The Self-Driving Lab: Unifying Hardware with AI

Author: Denis Avetisyan


A new operating system aims to orchestrate laboratory equipment, bringing the power of artificial intelligence to the entire scientific workflow.

UniLabOS establishes a unified laboratory management system by virtualizing heterogeneous scientific instruments as networked endpoints, structuring resources through a hierarchical tree encoding ownership and a physical graph mapping material pathways, and dynamically registering device capabilities via automated source code inspection-thereby enabling centralized control, AI integration, and streamlined workflow orchestration within a fully digitalized research environment.
UniLabOS establishes a unified laboratory management system by virtualizing heterogeneous scientific instruments as networked endpoints, structuring resources through a hierarchical tree encoding ownership and a physical graph mapping material pathways, and dynamically registering device capabilities via automated source code inspection-thereby enabling centralized control, AI integration, and streamlined workflow orchestration within a fully digitalized research environment.

UniLabOS introduces an AI-native platform for autonomous experimentation, leveraging ROS 2 and a novel CRUTD protocol to bridge high-level planning and low-level control of laboratory resources.

Despite the promise of accelerated discovery through robotic experimentation, realizing fully autonomous laboratories remains hindered by fragmented software architectures separating high-level planning from low-level execution. This work introduces UniLabOS: An AI-Native Operating System for Autonomous Laboratories, a unified platform that virtualizes laboratory hardware and reconciles digital decision-making with embodied experimentation via a novel Action/Resource model and transactional CRUTD protocol. By unifying disparate instruments within a distributed edge-cloud architecture, UniLabOS enables scalable, reproducible, and provenance-aware autonomous experimentation across reconfigurable topologies. Will this AI-native operating system pave the way for a new era of self-driving laboratories and accelerated scientific innovation?


The Inherent Bottlenecks of Contemporary Laboratory Practice

Laboratory research, despite advancements in instrumentation, often remains constrained by the sheer volume of manual steps required to progress experiments. Researchers frequently spend considerable time on tasks such as sample preparation, instrument setup, data collection initiation, and physical transfer of materials – actions that, while seemingly minor individually, accumulate to significantly impede overall throughput. This reliance on hands-on intervention not only limits the speed at which experiments can be completed, but also introduces potential for human error and inconsistencies between trials. Consequently, the pace of scientific discovery is often dictated not by the capabilities of the equipment, but by the logistical challenges of coordinating and executing these predominantly manual workflows, creating a substantial bottleneck in modern laboratory settings.

Current laboratory data management relies heavily on systems designed primarily for post-experiment documentation. While Laboratory Information Management Systems (LIMS) and Electronic Laboratory Notebooks (ELNs) are robust at archiving data, tracking samples, and ensuring reproducibility, they often function as disconnected silos. These platforms typically lack the capacity for real-time interaction with laboratory hardware – robotic liquid handlers, automated microscopes, or high-throughput screening devices – meaning researchers must manually input data or rely on cumbersome middleware. This separation prevents a truly integrated workflow where experimental parameters can be adjusted dynamically based on incoming results, hindering rapid iteration and the full exploitation of advanced instrumentation’s capabilities. The result is a significant bottleneck, limiting the speed at which scientific discovery can progress, as data isn’t seamlessly translated into immediate action or further experimentation.

The disconnect between data storage systems and actual laboratory hardware significantly restricts experimental agility. Modern scientific instrumentation – high-throughput screening devices, automated microscopes, and robotic liquid handlers – generates data at an unprecedented rate, yet this potential is often unrealized due to fragmented workflows. Researchers frequently spend considerable time manually transferring data between instruments, software platforms, and archival systems, introducing delays and potential errors. This lack of seamless integration prevents rapid iteration on experimental designs, hindering the ability to quickly analyze results, adjust parameters, and re-run experiments in a closed-loop fashion. Consequently, the full capabilities of advanced instrumentation remain untapped, slowing down the pace of discovery and limiting the efficiency of scientific research.

UniLabOS establishes universal hardware interoperability by abstracting diverse laboratory instruments and communication protocols-including Modbus, PLC, ROS 2, OPC UA, and TCP/IP-into a unified integration layer, enabling seamless device enrollment across vendor ecosystems.
UniLabOS establishes universal hardware interoperability by abstracting diverse laboratory instruments and communication protocols-including Modbus, PLC, ROS 2, OPC UA, and TCP/IP-into a unified integration layer, enabling seamless device enrollment across vendor ecosystems.

UniLabOS: A Paradigm Shift in Laboratory Automation

UniLabOS represents a shift in laboratory operating systems from passive data acquisition and storage to active workflow orchestration and real-time control. Traditional laboratory information management systems (LIMS) primarily focus on data handling after an experiment is completed. UniLabOS, however, is designed to manage and automate the entire experimental process. This includes dynamic scheduling of instruments, automated execution of protocols, and closed-loop control based on real-time sensor feedback. The system is built on the principle of integrating hardware and software control at a fundamental level, enabling automated experimentation and reducing manual intervention. This functionality moves beyond simply recording results to actively directing and adapting experimental parameters during runtime.

UniLabOS employs a standardized abstraction layer utilizing the A/R/A&R categorization to represent all laboratory components. ‘Actions’ define executable processes or functions within the lab environment, such as heating, mixing, or measurement. ‘Resources’ represent physical assets – instruments, reagents, samples, or containers – necessary for these actions. ‘A&R’ (Action & Resource) designates combined entities where an action is inherently linked to a specific resource, like an automated pipetting station. This unified categorization allows UniLabOS to create a cohesive digital representation of the lab, enabling software to interact with and control diverse hardware and data sources through a common interface, simplifying integration and automation of complex workflows.

The UniLabOS architecture is fundamentally structured around the Action Layer and the Physical Graph. The Action Layer facilitates dynamic control of laboratory devices, moving beyond static programming to enable real-time adjustments based on feedback from those devices; this includes managing parameters, initiating procedures, and interpreting sensor data. Complementing this is the Physical Graph, a computational representation of all feasible material pathways within the lab environment. This graph defines valid transitions between resources and actions, ensuring that automated workflows adhere to physical constraints and instrument capabilities. The combination of these two elements allows UniLabOS to not only execute commands but also to validate and adapt them based on the current state of the laboratory and its equipment.

UniLabOS facilitates closed-loop workflows by separating high-level experimental goals – the ‘intent’ – from the specific hardware and software configurations required to achieve them – the ‘implementation’. This decoupling is achieved through a standardized interface allowing users to define desired outcomes without prescribing precise operational steps. The system then autonomously selects, configures, and controls laboratory equipment to execute the experiment, monitoring results in real-time and adjusting parameters as needed based on feedback. This automated cycle of planning, execution, and analysis significantly reduces manual intervention, minimizes errors, and enables rapid iteration, thereby accelerating the pace of scientific discovery and validation.

UniLabOS virtualizes liquid handling through a hierarchical object tree, a layered architecture compiling protocols into hardware-agnostic commands, an LLM agent utilizing the Model Context Protocol for planning, and a web interface providing real-time workflow visualization and resource tracking.
UniLabOS virtualizes liquid handling through a hierarchical object tree, a layered architecture compiling protocols into hardware-agnostic commands, an LLM agent utilizing the Model Context Protocol for planning, and a web interface providing real-time workflow visualization and resource tracking.

Underlying Infrastructure: Enabling Autonomous Laboratory Functionality

UniLabOS utilizes the Robot Operating System 2 (ROS 2) and Data Distribution Service (DDS) as its core communication layers, facilitating a decentralized architecture for laboratory instrument control and data acquisition. ROS 2 provides a flexible framework for managing instrument nodes and their interactions, while DDS ensures reliable, real-time data exchange with low latency and high throughput. This combination allows for instruments to operate autonomously and communicate directly with each other, eliminating the need for a central controller and enabling scalable, distributed workflows. The infrastructure supports both publish-subscribe and request-reply communication patterns, crucial for coordinating complex experimental procedures and handling diverse data streams from various instruments within the autonomous lab environment.

AST Inspection is a critical component of UniLabOS, providing dynamic driver registration and validation to ensure safe and reliable operation of automated laboratory equipment. This process involves verifying the functionality and compatibility of newly integrated instrument drivers before they are permitted to control hardware. Specifically, AST Inspection confirms that drivers adhere to pre-defined safety protocols and communication standards, mitigating the risk of unintended actions or system instability. The system automatically checks driver signatures, memory access patterns, and communication interfaces, rejecting any driver that fails these validation checks. This dynamic approach allows for the addition of new instruments and capabilities without compromising system integrity, and provides a mechanism for ongoing monitoring of driver health.

UniLabOS facilitates direct integration of automated laboratory workstations, specifically the Liquid Handling Workstation and Organic Synthesis Workstation, by providing a unified software interface and communication protocol. This integration extends the native capabilities of these instruments, enabling programmatic control of instrument parameters, automated experiment sequencing, and real-time data acquisition. Through UniLabOS, these workstations are no longer isolated entities but become components of a larger, interconnected system capable of executing complex, multi-step workflows without manual intervention. This approach allows for increased throughput, improved reproducibility, and the potential for remote operation and monitoring of laboratory processes.

The UniLabOS platform demonstrated its capacity for high-throughput experimentation by autonomously synthesizing 41 unique compounds over a continuous 17-day operational period. This automation encompassed all aspects of the synthesis workflow, including reagent dispensing, reaction monitoring, and product purification. The sustained operation highlights the system’s reliability and ability to manage complex, multi-step processes without human intervention, representing a significant increase in experimental efficiency and throughput compared to traditional manual methods.

The UniLabOS infrastructure extends to specialized facilities such as the Electrolyte Foundry, enabling automated maintenance of critical process parameters. Specifically, the system provides closed-loop control of Total Dissolved Solids (TDS) within electrolyte solutions. This is achieved through integration with appropriate sensors and actuators, allowing for real-time monitoring and adjustments to maintain desired concentrations. Automated TDS control is essential for ensuring the consistency and quality of electrolytes produced, which directly impacts the performance of batteries and other electrochemical devices.

UniLabOS facilitates host-device communication through a layered architecture utilizing networking, material synchronization, ROS 2 Actions, and telemetry topics to manage device nodes and bridge external connections.
UniLabOS facilitates host-device communication through a layered architecture utilizing networking, material synchronization, ROS 2 Actions, and telemetry topics to manage device nodes and bridge external connections.

Towards Laboratory as a Service: Democratizing Scientific Discovery

UniLabOS establishes a Digital Twin – a dynamic, virtual replica of the physical laboratory environment. This isn’t merely a visual representation; it’s a fully functional simulation that mirrors the lab’s state in real-time, allowing researchers to monitor ongoing experiments, predict outcomes, and even control instruments remotely. By creating this virtual counterpart, UniLabOS enables comprehensive ‘what-if’ scenarios to be tested in silico before committing resources to physical experimentation, dramatically reducing wasted materials and accelerating the pace of scientific validation. The Digital Twin actively reconciles data from all connected instruments and the CRUTD protocol, providing a single, authoritative source of truth for the entire experimental process and facilitating a level of control previously unattainable in traditional lab settings.

The integrity of scientific experimentation hinges on precise material tracking, and the CRUTD Protocol addresses this critical need by meticulously managing a material’s complete lifecycle – from initial sourcing and modification to its ultimate state after each experimental step. This protocol doesn’t merely log material usage; it actively reconciles the intended state of a substance with its actual state, identifying discrepancies that might compromise results. By establishing a continuous feedback loop and employing robust error detection, CRUTD ensures accurate data provenance and reproducibility. This level of granular tracking not only minimizes experimental errors but also facilitates optimization; researchers can pinpoint inefficiencies in material handling and refine processes for maximum yield and reduced waste, ultimately accelerating the pace of scientific advancement.

The advent of fully automated laboratory systems facilitated by platforms like UniLabOS is poised to fundamentally reshape scientific access through the emergence of Laboratory as a Service (LaaS). This innovative model transcends traditional limitations by offering on-demand access to sophisticated instrumentation, robotic automation, and specialized expertise, effectively democratizing research capabilities. Rather than requiring substantial capital investment and dedicated personnel, scientists can remotely design, execute, and analyze experiments, paying only for the resources consumed. LaaS promises to accelerate discovery by removing logistical hurdles, enabling collaborations across geographical boundaries, and empowering researchers – regardless of institutional affiliation or funding level – to pursue cutting-edge investigations with unprecedented efficiency and scalability.

UniLabOS’s architecture isn’t confined by the limitations of a single laboratory; its scalability has been successfully demonstrated through deployment across a geographically distributed system comprising three physically isolated workstations. This distributed configuration proves the platform’s capacity to integrate and coordinate experimental processes regardless of physical location, facilitating collaboration and resource sharing. By operating effectively across these independent nodes, UniLabOS establishes a foundation for a network of automated laboratories, capable of tackling complex scientific challenges through parallelized experimentation and increased throughput – a significant step towards democratizing access to advanced scientific infrastructure and accelerating the pace of discovery.

Recent advancements in automated laboratory workflows have demonstrated a remarkable tenfold increase in throughput for organic synthesis, signifying a substantial acceleration in the pace of scientific discovery. This improvement wasn’t achieved through incremental changes, but via a fully integrated system capable of autonomously managing complex experimental procedures. By automating reagent handling, reaction monitoring, and data analysis, the system minimizes human error and drastically reduces the time required to complete iterative experiments. The implications extend beyond simply performing more experiments; the increased speed allows researchers to explore a wider range of chemical possibilities and rapidly validate hypotheses, potentially unlocking breakthroughs in fields like drug discovery and materials science. This heightened efficiency promises a future where scientific innovation is no longer limited by the constraints of manual labor and time-consuming processes.

Representing laboratory workflows as sequences of Create, Read, Update, Transfer, and Delete (CRUTD) operations provides a unified, provenance-aware record of material lifecycle management, explicitly capturing spatiotemporal constraints during material relocation.
Representing laboratory workflows as sequences of Create, Read, Update, Transfer, and Delete (CRUTD) operations provides a unified, provenance-aware record of material lifecycle management, explicitly capturing spatiotemporal constraints during material relocation.

The development of UniLabOS, as detailed in the paper, exemplifies a commitment to rigorous system design. It isn’t merely about automating existing laboratory procedures; it’s about constructing a fundamentally correct and verifiable platform for scientific discovery. This echoes the sentiment of Edsger W. Dijkstra: “Program testing can be a very effective way to find errors, but it is hopelessly inadequate for confirming the absence of errors.” UniLabOS, with its CRUTD protocol and focus on virtualization, strives for that absence – a provably correct system bridging the gap between high-level experimental planning and low-level hardware execution. The architecture isn’t simply about what works but about establishing a logically complete foundation for autonomous experimentation.

Future Trajectories

The presentation of UniLabOS, while a logical step towards fully automated scientific inquiry, merely clarifies the scope of the remaining challenges. The CRUTD protocol, for instance, represents a functional abstraction, yet its formal verification regarding deadlock freedom and composability remains conspicuously absent. Asserting ‘autonomy’ without provable guarantees of safe and correct operation is, at best, optimistic. The system’s scalability, currently demonstrated within a constrained laboratory environment, will necessitate rigorous analysis of its asymptotic complexity as the number of virtualized instruments and concurrent experiments increases. A doubling of complexity does not necessarily imply a doubling of computational cost; rather, a careful examination of the logarithmic or polynomial dependencies will be crucial.

Furthermore, the digital twin, currently functioning as a reflective model, lacks the capacity for predictive behavior. True autonomy demands not merely mirroring the physical state, but anticipating it. This requires integration with probabilistic modeling frameworks and the development of algorithms capable of quantifying uncertainty. The current reliance on ROS 2, while pragmatic, introduces inherent limitations in real-time determinism. A move towards a formally verified, real-time operating system kernel may ultimately be unavoidable, even at the cost of increased development complexity.

Ultimately, the pursuit of a self-driving laboratory is not simply an engineering problem. It is an exercise in applied epistemology – a forced articulation of what constitutes ‘knowledge’ within a machine. The question is not whether such a system can be built, but whether its outputs are meaningfully distinguishable from random noise, and whether a formal proof of correctness is even attainable, given the inherent complexity of the scientific method itself.


Original article: https://arxiv.org/pdf/2512.21766.pdf

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

See also:

2025-12-29 15:22