LabOS: Autonomous Laboratory OS

• LabOS is a laboratory operating system paradigm that virtualizes experimental hardware, data, and workflows to streamline lab processes.
• It employs typed stateful abstractions, transactional workflow management, and directed acyclic graphs for efficient resource scheduling and error recovery.
• By integrating AI and XR interfaces, LabOS bridges digital decision-making with physical experimentation, leading to autonomous lab operations.

LabOS (“Laboratory Operating System”) is a software paradigm that abstracts and orchestrates the hardware, data, and workflows of laboratory environments using principles analogous to computer operating systems. LabOS systems enable scalable, reproducible, and adaptive experimentation by providing typed stateful abstractions, transactional workflow management, resource arbitration, and interfaces for both human and AI agents. This paradigm underpins a new generation of AI-native autonomous laboratory platforms, bridging the gap between digital decision-making and embodied experimental execution (Fei et al., 2024, Cong et al., 16 Oct 2025, Olowe et al., 2024, Gao et al., 25 Dec 2025).

1. Definition and Conceptual Framework

LabOS is defined by its role as an orchestrating “kernel” that manages the interplay between decision-making agents (AIs or scientists), heterogeneous experimental hardware, and digital state representations (“digital twins”). The laboratory is virtualized so that:

• Hardware devices (furnaces, pumps, robotic arms, sensors) function as peripherals or I/O units.
• Samples, reagents, and containers act as data objects or files.
• Tasks (heating, mixing, measurement, etc.) correspond to processes or threads.
• The central kernel schedules tasks, allocates resources, resolves conflicts, logs provenance, and enforces system-wide invariants (Fei et al., 2024, Gao et al., 25 Dec 2025).

LabOS architectures expose minimal, typed APIs for declaring actions (operations) and resources (materials, vessels), ensuring strong separation of scientific intent from hardware specifics.

2. Core Models and Abstractions

Key abstractions unify diverse LabOS implementations:

A. Action/Resource/Action&Resource (A/R/A&R) Model:

All laboratory entities are classified as:

• Actions (A): Operations invoked by the kernel (e.g., pipetting, stirring).
• Resources (R): Passive objects owning material states (e.g., vials, tanks).
• Action&Resource composites (A × R): E.g., heater–stirrers, combining ownership and action interfaces.

Formally,

$$A = \{a\,|\,a \text{ is an executable capability}\},\quad R = \{r\,|\,r \text{ is a vessel or material container}\},\quad AR = \{(a,r)\,|\,a \in A, r \in R\}$$

Type constraints, configuration schemas, and provenance are encoded as part of each entity's definition, supporting introspection and transactional state changes (Gao et al., 25 Dec 2025).

B. Workflow Graphs and Topologies:

Laboratory workflows are encoded as composite directed acyclic graphs (DAGs) $(T, E)$, where task nodes have attributes such as capacity, required device types/samples, and parameter sets. The global workflow is dynamically reconfigurable subject to acyclicity (Fei et al., 2024).

UniLabOS further represents lab structure as coupled graphs:

• A logical resource tree $T = (V, E_\mathrm{tree})$ for ownership relations.
• A physical connectivity graph $G = (V, E_\mathrm{phy})$ for data/material flow, enabling automated protocol mobility and safety verification (Gao et al., 25 Dec 2025).

3. Resource Management and Transactional Guarantees

Resource arbitration is central to all LabOS platforms. A typical mechanism involves:

• Cooperative, non-preemptive resource reservation: Tasks request resource bundles (device types/IDs, sample positions), queued and scheduled by a centralized manager with first-come-first-served, priority-aware policies (Fei et al., 2024).
• Resource status is tracked in real time (e.g., via MongoDB collections) and released atomically upon task completion or failure.

In UniLabOS, the CRUTD protocol extends basic CRUD with:

• Transfer: atomic, feedback-guarded movement of material/vessels within the resource tree.
• Guarantees that digital and real-world states remain synchronized, with rollback/recovery on actuation failure (Gao et al., 25 Dec 2025).

State transitions and provenance are enforced at each step, preventing race conditions and orphaned events.

4. System Architectures and Interfaces

LabOS implementations vary, but share fundamental design motifs:

• Manager–Worker Pattern: Separate experiment managers, task managers, resource managers, and device managers coordinated via a shared state bus (e.g., MongoDB, ROS 2/DDS) (Fei et al., 2024, Gao et al., 25 Dec 2025).
• API/GUI Access: Python experiment builders, REST/JSON APIs, and browser-based graphical dashboards support job submission, monitoring, human-in-the-loop intervention, and notification (Fei et al., 2024).
• AI Integration: LLMs and self-evolving agent populations (e.g., STELLA in (Cong et al., 16 Oct 2025)) provide co-scientist capabilities—planning, code generation, adaptive experiment design, error detection—often via function-calling APIs and natural-language interfaces (Olowe et al., 2024).
• XR Collaboration: Extended reality (XR)-enabled human-AI interfaces leverage AR glasses (e.g., Viture, Unity-based overlay systems) for egocentric data streaming, context-aware guidance, and protocol step synchronization (Cong et al., 16 Oct 2025).
• Distributed Topologies: Decentralized edge–cloud architectures support self-organizing device discovery, cross-lab federation, and resilience to network disruptions. ROS 2/DDS enables peer-to-peer actions and global state streaming at 10–100 Hz (Gao et al., 25 Dec 2025).

5. Practical Deployments and Case Studies

The LabOS paradigm has been validated across multiple autonomous laboratory environments:

Platform	Domain	Distinct Features
AlabOS	Materials synthesis	Python-based DAG workflow, FCFS resource scheduling, proven scalability in A-Lab instance (3,500+ samples) (Fei et al., 2024)
UniLabOS	General/autonomous laboratory	Typed A/R/A&R model, dual-topology, CRUTD protocol, supports cross-host mobility, validated in liquid handling, organic synthesis, distributed foundries (Gao et al., 25 Dec 2025)
LabOS (AI-XR Co-scientist)	Biological R&D; AI-human collab	Multimodal VLMs, LLM agent planning, XR-guided execution, real-time human-AI collaboration, task-specific performance benchmarks (Cong et al., 16 Oct 2025)
LABIIUM	Measurement automation	Zero-config device bridging, LLM-based code assistant, integration with VSCode/Python, highlights current LLM limitations for adaptive sampling (Olowe et al., 2024)

Each deployment emphasizes robust orchestration, resource synchronization, provenance logging, and system resilience.

6. Strengths, Limitations, and Extensions

Strengths:

• Abstracts laboratory complexity, decoupling scientific workflows from hardware specifics.
• Enables concurrency, conflict-free resource usage, and maximal device/sample utilization.
• Supports extensible integrations with AI planners, digital twins, and human interfaces.
• Demonstrated fault tolerance and adaptability to topological changes and network failures (Gao et al., 25 Dec 2025).

Limitations:

• Non-preemptive resource scheduling can cause bottlenecks; most platforms lack optimal time-windowed schedulers (Fei et al., 2024).
• Current priority assignment mechanisms are coarse, omitting fine-grained user or experiment preferences.
• Adaptive AI for real-time protocol design remains constrained by LLM representational limits and feedback integration capabilities (Olowe et al., 2024).
• Cross-lab orchestration and federated operation require further standardization and robust security models.

Potential Extensions:

• Integration of advanced ML-based scheduling, predictive duration modeling, and closed-loop AI reagent handling.
• Enhanced provenance and reproducibility support (native AiiDA, cross-protocol audit trails).
• Tighter fusion of physical actuation control (robotics) with AI agent planning layers.
• Improved XR hardware/software alignment for richer digital-physical overlays in real-time collaboration scenarios.

7. Impact and Outlook

LabOS frameworks are reshaping experimental science by standardizing the abstraction, execution, and oversight of complex laboratory operations. The kernelized, transactional approach enables reproducibility, safety, and agent readiness, offering robust foundations for the next generation of self-driving, collaborative laboratories across domains as diverse as materials, biology, and measurement science. Future advances hinge on deeper integration of adaptive AI, cross-site federation, and unified provenance modeling, positioning LabOS as a cornerstone for agent-native, reproducible scientific discovery (Fei et al., 2024, Cong et al., 16 Oct 2025, Olowe et al., 2024, Gao et al., 25 Dec 2025).