LeRobot v2.1 Network Format

• LeRobot v2.1 Format is a JSON-based network description model that unifies primary network data with metadata for reproducibility and sustainable usability.
• It decomposes networks into nodes, links, properties, and weights to support analytical tractability and efficient data factorization.
• The format incorporates FAIR principles, temporal data, and advanced serialization techniques to optimize integration with dynamic robotic sensor networks.

A common format for describing networks, as elaborated in "Towards a format for describing networks / 2. Format elements" (Batagelj et al., 1 May 2025), is defined by a rich, self-contained structure unifying both the primary network data and its metadata. This format aims for data integrity, reproducibility, and sustainable usability, benefiting applications such as robotics (e.g., LeRobot v2.1), network science, and broad scholarly data exchange. The critical elements, mathematical foundations, data structures, metadata norms, and technical design choices underlying this format are summarized below.

1. Mathematical Foundations and Component Decomposition

A network is formally represented as a quadruple: $N = (V, L, P, W)$ where:

• $V$ is the set of nodes (vertices),
• $L$ the set of links (edges or arcs),
• $P$ denotes node (and optionally link) properties,
• $W$ comprises weights or other quantitative link attributes.

This explicit decomposition leads to a tabular representation split into two core tables: one for nodes $(V, P)$, and another for links $(L, W)$. Such structure supports analytical tractability and facilitates transformation between formats, directly enabling factorization and schema extension.

2. Data Structure and Serialization Principles

The format is built upon JSON for its seamless mapping to modern programming data structures. The prototypical NetsJSON structure is:

{ "netsJSON": "basic", "info": { ...metadata fields... }, "nodes": [ { "id": nodeId, "lab": label, "x": x, "y": y, ... }, ... ], "links": [ { "type": "arc"/"edge", "n1": nodeID1, "n2": nodeID2, "rel": r, ... }, ... ] }

Key characteristics:

• Data and metadata are co-located, ensuring context (creation date, version, provenance) is inseparable from the network itself.
• Supports structured values well beyond scalars, such as time series and functional properties (essential for dynamic or temporal networks).

This serialization principle enables the description of complex, multi-modal networks in a single, portable file, supporting archival and reproducibility.

3. Metadata and FAIR Principles

A central feature is the rigorous inclusion of metadata under the "info" key. Metadata requirements encompass:

• Origin, title, and creation/modification dates,
• Provenance and usage history, often as an events list with fields (date, title, author, description, URL, citation),
• Adherence to FAIR (Findable, Accessible, Interoperable, Reusable) norms, such as persistent identifiers, standardized schemas (Dublin Core, Schema.org), and rich contextual information.

These principles ensure datasets are catalogued, citable, and reusable. Detailed provenance supports integrity checking and long-term accessibility, directly benefiting collaborative research platforms and robotic network deployments (as with LeRobot v2.1).

4. Versatility for Diverse Network Types

The format design accommodates:

• Simple and complex networks, including 2-mode (bipartite), multi-relational, multi-level, and temporal networks (dynamic graphs),
• Arcs (directed links), undirected edges, and multiple (parallel) links,
• Special representations (e.g., molecular graphs, genealogies, topologies).

Support for structured properties (e.g., time series, functional data) under node or link attributes enables encoding networks where quantities go beyond ratio or nominal scales—essential when representing, for example, evolving robotic sensor networks.

5. Factorization and Default Values

Factorization is a technical design choice for storage and computational optimizations. It involves:

• Transformation of long or categorical labels/properties into integer codes, yielding reduced file size and faster processing.
• Maintenance of coding tables within metadata for recovery and mapping.
• Example: In bibliographic networks, node types ("person," "paper," "journal") are factorized; analogous approaches apply to robotic device roles or states.

Default values may be specified to further compress records, and abbreviated labels can be used to facilitate visualization or mitigate display clutter. A switch enables choosing between factorized/coded and human-readable formats.

6. Technical Flexibility and Integration with Robotics Formats

Additional specifications include:

• Indication of whether factorization is active,
• Allowance for abbreviated labels or code-based node/link fields,
• Structured value support for extended types or temporal data.

When ported to robotics formats like LeRobot v2.1, these features:

• Support self-description and integrated metadata for autonomous operation,
• Optimize performance for large-scale networks (e.g., communication graphs),
• Guarantee interoperability, especially when adhering to FAIR standards,
• Accurately model dynamic robotic systems through temporal and structured fields.

7. Applications and Implications for LeRobot v2.1

The described format is not limited to classical network analysis; many elements transfer directly to robotic or sensor network contexts:

• Self-contained files with bundled data and metadata aid autonomous decision-making and error recovery,
• Factorization and structured values allow for efficient analytics even in real-time robotics scenarios,
• Temporal and dynamic extensions cater to evolving robotic communication or collaboration networks.

In summary, a robust format for network description—modeled as $N = (V, L, P, W)$, serialized in JSON, and designed for technical flexibility—achieves both analytical depth and practical usability across scientific domains. This approach ensures rich, portable, and interoperable representations for diverse networks, including those within advanced robot frameworks such as LeRobot v2.1, thereby supporting reproducibility, efficient computation, and adaptive analytics.