Teamwork Makes the Dream Work: Coordinating Robots for Lunar Exploration

Author: Denis Avetisyan

A new framework, MOSAIC, is enabling heterogeneous robotic teams to work together more effectively, reducing the burden on human operators during complex missions.

MOSAIC is a modular and scalable autonomy system built on ROS 2 for intelligent coordination of diverse robots in challenging environments.

While increasingly capable, mobile robots exploring challenging environments like disaster zones or distant planets often remain tethered to continuous human supervision, limiting scalability and operational reach. This paper introduces ‘MOSAIC: Modular Scalable Autonomy for Intelligent Coordination of Heterogeneous Robotic Teams’, a novel framework designed to enable robust, multi-robot scientific exploration through a unified mission abstraction and layered autonomy. Validated in a space-analog field experiment with a team of five robots, MOSAIC achieved 82.3% task completion with a high autonomy ratio and minimal operator workload, even after a robot failure. How can such scalable, resilient autonomy architectures further unlock the potential for collaborative robotic exploration of previously inaccessible environments?

Beyond Predictability: Embracing Adaptability in Robotic Exploration

Historically, robotic exploration has been characterized by meticulously planned sequences of actions, a methodology that prioritizes predictability over responsiveness. These pre-programmed directives, while effective in controlled settings, present a critical limitation when encountering the unpredictable realities of extraterrestrial environments. A rover operating on Mars, for example, might be instructed to analyze a specific rock formation; however, an unexpected dust storm or an intriguing geological feature outside its designated path would likely be ignored. This reliance on rigid programming restricts a robot’s ability to capitalize on serendipitous discoveries or react effectively to hazards, ultimately hindering the scope and efficiency of scientific investigation and demanding increasingly detailed pre-mission planning to anticipate every possible scenario.

The inherent rigidity of pre-programmed robotic systems presents a considerable hurdle for ventures into the unpredictable terrains of the Moon and Mars. Unlike Earth-based automation operating within structured environments, extraterrestrial exploration demands responses to constantly shifting geological features, unanticipated obstacles, and variable lighting conditions. A rover following a fixed path, for example, may become immobilized by an unmapped boulder or unable to analyze a scientifically compelling feature slightly off-course. This limitation isn’t merely a matter of convenience; it directly impacts mission success, potentially preventing the collection of crucial data or even jeopardizing the entire operation as pre-defined contingency plans prove inadequate for truly novel situations. Consequently, the inability of robots to gracefully handle the unexpected dramatically increases the risk and cost associated with complex, long-duration planetary science missions.

For robotic exploration to truly flourish beyond carefully controlled environments, a paradigm shift towards autonomous decision-making is essential. Current robotic systems frequently operate on pre-programmed instructions, proving inadequate when confronted with the unpredictable nature of extraterrestrial landscapes or unexpected operational challenges. A robust solution necessitates integrating advanced algorithms – including machine learning and real-time sensor data analysis – allowing robots to independently assess situations, modify plans, and execute actions without constant human intervention. This capacity for adaptation isn’t simply about avoiding obstacles; it’s about proactively identifying scientific opportunities, optimizing resource utilization, and ensuring mission success even when faced with unforeseen dynamic conditions – a key step towards self-sufficient, long-duration exploration.

MOSAIC: A Framework for Coordinated Autonomy

The MOSAIC framework employs a modular architecture built upon independent, yet interoperable, software components. This design facilitates scalability by allowing the addition or removal of robotic units and capabilities without requiring substantial modifications to the core system. Each module encapsulates specific functionalities – such as perception, planning, or control – and communicates with others via well-defined interfaces. This separation of concerns enhances maintainability and allows for the integration of robots possessing heterogeneous hardware and software configurations. The framework supports teams ranging in size and complexity, from a few specialized robots to large-scale deployments with diverse robotic assets, all operating within a coordinated system.

The MOSAIC framework employs a Unified Mission Abstraction (UMA) as an intermediary step between high-level mission goals and individual robot actions. This abstraction defines mission objectives in a generalized, robot-agnostic manner, allowing the system to decompose these objectives into a series of abstract tasks. A planning module then translates these abstract tasks into concrete, executable actions specific to each robot’s capabilities and constraints. This process ensures coherent execution by providing a consistent interpretation of the mission across the entire robotic team, and facilitates adaptability as the UMA can be updated to reflect changing priorities or environmental conditions without requiring modifications to individual robot control systems.

Layered Autonomy within the MOSAIC framework establishes a tiered system for robot control, ranging from fully autonomous operation to direct human intervention. This approach allows robots to independently execute pre-defined tasks and react to predictable scenarios without constant oversight. However, when faced with unexpected events, ambiguous situations, or tasks requiring complex judgment, the system enables a human operator to assume control at any level – from modifying individual robot actions to re-tasking the entire team. This balance is achieved through a shared understanding of the mission objectives and a communication infrastructure that facilitates seamless transitions between autonomous and manual control, optimizing both efficiency and safety.

Decentralized Task Management within the MOSAIC framework distributes task allocation and execution responsibilities directly to individual robots, eliminating the need for a central coordinating entity. Each robot maintains a local world model and, based on its capabilities and perceived environment, autonomously selects and executes tasks from a shared task list or generates new tasks based on overarching mission objectives. This approach enhances robustness by mitigating single points of failure and improves scalability, as the computational burden of task management is distributed across the robotic team. Robots communicate directly with each other to negotiate task assignments, share information about task progress, and resolve conflicts, ensuring coordinated action without reliance on centralized intervention.

Underpinning Technologies: Perception, Communication, and Control

MOSAIC utilizes Simultaneous Localization and Mapping (SLAM) algorithms to enable robots to concurrently build a map of their surroundings while simultaneously determining their own location within that map. This is achieved through the processing of sensor data – typically from cameras, LiDAR, and inertial measurement units – to identify landmarks and features. The resulting map is not merely a visual representation; it’s a probabilistic model that allows multiple robots within the MOSAIC system to share a consistent understanding of the environment, facilitating coordinated navigation and task execution. This shared environmental model is crucial for collaborative robotics applications, enabling robots to operate effectively in dynamic and unstructured spaces without requiring pre-programmed paths or external guidance.

Mesh communication within the MOSAIC framework establishes decentralized, peer-to-peer connections between robots, eliminating the single point of failure inherent in centralized communication systems. This architecture allows robots to relay messages to each other, extending the communication range beyond direct transmission capabilities and maintaining connectivity even if individual robots lose connection to the primary network. Data transmission utilizes flooding or routing protocols to ensure message delivery, with redundancy built-in to mitigate signal interference or robot failure. The resulting network demonstrates increased robustness and scalability compared to traditional hub-and-spoke configurations, allowing for reliable operation in dynamic and potentially unpredictable environments.

Behavior Trees (BTs) offer a modular and hierarchical method for designing robot control systems. Unlike finite state machines, BTs allow for concurrent execution of tasks and facilitate easy modification and extension of robot behaviors without significant code restructuring. A BT is constructed from nodes representing actions, conditions, and control flow, enabling complex behaviors to be broken down into manageable, reusable components. This structure promotes code maintainability, scalability, and allows for dynamic behavior adaptation based on sensor input and environmental conditions. The framework leverages the inherent advantages of BTs for defining sequential and parallel actions, conditional branching, and iterative loops within a robot’s operational logic.

The MOSAIC framework utilizes ROS 2, a production-grade robotics middleware, to provide a standardized communication layer and essential services for distributed robot systems. ROS 2 enables inter-process communication via data distribution service (DDS), offering quality of service (QoS) configurations for reliable data exchange in potentially unreliable network conditions. This includes features such as publish-subscribe messaging, service calls, and parameter management. By leveraging ROS 2, MOSAIC gains access to a mature ecosystem of tools for robot development, simulation, and debugging, alongside established packages for perception, planning, and control. Furthermore, ROS 2’s support for multiple operating systems and hardware platforms enhances the portability and scalability of the framework.

Field Validation: Performance in a Lunar Analog

Field experiments were conducted within a lunar analog environment to evaluate the MOSAIC framework’s performance under conditions simulating a lunar surface. This analog scenario allowed for testing of the system’s capabilities in a realistic setting, prior to potential deployment in actual lunar missions. The environment was specifically chosen to replicate key aspects of lunar terrain and operational constraints, including limited communication bandwidth and challenging lighting conditions. Data collected during these field tests informed refinements to the MOSAIC framework and validated its operational parameters in a pre-flight assessment.

The system’s operational autonomy was quantified using the Autonomy Ratio, a metric representing the proportion of tasks completed without direct human intervention. During field trials in a lunar analog environment, the MOSAIC framework achieved an Autonomy Ratio of 86%. This indicates that the system was able to independently execute a substantial portion of its programmed functions, including navigation, data acquisition, and task planning, requiring minimal external control or assistance to maintain operation and fulfill objectives.

During field experiments conducted in a lunar analog environment, the MOSAIC framework achieved an 82.3% task completion rate. This performance was maintained despite the simulated failure of one robotic unit during operation. This result indicates a significant degree of robustness and resilience within the system, demonstrating its ability to continue functioning and achieving objectives even with component-level failures. The system’s architecture allows for continued operation by redistributing tasks and adapting to the reduced operational capacity, minimizing the impact of the robot failure on overall mission success.

During field experiments conducted in a lunar analog environment, the MOSAIC framework explored a total area of 758 m². The system achieved a mapping rate of 0.313 m²/s, indicating the speed at which the environment was digitally reconstructed. This rate reflects the efficiency of the robotic platform and onboard processing capabilities in acquiring and integrating sensor data to build a coherent map of the surveyed area.

Data collected during field validation indicates a measurement success ratio of 82.3%. While the initial measurement attempt succeeded in the majority of cases, a measurement retry was required for 19.4% of data points, suggesting occasional environmental interference or sensor limitations. The accuracy of the resulting maps was quantified using mean chamfer distance, which measured 0.145 m. This value represents the average distance between points on the generated map and their corresponding locations in ground truth, providing a metric for the system’s mapping error.

Future Implications: Expanding the Boundaries of Robotic Exploration

The MOSAIC framework represents a significant advancement in robotic systems design, establishing a robust foundation for creating machines capable of thriving in unpredictable environments. By prioritizing modularity and adaptability, this approach moves beyond traditionally rigid robotic architectures, allowing for easier reconfiguration and repair in the field. This inherent resilience isn’t limited to space exploration; the framework’s principles are readily applicable to terrestrial challenges such as disaster response, deep-sea exploration, and even complex industrial tasks. The resulting systems demonstrate improved operational longevity and reduced downtime, ultimately lowering the overall cost and risk associated with deploying robots in demanding conditions. Consequently, MOSAIC fosters a new era of robotic capability, promising machines that are not simply programmed to perform tasks, but are engineered to endure and evolve within dynamic real-world scenarios.

The MOSAIC framework’s design prioritizes both modularity and scalability, fundamentally easing the incorporation of technological advancements and novel capabilities. This architecture allows components – from sensors and actuators to software algorithms – to be swapped, upgraded, or augmented without necessitating a complete system overhaul. Consequently, robotic platforms built upon MOSAIC can readily benefit from breakthroughs in areas like artificial intelligence, materials science, or power systems. This adaptability isn’t simply about adding features; it’s about future-proofing robotic explorers, ensuring they remain at the cutting edge of exploration technology and capable of tackling increasingly complex challenges with minimal redesign and cost.

The development of increasingly autonomous robotic systems holds the potential to fundamentally alter the economics and scope of space exploration. By diminishing the need for constant, real-time human intervention – which is both expensive and subject to communication delays over vast distances – missions can be undertaken with greater efficiency and at a reduced cost. This shift allows for the deployment of robotic explorers to more remote and challenging environments, and enables longer-duration missions focused on complex scientific objectives. Consequently, a decreased reliance on direct human control opens doors to ambitious undertakings previously considered impractical, such as establishing persistent robotic outposts on distant celestial bodies or conducting extensive surveys of unexplored regions of space.

Ongoing development of the MOSAIC framework prioritizes imbuing robotic systems with robust learning capabilities and adaptive behaviors, moving beyond pre-programmed responses. Researchers are actively investigating methods for enabling robots to not only recognize unexpected situations, but also to formulate and implement novel solutions without human intervention. This involves integrating advanced machine learning algorithms, particularly reinforcement learning and evolutionary strategies, to allow robots to refine their performance through experience and optimize for success in dynamic, unpredictable environments. The ultimate goal is to create robotic explorers capable of independent decision-making, self-diagnosis, and autonomous problem-solving, dramatically expanding the scope and efficiency of future exploration missions to even the most challenging terrains and circumstances.

The MOSAIC framework, as detailed in the study, prioritizes a systemic approach to multi-robot coordination – a principle elegantly mirrored in Ada Lovelace’s observation: “The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform.” This highlights that even sophisticated autonomy, like that demonstrated by MOSAIC in lunar exploration scenarios, is fundamentally reliant on well-defined structures and instructions. The framework’s scalability and reduction of operator workload aren’t emergent properties, but rather the result of carefully designed modularity – a system where each component’s behavior contributes to the overall coordinated action. Just as Lovelace noted the engine’s dependence on human instruction, MOSAIC’s success rests on the clarity and precision of its underlying architecture.

The Road Ahead

The presentation of MOSAIC, a framework for coordinating robotic teams, feels less like a solution and more like a carefully charted observation point. One builds a nervous system – a modular architecture for autonomy – only to realize the true complexity resides not within the connections, but in the environment itself. Scalability, as demonstrated, addresses the multiplication of actors, but the unpredictable nature of lunar regolith, the subtleties of dust adhesion, or even the unforeseen interactions between robots remain largely external to the core system. The challenge, then, is not simply to add more nodes to the network, but to account for the emergent properties of the whole.

Future work must confront the inherent limitations of simulation. A perfect digital replica, capable of anticipating every contingency, remains a philosophical ideal. The true test of MOSAIC, and systems like it, will lie in prolonged deployment – observing how the framework adapts, or fails to adapt, to the unforgiving reality of an extraterrestrial landscape. One does not ‘fix’ a broken wheel; one redesigns the entire vehicle to navigate the terrain.

Ultimately, the reduction of operator workload, while a laudable goal, risks obscuring a fundamental truth: intelligent systems should augment human capabilities, not replace them. The most elegant automation is not that which eliminates the need for oversight, but that which allows for more insightful, more considered intervention. The question is not how to make robots independent, but how to foster a symbiotic relationship between human intellect and robotic execution.

Original article: https://arxiv.org/pdf/2601.23038.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/