorporate improved sealing and possibly power and signal transfer across connections. This suggests that, in MoonBot, modularity is not treated as a purely geometric docking problem; it is also a contamination, sensing, and maintainability problem (Uno et al., 26 Dec 2025).

3. Software architecture, communication, and deployment

MoonBot’s control software is the Motion Stack, described as hierarchically modular, distributed, and aware of different module types and configurations. In the hardware-oriented paper, the control hierarchy is organized into five levels: Level 1: Individual joint control, Level 2: Inverse kinematics target computation, Level 3: Manage limb trajectories, Level 4: Synchronize multi-limb, and Level 5: Operator input, flexible mapping to subsystems. Teleoperation is primarily via joystick and keyboard, with visual feedback from wrist-mounted and field cameras. The reported joint controller is a “clamped integral” controller,

$$u_k = \max\!\left(y_k - \delta_e,\ \min \left( y_k + \delta_e,\ u_{k-1} + \dot{r}_k (t_k - t_{k-1}) \right)\right),$$

where $u_k$ is the position command, $\dot{r}_k$ is the input velocity, $y_k$ is the current joint measurement, and $\delta_e$ is the maximum allowed error. The stated fail-safe property is that, if communication is lost, joints do not move (Uno et al., 26 Dec 2025).

The software-architecture paper generalizes this into a broader framework of distributed, heterogeneous modularity, extending modular robotics beyond physical reconfiguration to software, communication, and orchestration. Its component structure separates Core, Injection/Override, API, Interface, Executor, and Communication blocks, with hardware- and robot-specific code isolated from the main logic. Communication is explicitly data-centric rather than node-centric, using ROS2 with Zenoh as middleware in place of DDS. The paper reports lower startup times and bandwidth requirements, better stability at 10+ robots/modules vs. 4 maximum for DDS, and faster connectivity recovery when switching from DDS to Zenoh (Neppel et al., 3 Nov 2025).

Deployment is handled by an orchestrator that manages version control, build, assembly-specific configuration, and process launch. Its Robot Assembler computes kinematics, URDFs, host- and hardware-specific settings, and per-module calibration for a given assembly; its launcher starts only the required processes for each hardware node. The reported operator interface reduces routine deployment to three actions per robot: update, build, and start. In field deployment, the system supported multiple robot families + 7 module types, assemblies with >60 live ROS2 components, and routine assembly changes reduced from 1-2 weeks to 20 seconds (Neppel et al., 3 Nov 2025).

These software results matter because MoonBot is presented not merely as a modular mechanism, but as a modular robotic system that must scale across time, hardware, teams, and operational environments. The platform’s architectural claim is that dynamic reconfiguration, decentralized control, self-assembly, inter-robot cooperation, and remote operation become operationally manageable only when hardware modularity is matched by communication and deployment modularity (Neppel et al., 3 Nov 2025).

4. Autonomous alignment for self-assembly

For autonomous assembly and reconfiguration, MoonBot is paired with an agnostic end-effector alignment controller that does not rely on robot-specific kinematics, joint sensing, or force feedback. The controller uses only measured end-effector pose $(x_e, q_e)$ and measured target pose $(x_t, q_t)$, as provided by external vision or motion-capture systems. Its raw pose errors are defined as

$$\Delta d = \|x_t - x_e\|, \qquad \Delta\theta = 2 \arccos\left(|q_t^{-1} \cdot q_e|\right).$$

Translational and rotational bounds are then interpolated from the current error magnitude and modulated by shrink factors $f_j$, $f_k$, and $u_k$0, yielding effective bounds $u_k$1 and $u_k$2. The controller clamps the combined translational and rotational command to a 6D ellipsoidal hypersphere,

$u_k$3

and outputs

$u_k$4

A low-pass filter is applied to prevent abrupt changes between controller ticks (Karimov et al., 24 Oct 2025).

Two controller variants are reported. Version 1 is a discrete step-and-settle controller that issues a small pose increment and waits for stabilization before proceeding. Version 2 is a continuous velocity controller that applies the adaptive hypersphere clamp at every tick. Both were implemented without modification on two different 7-DOF MoonBot limbs in JAXA’s lunar environment simulator, using only external pose feedback from OptiTrack motion capture at 180 Hz, and validated on MoonBot Minimal and Dragon assembly targets (Karimov et al., 24 Oct 2025).

Version / Limb	Duration	Final errors
Version 1 (Limb V1)	$u_k$5 s	$u_k$6 mm, $u_k$7
Version 2 (Limb V1)	$u_k$8 s	$u_k$9 mm, $\dot{r}_k$0
Version 1 (Limb V2)	$\dot{r}_k$1 s	$\dot{r}_k$2 mm, $\dot{r}_k$3
Version 2 (Limb V2)	$\dot{r}_k$4 s	$\dot{r}_k$5 mm, $\dot{r}_k$6

The qualitative comparison is explicit. Version 1 produced highly smooth, monotonic motion with minimal lateral wobble and little to no oscillation. Version 2 converged more quickly, typically reducing alignment time by 20–25%, and generally outperformed Version 1 in translation at minor cost to final orientation accuracy, which remained sub-degree. Both versions remained robust under substantial imperfections, including up to 10° joint backlash and 10 cm flex/wobble, which is directly relevant to self-assembly under lunar-relevant disturbances (Karimov et al., 24 Oct 2025).

5. Excavation as an ISRU payload

MoonBot is also presented as a carrier and integration platform for lunar excavation hardware. The bucket drum excavator paper describes a tool developed for MoonBot as part of the Japanese Moonshot program, with a centrally located mechanical interface on the shaft between two drum assemblies so that it can attach to MoonBot’s arm modules. The excavator uses two drums, each with four pairs of opposing spiral-shaped buckets spanning 160°, and the buckets are rotationally offset so that only one bucket engages regolith at a time. The stated purpose is to minimize instantaneous mechanical loads and reaction forces, a low-gravity consideration. Felt seals are used at moving interfaces to reduce dust ingress, and the prototype is primarily PLA, with gears and structural parts reinforced by carbon fibers (Giel et al., 1 Nov 2025).

The prototype tool weighs 4.8 kg and has a volume of 14.06 L. Reported actuation consists of a 24 V, 50 W EC motor (Maxon), a 936:1 ceramic planetary gearhead, and a maximum drum speed of ~7.5 rpm, controlled by a Maxon EPOS4 Disk 60/8 driver in Profile Velocity mode. The integrated Dragon configuration has a total system mass of 95 kg, including the MoonBot platform. Validation was performed in a 3 m × 4 m sandbox with Tohoku Silica Sand, 8 cm deep, using both hand-operated and MoonBot-operated trials (Giel et al., 1 Nov 2025).

Mode	Excavation rate	Normalized energy consumption
Continuous excavation	$\dot{r}_k$7	$\dot{r}_k$8
Batch operation	$\dot{r}_k$9	$y_k$0

Additional reported metrics clarify the performance envelope. In continuous mode, the volumetric rate is 0.53 m$y_k$1/h, average power draw is ~16.74 W, and the operation time to material-loss threshold is ~46 s per cycle at full fill. In batch operation, the reported cycle breakdown is 46.08 s at 146.74 W for excavation, 23.3 s at 144.13 W for unloading, and 138.89 s at 150 W for driving 10 m @ 0.14 m/s. The limiting factor is stated to be transit time and process overheads outside raw excavation. Capacity measurements report a maximum mass load of 12.87 kg, an operational (no loss) mass load of 9.95 kg, and filling efficiencies ranging from 62% at max fill for the drum only to 32% at onset of loss when all buckets are included. The design is described as comparable to NASA RASSOR 2.0 in continuous and batch excavation rates and in low normalized energy use (Giel et al., 1 Nov 2025).

The significance for MoonBot is architectural as much as mechanical. The excavator is characterized as a removable payload compatible with the modular platform, enabling alternative ISRU mission scenarios such as mobile excavators depositing regolith at a staging point while a dedicated transport rover shuttles material to processing sites. Future improvements are stated to include force and torque sensors, fill-level or material-mass sensing, autonomous control for depth-of-cut regulation and full-cycle autonomy, structural redesign for lunar vacuum and thermal cycling, and the use of FEM and DEM for further design iteration (Giel et al., 1 Nov 2025).

6. Demonstrations, performance envelope, and lessons learned

The principal field demonstration reported for MoonBot took place at the JAXA Advanced Facility for Space Exploration in a 20 m × 20 m analog field of silica sand, with additional environments containing actual regolith simulant, over three weeks continuous testing. Operation was primarily teleoperated, while on-palette assembly was semi-automated. The milestone tasks were construction-oriented: on-palette assembly of robots, mobility and traversability tests, rock and boulder removal, terrain leveling, transport and deployment of infrastructure, and assistive operations for inflatable habitat deployment (Uno et al., 26 Dec 2025).

The demonstrated mobility envelope depends on morphology. Minimal, Vehicle, Dragon, and Multicycle configurations were tested. The system demonstrated adaptive locomotion through wheel–limb collaboration, slope climbing up to 20°, and traversal of sandy and rough terrain. In civil engineering tasks, the Dragon configuration used its front limb as a manipulator for rock removal, while a Minimal configuration transported sleds. Terrain leveling was performed with a raking tool and wrist-mounted camera feedback. For infrastructure handling, a Dragon configuration transported and deployed an extendible >30 kg tower representing a solar or communications mast, and collaborative operations were used to erect and manipulate panel-type components (Uno et al., 26 Dec 2025).

The assistive habitat deployment scenario further broadens the claimed operational scope. In that task, a Dragon configuration monitored inflatable module inflation with a hand-eye camera, checked for leaks via pressure feedback, and placed restraint stoppers to stabilize the pressurized module after deployment. Separate durability tests at DLR/ESA LUNA reported reliable operation for 3 days in a high-dust environment, including descending and ascending crater terrain with 10°–20° slopes (Uno et al., 26 Dec 2025).

The reported hardware performance figures include 2 kg at full limb extension for each module under Earth gravity, scaling to >12 kg on lunar gravity, >2 tons grip force at the end effector, 480 mm wheels, and 30 kg payload capacity for the wheel module. Connection success was reported as high for gripper and screw-type connectors, while the diaphragm-type connector required higher precision in alignment but improved notably with visual feedback. In parallel, the software-architecture paper reports months of field deployment with self-assembling robots, inter-robot cooperation, and remote operation, including teleoperation across WiFi, internet, cellular 5G, and back to WiFi with negligible added architectural or operational complexity (Uno et al., 26 Dec 2025, Neppel et al., 3 Nov 2025).

Taken together, these results position MoonBot as a modular robotic system for lunar infrastructure rather than a single fixed robot. The literature emphasizes that the central challenge is not only physical reconfiguration, but also the coordinated design of connectors, control, communication, orchestration, and task-specific payloads. A plausible implication is that MoonBot’s primary research contribution lies in treating lunar construction robotics as a problem of system-level modularity across hardware, software, and operations rather than as a problem of locomotion or manipulation in isolation.