Datasets:

Berkeley-ICON-Lab
/

ALTER-data

The pretraining selection contains 400 demonstrations: 200 place-return and 200 wipe. Adaptation includes 60 selected multi-arm demonstrations and 60 distilled single-arm rollouts at the largest budget. Nested manifests retain 20+20, 40+40, and 60+60 membership, mode grouping and ordering. FT-multi omits the single-arm adaptation domain. Counts exclude base-policy pretraining.

Data and provenance retain distinct namespaces. Portable manifests use logical artifact references; materialization checks checksums and preserves list order. Do not rerank examples from relocated absolute paths. Original private records remain retained separately, and exports record original and export hashes.

Use only with the matching model/preprocessing, camera and action conventions in the release source. Data are trusted Python pickle artifacts, not a safe format for arbitrary untrusted downloads. Hardware recordings are released separately under hardware/v1/; scored physical-trial evidence is not included. Original ALTER demonstrations, replay and accompanying original records are released under CC BY 4.0; see LICENSE and NOTICE. Third-party software, assets and the paper retain their separate terms.

The provenance bundle contains selected evaluation/selection evidence, not new experimental results. The original base-training contract is absent, and the FT-mixed selection-panel wording requires author review. No complete paper retraining or full evaluation rerun is claimed by preparation tests.

See USAGE.md for downloads and the accompanying code release status.

Paper: Residual Denoising Enables Sample-Efficient Multi-Agent Coordination on Demand.

Hardware v1

Selected hardware models, source data and exact prepared caches are now available under hardware/v1. See the hardware download guide for the separate pinned manifest, selective downloads and offline validation. Simulation payloads and their pinned revisions remain unchanged.