Choose a version to compare

  • ABC-DiT XL Lego Three All-data Fine-tune (edisonz2103)
  • ABC-DiT XL 200k

ABC-DiT diffusion policies for bimanual manipulation and research derivatives. The base project is a collaboration including Amazon FAR. Official base weights and community fine-tunes are separate versions; family developers do not establish the affiliation or endorsement of community publishers. Base-project authors list UC Berkeley,MIT,Amazon FAR,XDOF and CMU; affiliation links preserve provenance without asserting ownership.

Overview

Model type
Action policy
Listed versions
ABC-DiT XL Lego Three All-data Fine-tune (edisonz2103) / ABC-DiT XL 200k

Papers & reports (1)

Scalable Behavior Cloning with Open Data, Training, and Evaluation

Paper available

arXiv: 2606.27375DOI: 10.48550/arxiv.2606.27375
Authors & affiliations (18)

Author order: Original order

  1. Arthur Allshire

    UC Berkeley

  2. Himanshu Gaurav Singh

    UC Berkeley

  3. Ritvik Singh

    UC Berkeley

  4. Adam Rashid

    MIT

  5. Hongsuk Choi

    UC Berkeley

  6. David McAllister

    UC Berkeley

  7. Justin Yu

    UC Berkeley · XDOF

  8. Yiyuan Chen

    XDOF

  9. Huang Huang

    XDOF

First published
2026-06-25
Latest paper revision
v1 · 2026-06-25
Venue
Not provided

Covered model versions

  • abc · ABC-DiT XL 200k

    Official released checkpoint; the publisher explicitly cites this model paper. Exact checkpoint-specific performance is not inferred.

Paper ↗PDF ↗

Official resources

Submit a source or correction