empa-aurora-2025-raw

Tier 1 raw mirror of the Empa / ETH Zürich / EPFL / SINTEF Aurora 2025 coin-cell cycling dataset, hosted under the BSEBench organization for reproducibility and provenance audit purposes. Files are preserved exactly as published on Zenodo, in the original RO-Crate .zip archive.

This dataset is the first publicly-released, large-scale battery cycling release natively conforming to the LF Energy Battery Data Alliance Battery Data Format (BDF). As such it occupies a special role in BSEBench : it is our gold-standard compatibility check for the BDF-aligned schema in bsebench-specs.

Status

Upload pending — placeholder dataset card. The Zenodo RO-Crate (a single Dataset-rocrate.zip, approx. 2.5 GB on disk, expanding to a substantially larger directory of per-cell .json, .csv, and .parquet files) will be uploaded once SHA-256 verified against the original Zenodo record at DOI 10.5281/zenodo.15481956.

What this is

A bit-exact mirror of the dataset accompanying :

Svaluto-Ferro, E., Kimbell, G., Kim, Y., Plainpan, N., Kunz, B., Scholz, L., Laeubli, R., Becker, M., Reber, D., Kraus, P., Kühnel, R.-S., Clark, S., Battaglia, C. (2025). Toward an Autonomous Robotic Battery Materials Research Platform Powered by Automated Workflow and Ontologized Findable, Accessible, Interoperable, and Reusable Data Management. Batteries & Supercaps, 2025. doi:10.1002/batt.202500155

@article{svalutoferro2025aurora,
  author  = {Svaluto-Ferro, Enea and Kimbell, Graham and Kim, YeonJu
             and Plainpan, Nukorn and Kunz, Benjamin and Scholz, Lina
             and Laeubli, Raphael and Becker, Maximilian and Reber, David
             and Kraus, Peter and K{\"u}hnel, Ruben-Simon
             and Clark, Simon and Battaglia, Corsin},
  title   = {Toward an Autonomous Robotic Battery Materials Research
             Platform Powered by Automated Workflow and Ontologized
             Findable, Accessible, Interoperable, and Reusable Data
             Management},
  journal = {Batteries \& Supercaps},
  year    = {2025},
  doi     = {10.1002/batt.202500155}
}

@dataset{svalutoferro2025aurora_zenodo,
  author    = {Svaluto-Ferro, Enea and Kimbell, Graham and Kim, YeonJu
               and Plainpan, Nukorn and Kunz, Benjamin and Scholz, Lina
               and Laeubli, Raphael and Becker, Maximilian and Reber, David
               and Kraus, Peter and K{\"u}hnel, Ruben-Simon
               and Clark, Simon and Battaglia, Corsin},
  title     = {Dataset for publication ``Toward an Autonomous Robotic
               Battery Materials Research Platform Powered by Automated
               Workflow and Ontologized Findable, Accessible,
               Interoperable, and Reusable Data Management''},
  publisher = {Zenodo},
  year      = {2025},
  doi       = {10.5281/zenodo.15481956},
  url       = {https://zenodo.org/records/15481956}
}

The data was generated by the Aurora battery robot platform at Empa (Swiss Federal Laboratories for Materials Science and Technology, Laboratory for Materials for Energy Conversion). Aurora is an automated coin-cell assembly + cycling robot, developed in collaboration between Empa and Chemspeed Technologies, that produces and tests coin cells with anode/cathode capacity balancing reported to a precision of approximately 0.01 mg. Experiments are orchestrated via the AiiDA workflow engine and AiiDAlab web-interface layer (originally developed for atomistic simulations by EPFL / NCCR MARVEL) — Aurora is reportedly the first documented use of AiiDA to drive experimental (rather than simulated) work.

Note : the Aurora robot, AiiDA, and AiiDAlab are three distinct components of the platform stack — they are sometimes informally bundled under the name AiiDAlab-Aurora, but the Zenodo metadata and the accompanying publication reference them separately.

Significance for BSEBench

This is the only dataset in our corpus that is BDF-native at source.

The LF Energy Battery Data Alliance announced the Battery Data Format (BDF) on 22 December 2025 as an open standard for battery time-series interoperability, aligned with the BattINFO domain ontology and supported by Empa, ETH Zürich, EPFL, SINTEF, the Faraday Institution, Imperial College London, and Microsoft (among others). The Aurora 2025 dataset was released earlier (Zenodo publication date 1 July 2025) but uses the same .bdf.parquet / .bdf.csv schema that BDF would later codify, and is cited in the BDA announcement as a flagship reference release.

For BSEBench, this dataset plays a schema-validation role :

1. The BSEBench-canonical TimeSeriesSchema (planned for bsebench-specs v0.2.0) is intentionally designed to be a strict subset / rename of BDF, not a competing schema. If the harmonized Tier 2 view of this dataset cannot be produced by a near-passthrough from *.bdf.parquet to BSEBench Parquet, then our schema is out of step with the community standard, and we should adjust BSEBench (not the data).
2. Any sign-convention surprises observed during harmonization (e.g. discharge sign, idle-step polarity) reveal underspecified areas of BDF where we need to commit to a documented convention (BPX-1.1 in our case).

In short : this dataset audits us, not the other way around.

Cell specifications

• Total cells : 199 (verified from Zenodo abstract).
• Format : coin cell. CR2032 form factor inferred from the Aurora platform's documented assembly capacity (32 cells per generation in CR2032 carriers reported in 2024 Empa communications) but NOT VERIFIED against the dataset's per-cell metadata at upload time. Confidence : MEDIUM.
• Chemistries : two families.
  • NMC // graphite
  • LFP // graphite Specific split between the two chemistries (e.g. how many of the 199 cells are NMC vs LFP) is NOT STATED in the Zenodo abstract or the available secondary sources, and must be derived per-cell from the metadata.json files at consumption time. Confidence : LOW until enumerated.
• Anode : graphite for both chemistry families (verified).
• Cathode : NMC (lithium nickel-manganese-cobalt oxide) and LFP (lithium iron phosphate) — verified from the abstract.
• Capacity : per-cell nominal capacities are reported in the per-cell BattINFO-ontologized JSON-LD metadata files (one per cell); exact range not stated in the dataset summary.

Cycling protocol

• Cycles per cell : 1000 (verified from the Zenodo abstract).
• Temperature : NOT STATED in the Zenodo abstract or the available secondary sources. Aurora cells are cycled in a temperature-controlled environment per Empa's standard practice, but the exact set-point (e.g. 25 °C, 30 °C) for this release is NOT VERIFIED and must be read from the per-cell metadata at upload time. Confidence : LOW.
• Charge / discharge protocol : standard CCCV (constant-current constant-voltage) per the Aurora cycler's default routine, with formation cycles followed by long-term cycling. Specific C-rate and voltage cutoffs are NOT STATED in the available summaries and must be derived from the per-cell metadata. Confidence : LOW.
• Sampling rate : NOT STATED in the available sources. Often approximately 1 Hz on coin-cell cyclers, but NOT VERIFIED for this dataset. Confidence : LOW until empirically inspected on the per-cell .bdf.parquet files.

All four "NOT STATED" items above are first-class issues to resolve during the harmonization-research phase ; see bsebench-datasets/docs/research/per-dataset/empa_aurora_2025.md.

File inventory

The Zenodo record ships the dataset as a single archive :

File	Size	Format
Dataset-rocrate.zip	approx. 2.5 GB	RO-Crate, ZIP-compressed

When extracted, the RO-Crate contains, per cell (199 cells total), three files :

Per-cell file	Format	Role
empa__ccid000XXX.metadata.json	JSON-LD	BattINFO-ontologized cell metadata (chemistry, capacity, electrodes, electrolyte, assembly date, operator, etc.)
empa__ccid000XXX.bdf.csv	CSV (BDF)	Cycling time-series, BDF schema
empa__ccid000XXX.bdf.parquet	Parquet (BDF)	Cycling time-series, BDF schema, identical content to the CSV — provided for fast columnar access

Plus the standard RO-Crate ro-crate-metadata.json at the archive root describing the relationships between files, authors, and the dataset itself (Research Object Crate convention).

The Tier 1 mirror preserves the .zip byte-for-byte ; users with a columnar workflow are expected to extract and read the .bdf.parquet files directly.

Why "raw mirror" tier

BSEBench follows a dual-tier dataset strategy :

• Tier 1 (this repo) — original RO-Crate .zip, SHA-256 verified against the Zenodo record. Used for provenance verification, audit reproduction, and as a downtime-resilient backup of the canonical Zenodo source.
• Tier 2 — see bsebench-org/empa-aurora-2025 for the harmonized BSEBench-canonical Parquet consumable directly by bsebench_specs.timeseries.TimeSeriesSchema validators, with per-cell rows joined into a long format and BPX-1.1 sign convention applied.

Even though the source is already BDF, Tier 1 still exists. The reasons :

1. Single source of truth for fetching — independent of Zenodo uptime and Zenodo's "latest version" semantics. Zenodo records are versioned, but our Hub mirror pins a SHA-256.
2. RO-Crate preservation — the metadata.json files contain BattINFO-ontologized cell genealogy that downstream consumers may want to query directly. Flattening to a single Parquet table loses the JSON-LD provenance graph.
3. Forensic audit — if a future BSEBench user disputes a number in Tier 2 (e.g. a sign-flip during harmonization), the Tier 1 mirror is the definitive byte-level reference.
4. Format evolution — the BDF spec is at an early stage (v1 announced December 2025) and may extend ; preserving the original .bdf.csv / .bdf.parquet lets us re-harmonize if BSEBench's target schema later widens.

For most filter benchmarking work, you want Tier 2.

Original source

• Canonical Zenodo URL (citation only ; not a download endpoint consumed by BSEBench tooling) : https://zenodo.org/records/15481956
• Canonical DOI : 10.5281/zenodo.15481956
• Zenodo publication date : 1 July 2025 (created 30 June 2025, modified 4 July 2025).
• Version : v1.

The Zenodo record is the authoritative origin. BSEBench's HuggingFace mirror is the single source of truth for fetching in BSEBench pipelines, and is kept in lockstep with Zenodo via the manifest's SHA-256 entries. If Zenodo ever publishes a v2, BSEBench will mint a new dataset repo (e.g. empa-aurora-2025-v2-raw) rather than silently update.

License

Distributed under the Creative Commons Attribution 4.0 International license (CC-BY-4.0) — the same license as the Zenodo source. Verbatim from the Zenodo record :

"Creative Commons Attribution 4.0 International — The Creative Commons Attribution license allows re-distribution and re-use of a licensed work on the condition that the creator is appropriately credited."

You may redistribute, modify, and use commercially, provided you attribute the original authors as listed above and indicate any modifications.

How to use

from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    "bsebench-org/empa-aurora-2025-raw",
    repo_type="dataset",
)
# local_dir contains Dataset-rocrate.zip, SHA-256 verified vs Zenodo.

# Extract per-cell BDF parquet files :
import zipfile, pathlib
zip_path = pathlib.Path(local_dir) / "Dataset-rocrate.zip"
with zipfile.ZipFile(zip_path) as z:
    z.extractall(pathlib.Path(local_dir) / "extracted")

# Read one cell's BDF time-series :
import pandas as pd
df = pd.read_parquet(
    pathlib.Path(local_dir) / "extracted" /
    "empa__ccid000001.bdf.parquet"
)
print(df.columns)
# expected mandatory BDF columns : test_time_second, voltage_volt,
# current_ampere ; plus optional cycle_count, step_index, capacity
# columns, T1-T5 temperatures, etc.

For the BSEBench-harmonized Parquet view with a single canonical schema across all cells and all datasets in the BSEBench corpus, prefer the Tier 2 repository : bsebench-org/empa-aurora-2025.

Citation

Cite the original Svaluto-Ferro et al. 2025 paper and the Zenodo dataset record (BibTeX above) — not BSEBench's mirror. BSEBench's contribution is hosting + harmonization, not the data itself.

If you also use BSEBench tooling, additionally cite :

@misc{bsebench2026,
  author = {Akir, Oussama and BSEBench Contributors},
  title  = {{BSEBench}: an open-source benchmark for battery
            state-estimation filters},
  year   = {2026},
  url    = {https://bsebench.org}
}

Provenance manifest

A machine-readable manifest validating this dataset's metadata against the bsebench-dataset-manifest/v1 schema lives at bsebench-org/bsebench-datasets/manifests/empa_aurora_2025.yaml.

Including : source.canonical_url, source.canonical_doi, the RO-Crate ZIP's SHA-256 + size, license, citation BibTeX, and the expected per-cell file count (199 × 3 = 597 inner files plus the RO-Crate metadata).