JiajunLI/roboimi
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(vla): add SmolVLA conditioning and experiment artifacts

Browse Source
This commit is contained in:
Logic
2026-07-31 10:11:04 +08:00
parent acbd7c605a
commit 5ae9f5fa48
175 changed files with 317471 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files
workspace/inputs/README.md
+19
View File
@@ -0,0 +1,19 @@
# Inputs
The paper-orchestra pipeline expects the four files below in this
directory before you start. Optional figures go in `figures/`.
## Required
- `idea.md` — Idea Summary (Sparse or Dense; see io-contract.md)
- `experimental_log.md` — Setup, raw numeric data, qualitative observations
- `template.tex` — LaTeX template for the target conference
- `conference_guidelines.md` — Page limit, mandatory sections, formatting rules
## Optional
- `figures/` — Pre-existing figures (PNG/PDF). If empty, the
plotting agent generates everything from scratch.
See `skills/paper-orchestra/references/io-contract.md` in the repo for the
exact schemas of each file.
workspace/inputs/conference_guidelines.md
+151
View File
@@ -0,0 +1,151 @@
The Conference on Robot Learning (CoRL) is an annual international conference aiming to bring together the robotics and machine learning research communities. It focuses on the increasingly important role of learning in robotics and its interaction with other areas of robotics. The aim of CoRL 2026 is to publish significant original research at the intersection of robotics and machine learning. CoRL is a selective, single-track international conference addressing theory and practice of machine learning for robots. CoRL welcomes papers in areas such as:
Learning representations for robotic perception and control
Learning robot foundation models or general-purpose knowledge systems for robotics
Imitation learning for robotics, e.g. by behavioral cloning and/or inverse reinforcement learning
Reinforcement learning for control of physical robots
Model-based and model-free learning for robotic control and decision-making
Combination of learning- and planning-based approaches in robotics
Probabilistic learning and representation of uncertainty in robotics
Automatic robotic data generation for learning methods in robotics
Learning for Robot Task and Motion Planning
Learning for multimodal robot perception, sensor fusion, and robot vision
Learning for human-robot interaction and robot instruction by natural language, gestures as well as alternative devices
Learning for hardware design and optimization
Learning approaches to robot safety and alignment; and safety approaches applicable to learning-based robotic systems
Applications of robot learning in robot manipulation, navigation, locomotion, driving, flight, and other areas of robotics
Robot systems, hardware, and sensors for learning and data-driven approaches
Theoretical foundations of robot learning, including generalization theory and uncertainty quantification
Video models and latent world models for robot learning
Benchmarks and datasets for robot learning
Submissions should focus on a core robotics problem and demonstrate the relevance of proposed models, algorithms, datasets, and benchmarks to robotics. Authors are encouraged to report real-robot experiments or provide convincing evidence that simulation experiments are transferable to real robots. Submissions without a robotics focus will be returned without review.
All submissions should include a Limitations section, explicitly describing limiting assumptions, failure modes, and other limitations of the results and experiments and how these might be addressed in the future.
Authors are also encouraged to submit video, code and data as supplementary materials.
For paper submission format and timeline, see: Instruction for Authors
New This Year (details see below)
Paper Submission Requirements and Instructions
Submission Policy
Reciprocal Reviewing Requirement
Rebuttal Process
Concurrent/contemporaneous Work
Camera Ready Paper Submission
Key Dates
New This Year (details see below)
An additional 9th page for camera-ready submissions to incorporate reviewer feedback (initial submission is 8 pages).
Paper Submission Requirements and Instructions
CoRL is double-blind, which means all papers must be anonymized. The accepted papers and reviews will be publicly accessible and de-anonymized after decisions are announced. Our aim is to have at least two reviewers per paper. Accepted papers will appear in the Proceedings of Machine Learning Research (formerly JMLR Workshop and Conference Proceedings).
Please keep in mind that the deadlines are final, and we cannot make any accommodations for missing the abstract deadline or paper deadline. Authors may update the title and abstract on OpenReview after the abstract submission deadline. However, authors cannot be added or removed after this deadline, as this information is required to facilitate reviewer assignments. If you need to request an exception, please email our OpenReview Chair Carlo Sferrazza (csferrazza@austin.utexas.edu).
Papers may be submitted through OpenReview via the link here (https://openreview.net/group?id=robot-learning.org/CoRL/2026/Conference). All submissions should comply with the format and length indicated below:
Page limit will be 8 pages for the main paper. Acknowledgments, References, and Appendix (optional) will not count towards the page limit. Please note that reviewers are not required to read the Appendix. The Appendix can be included with the main paper or with the Supplement.
Please use the LaTeX template: here (https://drive.google.com/file/d/1R9irIW0ImqeDHh5g8Ukm2XGy91sWxOsQ/view?usp=drive_link ).
Authors are encouraged to submit a supplementary file containing further details, which the reviewers may decide to consult.
Authors are highly encouraged to submit a video not exceeding 250 MB (strict) and not more than 3 minutes in length (suggested), providing an overview of the work.
All supplementary materials will be submitted through OpenReview as a single zip file.
All accepted papers will be presented in poster sessions, while selected papers will be invited for an oral spotlight presentation.
All submissions should include a Limitations section (counted toward the 8-page limit), explicitly describing limiting assumptions, failure modes, and other limitations of the results and experiments, and how these might be addressed in the future.
Authors are strongly encouraged to include all relevant details in the main paper and the appendix to ensure that future researchers can reproduce the methodology and results.
Submission Policy
We will not accept papers that are identical or substantially similar to papers that have previously been published or accepted for publication in an archival venue, nor papers submitted in parallel to other conferences or archival venues. Archival venues include conferences and journals with formally published proceedings, but do not include non-archival workshops. Submission is permitted for papers that have previously appeared only as a technical report, e.g., in arXiv.
Dual submission policy: Dual submission is not allowed by default. Submissions that are simultaneously under review at another archival venue will be desk-rejected without review. However, we recognize that some conferences may have decision timelines that slightly overlap with our review period (e.g., ICML decisions on May 1st). In such cases, authors must notify the Publication Chair before submission and clearly indicate the overlapping venue and relevant dates. Approval for such exceptions will be granted on a case-by-case basis. Failure to disclose dual submission may result in rejection and may be reported to both venues.
Generative AI policies: We understand that generative AI is a useful tool in the paper writing process. In order to ensure that the use of generative AI does not pose an undue burden on the review process, we are implementing the following policies this year:
Papers that contain citations to references that do not exist will be desk-rejected after review by the Program Committee.
Authors must take full responsibility for the content in submitted papers. LLMs or other generative AI tools do not qualify for authorship. During the paper submission process, authors will be asked to briefly disclose how generative AI tools were used for research and paper writing.
Authors who submit multiple versions of substantially the same paper (e.g., different papers with the same method, but slightly different ablations, baselines, or algorithmic variations) will face desk rejection of all papers they have submitted, and potentially also be prevented from submitting papers to future iterations of the conference. We are aware that generative AI tools have been used for this purpose in recent conferences, and we will implement checks to flag such papers.
With these policies, we aim to strike a balance between beneficial uses of generative AI (e.g., help with paper editing, figures, literature review, or brainstorming) and uses that degrade the overall quality and integrity of the peer-review process.
Reciprocal Reviewing Requirement
Qualified authors are required to review for CoRL, according to the policies below.
All submissions must include at least one author who agrees to serve as a CoRL reviewer. They are qualified if they have at least one accepted publication at a previous robotics (CoRL/ ICRA/ IROS/ RSS) or learning conference (ICLR/ NeurIPS/ ICML) or equivalent journal. If none of the authors are qualified under this definition or all authors are exempt (e.g., serving as AC or SAC), then they are exempt from this requirement. The abstract submission form will allow submitters to designate an author to fulfill this requirement or to indicate that the submission is exempt from the requirement.
Additionally, every author with 3 or more submissions must agree to serve as a reviewer. Authors are exempt from the review requirement if they serve as an AC, an SAC, or another organizing chair for CoRL 2026.
Submissions that do not meet this reciprocal review requirement may be desk-rejected. Additionally, reviewers who fail to adequately participate in the review process (e.g., not submitting reviews on time or submitting highly insufficient or inappropriate reviews) may have their own submissions desk rejected. The program chairs may grant exceptions on a case-by-case basis.
Rebuttal Process
In order to focus reviewer effort, papers that receive all reviews in the “reject” range in the initial round will not proceed to the rebuttal phase.
Authors of all other papers are invited to submit a 1-page rebuttal in pdf by the rebuttal deadline. You will not be able to give individual responses to the reviewer on OpenReview, nor will you be able to update the paper during the rebuttal process. The rebuttal should be focused on addressing any factual errors in the review.
All rebuttals will be reviewed by the reviewer, the original AC, and, if applicable, external reviewers. The AC reserves the right to invite new reviewers if needed. The results of the current review(s) will be shared with the new reviewers in such cases.
Concurrent/contemporaneous Work
We consider papers contemporaneous if they are published within the last 4 months, so authors do not need to compare their own work to that paper. Authors are encouraged to cite and discuss all relevant papers, but they may be excused for not knowing about papers not published in peer-reviewed conference proceedings or journals, which include papers exclusively available on arXiv.
Camera Ready Paper Submission
New this year — additional page for incorporating reviewer feedback
Congratulations again on getting your paper accepted to CoRL 2026! The camera-ready version of your paper is due by Oct 12, 2026 (23:59 Anywhere on Earth). Please submit your camera-ready PDF and signed permission form through OpenReview.
Here are the instructions for preparing and submitting your camera-ready paper:
1. The camera-ready paper should follow the template: 8-page main text + acknowledgment + references + appendix (optional). Note that this year, we are allowing an additional page to accommodate feedback from the review process. The appendix should be included at the end of the camera-ready PDF, rather than as a separate file. The template is available here. Please make sure to use “\usepackage[final]{corl_2026}.
2. One important note: PMLR does not allow videos to be submitted as supplementary material. If you have videos, code, datasets, and other supplementary materials, please host them on your own (e.g., YouTube, GitHub, etc). You should provide, in the main text of your paper, a link to this material or a link to your project website.
3. Please sign the permission form for publication in PMLR (form available here). Please rename the pdf to <paperid_firstname_lastname>.pdf.
Please ensure:
1. Paper length: 9 pages main text + Acknowledgement + References + Appendix (optional)
2. The author list is NOT anonymous.
3. Footer on the 1st page: "10th Conference on Robot Learning (CoRL 2026), Austin, Texas, USA."
4. Text density, fonts, and spacing should be the same as the provided template on the website.
5. The margin should be the same as the provided template on the website.
6. The OpenReview title is the same as the PDF title.
Please contact Yoonchang Sung (yoonchang.sung@ntu.edu.sg) if you have any questions.
workspace/inputs/corl_2026.sty
+472
View File
@@ -0,0 +1,472 @@
% File: corl_2026.sty
%
% Latex templates for the Conference on Robot Learning (CoRL)
%
% This template is heavily inspired by the NeurIPS, ICML, ICLR and IEEE Transactions latex templates.
% Hence we would like to thank: Roman Garnett and the previous mantainers of the NIPS style, Percy Liang and the previous mantainers of the ICML style, Hugo Larochelle for the ICLR style, and Michael Shell for the IEEE Transactions style.
%
% History:
% 2017/04/16 - First revision by Roberto Calandra (roberto.calandra@berkeley.edu).
% Main changes:
% - The abstract is more compact compared to NeurIPS/ICML
% - References are by default using natbib with squared numbers (e.g., [1])
% - DOI fields from the bibtex are automatically converted to hyperlinks
% to the corresponding page
% - acknowledgments are now a command, and the corresponding subsubsection is
% automatically included only in the final version
% 2017/06/12 - Modified to use corlabbrvnat.bst, which order the reference by order of appearance in the paper
% 2017/06/13 - fixed typo
% 2018/05/09 - Slightly modified for CoRL 2018 by Jun Morimoto (xmorimo@atr.jp)
% 2019/01/28 - Slightly modified for CoRL 2019 by Jun Nakanishi (jnakanis@meijo-u.ac.jp)
% 2020/02/02 - Slightly modified for CoRL 2020 by Cynthia Matuszek (cmat@umbc.edu)
% 2020/08/19 - Added preprint option by Roberto Calandra (rcalandra@fb.com)
% 2021/05/06 - Slightly modified for CoRL 2021 by Gerhard Neumann (gerhard.neumann@kit.edu)
% 2022/03/09 - Slightly modified for CoRL 2022 by Minas Liarokapis (minas.liarokapis@auckland.ac.nz)
% 2022/03/06 - Slightly modified for CoRL 2023 by Marc Toussaint (toussaint@tu-berlin.de)
% 2024/03/26 - Slightly modified for CoRL 2024 by David Held (dheld@andrew.cmu.edu)
% 2026/01/10 - Slightly modified for CoRL 2026 by Yoonchang Sung (yoonchang.sung@ntu.edu.sg)
%
% TODO: nohyperref is not working at the moment
%
\NeedsTeXFormat{LaTeX2e}
% Content to be changed from year to year
\ProvidesPackage{corl_2026}[2026/08/15 CORL2026 submission/preprint/camera-ready style file]
\newcommand{\@conferenceordinal}{10th}
\newcommand{\@conferenceyear}{2026}
\newcommand{\@conferencelocation}{Austin TX, USA}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Accepted options: [final,preprint,nonatbib,nohyperref]
% Declare the final option, which creates camera-ready copy
\newif\if@conferencefinal\@conferencefinalfalse
\DeclareOption{final}{
\@conferencefinaltrue
}
% Declare the preprint option, which creates a camera-ready copy without the corl footnote
\newif\if@preprinttype\@preprinttypefalse
\DeclareOption{preprint}{
\@preprinttypetrue
}
% The natbib package is loaded by default. Declaring the nonatbib option, does not load natbib in case of package clash (users can pass options to natbib via \PassOptionsToPackage)
\newif\if@natbib\@natbibtrue
\DeclareOption{nonatbib}{
\@natbibfalse
}
% The hyperref package is loaded by default. Declaring the nohyperref option, does not load the hyperref.
\DeclareOption{nohyperref}{%
\gdef\nohyperref{1}
}
% Activate the options
\ProcessOptions\relax
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Required packages:
\RequirePackage{lineno}
\RequirePackage{color}
% Load natbib unless told otherwise
\if@natbib
\RequirePackage[square,numbers]{natbib}
\bibliographystyle{corlabbrvnat}
\fi
% set page geometry
\RequirePackage{hyperref} % hyperlinks
\RequirePackage[verbose=true,letterpaper]{geometry}
\AtBeginDocument{
\newgeometry{
textheight=9in,
textwidth=5.5in,
top=1in,
headheight=12pt,
headsep=25pt,
footskip=30pt
}
\@ifpackageloaded{fullpage}
{\PackageWarning{corl_2026}{fullpage package not allowed! Overwriting formatting.}}
{}
}
\ifdefined\nohyperref\else\ifdefined\hypersetup
\definecolor{mydarkblue}{rgb}{0,0.08,0.45}
\hypersetup{ %
pdftitle={},
pdfauthor={},
pdfsubject={Proceedings of the \@conferenceordinal\/ Conference on Robot Learning (CoRL \@conferenceyear)},
pdfkeywords={},
pdfborder=0 0 0,
pdfpagemode=UseNone,
colorlinks=true,
linkcolor=mydarkblue,
citecolor=mydarkblue,
filecolor=mydarkblue,
urlcolor=mydarkblue,
pdfview=FitH}
\ifdefined\isaccepted \else
\hypersetup{pdfauthor={Anonymous Submission}}
\fi
\fi\fi
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% fonts
\renewcommand{\rmdefault}{ptm}
\renewcommand{\sfdefault}{phv}
% Create acknowledgments -- only if the option 'final' is activated
\providecommand{\acknowledgments}{}
\renewcommand{\acknowledgments}[1]{%
\if@conferencefinal%
\subsubsection*{Acknowledgments} #1
\fi
\if@preprinttype%
\subsubsection*{Acknowledgments} #1
\fi
}
% handle tweaks for camera-ready copy vs. submission copy
\if@conferencefinal
\newcommand{\@noticestring}{%
\@conferenceordinal\/ Conference on Robot Learning
(CoRL \@conferenceyear), \@conferencelocation.%
}
\else
\if@preprinttype
\newcommand{\@noticestring}{%
% Nothing here.
}
\else
\newcommand{\@noticestring}{%
Submitted to the \@conferenceordinal\/ Conference on Robot Learning (CoRL \@conferenceyear). Do not distribute.%
}
% line numbers for submission
\linenumbers
% fix incompatibilities between lineno and amsmath, if required, by
% transparently wrapping linenomath environments around amsmath
% environments
\AtBeginDocument{%
\@ifpackageloaded{amsmath}{%
\newcommand*\patchAmsMathEnvironmentForLineno[1]{%
\expandafter\let\csname old#1\expandafter\endcsname\csname #1\endcsname
\expandafter\let\csname oldend#1\expandafter\endcsname\csname end#1\endcsname
\renewenvironment{#1}%
{\linenomath\csname old#1\endcsname}%
{\csname oldend#1\endcsname\endlinenomath}%
}%
\newcommand*\patchBothAmsMathEnvironmentsForLineno[1]{%
\patchAmsMathEnvironmentForLineno{#1}%
\patchAmsMathEnvironmentForLineno{#1*}%
}%
\patchBothAmsMathEnvironmentsForLineno{equation}%
\patchBothAmsMathEnvironmentsForLineno{align}%
\patchBothAmsMathEnvironmentsForLineno{flalign}%
\patchBothAmsMathEnvironmentsForLineno{alignat}%
\patchBothAmsMathEnvironmentsForLineno{gather}%
\patchBothAmsMathEnvironmentsForLineno{multline}%
}{}
}
\fi
\fi
% The DOI will now automatically generate a URL link. Pretty cool!
% + Fix for hyperref and DOI (https://www.tug.org/pipermail/tex-live/2012-August/032161.html)
%-----------------------------
\makeatletter
\providecommand{\doi}[1]{%
\begingroup
\let\bibinfo\@secondoftwo
\urlstyle{rm}%
\href{http://dx.doi.org/#1}{%
doi:\discretionary{}{}{}%
\nolinkurl{#1}%
}%
\endgroup
}
% \makeatother
% %-----------------------------
\widowpenalty=10000
\clubpenalty=10000
\flushbottom
\sloppy
% font sizes with reduced leading
\renewcommand{\normalsize}{%
\@setfontsize\normalsize\@xpt\@xipt
\abovedisplayskip 7\p@ \@plus 2\p@ \@minus 5\p@
\abovedisplayshortskip \z@ \@plus 3\p@
\belowdisplayskip \abovedisplayskip
\belowdisplayshortskip 4\p@ \@plus 3\p@ \@minus 3\p@
}
\normalsize
\renewcommand{\small}{%
\@setfontsize\small\@ixpt\@xpt
\abovedisplayskip 6\p@ \@plus 1.5\p@ \@minus 4\p@
\abovedisplayshortskip \z@ \@plus 2\p@
\belowdisplayskip \abovedisplayskip
\belowdisplayshortskip 3\p@ \@plus 2\p@ \@minus 2\p@
}
\renewcommand{\footnotesize}{\@setfontsize\footnotesize\@ixpt\@xpt}
\renewcommand{\scriptsize}{\@setfontsize\scriptsize\@viipt\@viiipt}
\renewcommand{\tiny}{\@setfontsize\tiny\@vipt\@viipt}
\renewcommand{\large}{\@setfontsize\large\@xiipt{14}}
\renewcommand{\Large}{\@setfontsize\Large\@xivpt{16}}
\renewcommand{\LARGE}{\@setfontsize\LARGE\@xviipt{20}}
\renewcommand{\huge}{\@setfontsize\huge\@xxpt{23}}
\renewcommand{\Huge}{\@setfontsize\Huge\@xxvpt{28}}
% sections with less space
\providecommand{\section}{}
\renewcommand{\section}{%
\@startsection{section}{1}{\z@}%
{-2.0ex \@plus -0.5ex \@minus -0.2ex}%
{ 1.5ex \@plus 0.3ex \@minus 0.2ex}%
{\large\bf\raggedright}%
}
\providecommand{\subsection}{}
\renewcommand{\subsection}{%
\@startsection{subsection}{2}{\z@}%
{-1.8ex \@plus -0.5ex \@minus -0.2ex}%
{ 0.8ex \@plus 0.2ex}%
{\normalsize\bf\raggedright}%
}
\providecommand{\subsubsection}{}
\renewcommand{\subsubsection}{%
\@startsection{subsubsection}{3}{\z@}%
{-1.5ex \@plus -0.5ex \@minus -0.2ex}%
{ 0.5ex \@plus 0.2ex}%
{\normalsize\bf\raggedright}%
}
\providecommand{\paragraph}{}
\renewcommand{\paragraph}{%
\@startsection{paragraph}{4}{\z@}%
{1.5ex \@plus 0.5ex \@minus 0.2ex}%
{-1em}%
{\normalsize\bf}%
}
\providecommand{\subparagraph}{}
\renewcommand{\subparagraph}{%
\@startsection{subparagraph}{5}{\z@}%
{1.5ex \@plus 0.5ex \@minus 0.2ex}%
{-1em}%
{\normalsize\bf}%
}
\providecommand{\subsubsubsection}{}
\renewcommand{\subsubsubsection}{%
\vskip5pt{\noindent\normalsize\rm\raggedright}%
}
% float placement
\renewcommand{\topfraction }{0.85}
\renewcommand{\bottomfraction }{0.4}
\renewcommand{\textfraction }{0.1}
\renewcommand{\floatpagefraction}{0.7}
\newlength{\@nipsabovecaptionskip}\setlength{\@nipsabovecaptionskip}{7\p@}
\newlength{\@nipsbelowcaptionskip}\setlength{\@nipsbelowcaptionskip}{\z@}
\setlength{\abovecaptionskip}{\@nipsabovecaptionskip}
\setlength{\belowcaptionskip}{\@nipsbelowcaptionskip}
% swap above/belowcaptionskip lengths for tables
\renewenvironment{table}
{\setlength{\abovecaptionskip}{\@nipsbelowcaptionskip}%
\setlength{\belowcaptionskip}{\@nipsabovecaptionskip}%
\@float{table}}
{\end@float}
% footnote formatting
\setlength{\footnotesep }{6.65\p@}
\setlength{\skip\footins}{9\p@ \@plus 4\p@ \@minus 2\p@}
\renewcommand{\footnoterule}{\kern-3\p@ \hrule width 12pc \kern 2.6\p@}
\setcounter{footnote}{0}
% paragraph formatting
\setlength{\parindent}{\z@}
\setlength{\parskip }{5.5\p@}
% list formatting
\setlength{\topsep }{4\p@ \@plus 1\p@ \@minus 2\p@}
\setlength{\partopsep }{1\p@ \@plus 0.5\p@ \@minus 0.5\p@}
\setlength{\itemsep }{2\p@ \@plus 1\p@ \@minus 0.5\p@}
\setlength{\parsep }{2\p@ \@plus 1\p@ \@minus 0.5\p@}
\setlength{\leftmargin }{3pc}
\setlength{\leftmargini }{\leftmargin}
\setlength{\leftmarginii }{2em}
\setlength{\leftmarginiii}{1.5em}
\setlength{\leftmarginiv }{1.0em}
\setlength{\leftmarginv }{0.5em}
\def\@listi {\leftmargin\leftmargini}
\def\@listii {\leftmargin\leftmarginii
\labelwidth\leftmarginii
\advance\labelwidth-\labelsep
\topsep 2\p@ \@plus 1\p@ \@minus 0.5\p@
\parsep 1\p@ \@plus 0.5\p@ \@minus 0.5\p@
\itemsep \parsep}
\def\@listiii{\leftmargin\leftmarginiii
\labelwidth\leftmarginiii
\advance\labelwidth-\labelsep
\topsep 1\p@ \@plus 0.5\p@ \@minus 0.5\p@
\parsep \z@
\partopsep 0.5\p@ \@plus 0\p@ \@minus 0.5\p@
\itemsep \topsep}
\def\@listiv {\leftmargin\leftmarginiv
\labelwidth\leftmarginiv
\advance\labelwidth-\labelsep}
\def\@listv {\leftmargin\leftmarginv
\labelwidth\leftmarginv
\advance\labelwidth-\labelsep}
\def\@listvi {\leftmargin\leftmarginvi
\labelwidth\leftmarginvi
\advance\labelwidth-\labelsep}
% create title
\providecommand{\maketitle}{}
\renewcommand{\maketitle}{%
\par
\begingroup
\renewcommand{\thefootnote}{\fnsymbol{footnote}}
% for perfect author name centering
\renewcommand{\@makefnmark}{\hbox to \z@{$^{\@thefnmark}$\hss}}
% The footnote-mark was overlapping the footnote-text,
% added the following to fix this problem (MK)
\long\def\@makefntext##1{%
\parindent 1em\noindent
\hbox to 1.8em{\hss $\m@th ^{\@thefnmark}$}##1
}
\thispagestyle{empty}
\@maketitle
\@thanks
\@notice
\endgroup
\let\maketitle\relax
\let\thanks\relax
}
% rules for title box at top of first page
\newcommand{\@toptitlebar}{
\hrule height 4\p@
\vskip 0.25in
\vskip -\parskip%
}
\newcommand{\@bottomtitlebar}{
\vskip 0.29in
\vskip -\parskip
\hrule height 1\p@
\vskip 0.09in%
}
%% keywords as first class citizens
\def\keywords#1{%
% \ifdefined\isaccepted \else
% \par {\bf Keywords:} #1%
% \fi
% \ifdefined\nohyperref\else\ifdefined\hypersetup
% \hypersetup{pdfkeywords={#1}}
% \fi\fi
\ifdefined\isaccepted \else
\begin{quote}
\textbf{Keywords:} #1%
\end{quote}
\fi
\ifdefined\nohyperref\else\ifdefined\hypersetup
\hypersetup{pdfkeywords={#1}}
\fi\fi
}
% create title (includes both anonymized and non-anonymized versions)
\providecommand{\@maketitle}{}
\renewcommand{\@maketitle}{%
\vbox{%
\hsize\textwidth
\linewidth\hsize
\vskip 0.1in
% \@toptitlebar
\centering
{\LARGE\bf \@title\par}
% \@bottomtitlebar
\if@conferencefinal
\def\And{%
\end{tabular}\hfil\linebreak[0]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\def\AND{%
\end{tabular}\hfil\linebreak[4]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\@author\end{tabular}%
\else
\if@preprinttype
\def\And{%
\end{tabular}\hfil\linebreak[0]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\def\AND{%
\end{tabular}\hfil\linebreak[4]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\@author\end{tabular}%
\else
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}
Anonymous Author(s) \\
Affiliation \\
Address \\
\texttt{email} \\
\end{tabular}%
\fi
\fi
\vskip 0.3in \@minus 0.1in
}
}
% add conference notice to bottom of first page
\newcommand{\ftype@noticebox}{8}
\newcommand{\@notice}{%
% give a bit of extra room back to authors on first page
\enlargethispage{2\baselineskip}%
\@float{noticebox}[b]%
\footnotesize\@noticestring%
\end@float%
}
% abstract styling
\renewenvironment{abstract}%
{%
% \vskip 0.075in%
% \centerline%
% {\large\bf Abstract}%
% \vspace{0.5ex}%
\begin{quote}%
\textbf{Abstract:}%
}
{
\par%
\vskip 1ex%
% \ifdefined\keywords
% \textbf{Keywords:}%
% \@keywords%
% \else
% \fi
\end{quote}%
}
\endinput
workspace/inputs/corl_2026_template_submission.zip
BIN
View File
Binary file not shown.
workspace/inputs/corlabbrvnat.bst
+1442
View File
File diff suppressed because it is too large Load Diff
workspace/inputs/example.bib
+20
View File
@@ -0,0 +1,20 @@
% This file was created with JabRef 2.10.
% Encoding: UTF-8
@Article{Gauss1857,
Title = {Theory of the motion of the heavenly bodies moving about the sun in conic sections},
Author = {Carl Friedrich Gauss and Charles Henry Davis},
Journal = {Gauss's Theoria Motus},
Year = {1857},
Number = {1},
Pages = {5--23},
Volume = {76}
}
@Book{Lagrange1788,
title = {M{\'e}canique Analytique},
author = {Joseph-Louis Lagrange},
publisher = {Desaint, Paris},
year = {1788}
}
workspace/inputs/experimental_log.md
+92
View File
@@ -0,0 +1,92 @@
# Experimental Log: iMF-AttnRes for Fast Vision-Language-Action Imitation Learning in RoboIMI
## 1. Experimental Setup
We evaluate imitation-learning policies in the RoboIMI simulated manipulation benchmark. The study focuses on two environments:
- **socket_peg / socket-insert**: a socket peg insertion task.
- **sim_transfer / object transfer**: a block/object transfer manipulation task.
The central method is an iMF-AttnRes VLA policy that combines improved Mean Flow (iMF) training with Attention Residuals (AttnRes). The comparison set includes diffusion-policy-style DiT baselines, ACT, SmolVLA, and several ablations of the iMF-AttnRes architecture and horizon/execution settings. Each reported result is based on **100 simulation rollouts** unless otherwise stated.
Reported metrics:
- `avg_reward`: mean cumulative episode reward over 100 rollouts.
- `median_reward`: median cumulative episode reward.
- `avg_max_reward`: mean maximum per-episode reward statistic reported by the rollout evaluator.
- `max_reward`: highest cumulative episode reward among 100 rollouts.
- `nonzero_reward`: number of episodes with nonzero reward.
- `success_like`: number of episodes whose max reward crosses the environment threshold (`max_reward > 4` for socket-insert; `max_reward >= 4` for sim_transfer, as recorded in the logs).
- `avg_inference_fps`, `avg_control_fps`, and `avg_inference_time_ms`: rollout speed measurements. Some sim_transfer logs report FPS but not inference time in milliseconds.
Hardware varies across runs and is encoded in run names: 5880g0/5880g1 denotes RTX 5880 Ada runs, l20g* denotes L20 runs, and 5090 denotes RTX 5090 runs. Because hardware differs across experiments, speed comparisons should be interpreted as practical rollout measurements rather than perfectly hardware-normalized measurements.
## 2. Raw Numeric Data
### Table 1: Socket peg insertion, 100-rollout performance and speed
| run_name | method_family | horizon_or_steps | avg_reward | median_reward | avg_max_reward | max_reward | nonzero_reward | success_like | avg_inference_fps | avg_control_fps | avg_inference_time_ms |
|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
| socket-insert-imf-attnres-ph32-exec16-emb384-l12-infer1-unfreeze-step50k-roll1x10-5880g01-20260506-195806 | iMF-AttnRes | infer1, ph32, exec16, 50k | 1275.22 | 1490.5 | 3.12 | 2250.0 | 84/100 | 0/100 | 64.226 | 5.863 | 16.186 |
| socket-insert-no-pretrain-ph16-emb384-l18-infer100-unfreeze-step150k-roll1x10-5090-20260506-193753 | Diffusion-style VLA | infer100, ph16, 150k | 861.53 | 668.0 | 2.58 | 2241.0 | 97/100 | 2/100 | 2.956 | 2.591 | 338.291 |
| socket-insert-no-pretrain-ph32-exec16-emb384-l18-infer100-unfreeze-step150k-roll1x10-5880g1-20260507-092227 | Diffusion-style VLA | infer100, ph32, exec16, 150k | 1019.39 | 804.0 | 2.47 | 2385.0 | 92/100 | 4/100 | 2.516 | 1.928 | 397.912 |
| socket-insert-imf-attnres-ph32-exec16-emb384-l12-infer2-drop005-unfreeze-step150k-roll1x10-5880g01-20260508-165022 | iMF-AttnRes | infer2, ph32, exec16, 150k | 1472.62 | 1825.0 | 3.29 | 2365.0 | 83/100 | 5/100 | 71.579 | 7.181 | 14.409 |
| socket-insert-imf-attnres-ph32-exec16-emb384-l12-infer3-drop005-unfreeze-step150k-roll1x10-5880g01-20260509-172423 | iMF-AttnRes | infer3, ph32, exec16, 150k | 1513.56 | 1901.5 | 3.28 | 2285.0 | 83/100 | 3/100 | 67.891 | 7.247 | 15.122 |
| act-socket-peg-224-20260508-170237 | ACT | action chunking | 289.6 | 13.0 | 1.29 | 2091.0 | 55/100 | 1/100 | 10.5171 | 2.2782 | 100.2116 |
| smolvla_socket_peg_bs80_100k_20260511_145727 | SmolVLA | 100k | 466.16 | 106.0 | 1.72 | 4.0 | 89/100 | 0/100 | 382.53 | 16.84 | 2.614 |
### Table 2: Object transfer / sim_transfer, 100-rollout performance and speed
| run_name | method_family | horizon_or_steps | avg_reward | median_reward | avg_max_reward | max_reward | nonzero_reward | success_like | avg_inference_fps | avg_control_fps | avg_inference_time_ms |
|---|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
| diffusion_policy_native_dit_ddpm_resnet_best | Diffusion Policy native DiT + DDPM + ResNet | best checkpoint | 319.2 | 6.0 | 1.68 | 1416.0 | 55/100 | 29/100 | 32.09 | 16.70 | n/a |
| embed384_layer18_best_checkpoint | Diffusion-style VLA | emb384, layer18 | 233.52 | 0.0 | 1.12 | 1436.0 | 39/100 | 17/100 | 1.859 | 1.649 | n/a |
| resnet18-multitoken-imf-emb256-l16-ph16-ex08-roll10-l20g2-20260406-112815 | iMF multi-token ResNet18 | step34999, ph16, exec08 | 260.66 | 0.0 | 1.12 | 1422.0 | 33/100 | 23/100 | 441.80 | 13.37 | n/a |
| imf-p2-full-attnres-vision-ph16-ex08-emb384-l12-b40-lr1p25e4-ms50k-l20g3-20260405-002424 | iMF full AttnRes vision | ph16, exec08, 50k | 228.42 | 0.0 | 0.94 | 1482.0 | 31/100 | 16/100 | 55.997 | 4.301 | n/a |
| imf-p1-ph16-ex16-emb384-l12-ms50k-l20g0-20260404-131223 | iMF-AttnRes DiT only | ph16, exec16, 50k | 240.64 | 0.0 | 1.02 | 1428.0 | 32/100 | 19/100 | 137.996 | 4.523 | n/a |
| imf-p1-ph32-ex08-emb384-l12-ms50k-l20g1-20260404-131223 | iMF-AttnRes DiT only | ph32, exec08, 50k | 163.28 | 0.0 | 0.76 | 1566.0 | 26/100 | 12/100 | 85.638 | 4.317 | n/a |
| imf-p1-ph32-ex32-emb384-l12-ms50k-5090-20260404-13122 | iMF-AttnRes DiT only | ph32, exec32, 50k | 260.72 | 0.0 | 1.18 | 1342.0 | 38/100 | 21/100 | 9.557 | 3.299 | n/a |
| imf-p1-ph16-ex08-emb384-l12-ms50k-5880g1-20260404-131223 | iMF-AttnRes DiT only | ph16, exec08, 50k | 229.56 | 0.0 | 1.18 | 1564.0 | 41/100 | 18/100 | 69.984 | 8.053 | n/a |
| imf-p1-ph08-ex08-emb384-l12-ms50k-5880g0-20260404-131223 | iMF-AttnRes DiT only | ph08, exec08, 50k | 237.02 | 0.0 | 1.32 | 1400.0 | 48/100 | 18/100 | 69.192 | 8.173 | n/a |
| imf-p1-ph32-ex16-emb384-l12-ms50k-l20g2-20260404-131223 | iMF-AttnRes DiT only | ph32, exec16, 50k | 49.88 | 0.0 | 0.32 | 1284.0 | 13/100 | 4/100 | 138.903 | 4.482 | n/a |
| sim-transfer-imf-attnres-ph32-exec16-emb384-l12-infer1-unfreeze-step50k-roll5x5-20260403-14130 | iMF-AttnRes | infer1, ph32, exec16, 50k | 526.22 | 86.0 | 2.14 | 1460.0 | 63/100 | 44/100 | 8.911 | 3.193 | n/a |
### Table 3: Derived comparisons used in the paper draft
| comparison | metric | numerator_run | denominator_run | numerator_value | denominator_value | derived_ratio_or_delta |
|---|---|---|---|---:|---:|---:|
| socket best iMF vs ph16 diffusion baseline | avg_reward ratio | iMF infer3 150k | diffusion ph16 infer100 150k | 1513.56 | 861.53 | 1.76x |
| socket best iMF vs ph32 diffusion baseline | avg_reward ratio | iMF infer3 150k | diffusion ph32 infer100 150k | 1513.56 | 1019.39 | 1.48x |
| socket best iMF vs ACT | avg_reward ratio | iMF infer3 150k | ACT | 1513.56 | 289.6 | 5.23x |
| socket best iMF vs SmolVLA | avg_reward ratio | iMF infer3 150k | SmolVLA | 1513.56 | 466.16 | 3.25x |
| socket iMF infer2 vs ph16 diffusion baseline | inference time reduction | iMF infer2 150k | diffusion ph16 infer100 150k | 14.409 | 338.291 | 23.48x lower latency |
| socket iMF infer3 vs ph16 diffusion baseline | inference fps ratio | iMF infer3 150k | diffusion ph16 infer100 150k | 67.891 | 2.956 | 22.97x |
| socket iMF infer2 vs ph32 diffusion baseline | inference fps ratio | iMF infer2 150k | diffusion ph32 infer100 150k | 71.579 | 2.516 | 28.45x |
| sim_transfer best iMF vs native diffusion policy | avg_reward ratio | sim-transfer iMF infer1 | native diffusion policy | 526.22 | 319.2 | 1.65x |
| sim_transfer best iMF vs native diffusion policy | success_like delta | sim-transfer iMF infer1 | native diffusion policy | 44 | 29 | +15 episodes |
| sim_transfer ResNet18 multitoken iMF vs embed384 layer18 baseline | inference fps ratio | ResNet18 multitoken iMF | embed384 layer18 | 441.80 | 1.859 | 237.65x |
## 3. Qualitative Observations
Socket peg insertion shows the clearest quality-speed tradeoff improvement. The iMF-AttnRes variants with ph32/exec16 and only 1--3 inference evaluations achieve substantially higher average and median rewards than the diffusion-style infer100 baselines, while reducing latency from hundreds of milliseconds to roughly 14--16 ms. The infer2 and infer3 variants are the strongest socket policies by average reward, with the infer3 run obtaining avg_reward 1513.56 and median_reward 1901.5. The infer2 run has the fastest iMF-AttnRes socket latency among the high-reward variants at 14.409 ms.
The socket experiments also reveal that nonzero reward alone is not a sufficient quality metric. The diffusion-style ph16 baseline has 97/100 nonzero episodes but much lower median reward than the best iMF-AttnRes runs. Conversely, iMF-AttnRes has 83/100 nonzero episodes but much higher median and average reward, suggesting that when it engages successfully with the task it produces stronger progress toward insertion.
SmolVLA is extremely fast in inference FPS and control FPS, but its reward is lower than iMF-AttnRes in these experiments. ACT is slower than SmolVLA and underperforms iMF-AttnRes in average and median reward.
In sim_transfer, the best recorded iMF-AttnRes run, `sim-transfer-imf-attnres-ph32-exec16-emb384-l12-infer1-unfreeze-step50k-roll5x5-20260403-14130`, reaches avg_reward 526.22 and 44/100 success-like episodes, exceeding the native diffusion-policy DiT/DDPM/ResNet baseline with avg_reward 319.2 and 29/100 success-like episodes. However, several iMF ablations underperform the native diffusion-policy baseline on average reward, indicating that the method is sensitive to horizon/execution choices, vision-token design, and whether AttnRes is applied only in the policy transformer or throughout the vision stack.
For sim_transfer, the ResNet18 multi-token iMF variant is a particularly fast policy at 441.80 inference FPS but does not match the best iMF-AttnRes reward. The full-AttnRes vision variant underperforms the more conservative DiT-only AttnRes settings, suggesting that replacing residual pathways inside the vision encoder may be harmful or require different training hyperparameters.
### Iteration History
1. Early sim_transfer experiments compared diffusion-style baselines against initial iMF-AttnRes policy variants and explored policy horizon/execution length combinations.
2. Subsequent ablations tested multi-token ResNet18 visual encoding, full AttnRes in the vision backbone, and DiT-only AttnRes. These results showed that iMF speedups are robust but reward quality is sensitive to model and horizon choices.
3. Later socket-insert experiments scaled to 100-rollout evaluation and compared iMF-AttnRes against diffusion-style VLA, ACT, and SmolVLA. These experiments provided the strongest evidence for the paper's main claim: iMF-AttnRes can substantially reduce inference latency while improving or preserving manipulation reward.
## 4. Caveats for Paper Writing
- Do not claim real-robot validation; all reported experiments are simulation rollouts.
- Hardware differs across runs, so speed measurements are useful but not fully hardware-normalized.
- Success-like episode counts use thresholds as recorded by each environment's evaluation script: socket-insert logs report `max_reward > 4`, whereas sim_transfer logs report `max_reward >= 4`.
- The reported data are sufficient for a first paper draft but should be supplemented later with controlled hardware-normalized speed measurements and more seeds if the paper is prepared for formal submission.
workspace/inputs/idea.md
+73
View File
@@ -0,0 +1,73 @@
本文的主要idea是使用imf和attnres来加速流匹配的推理速度，同时保持相近的推理质量，imf来自于计算机视觉中的he kaiming的研究Improved mean flows: On the challenges of fastforward generative models,之前是用于流匹配图像生成模型的改进，其前述相关研究如下
## MeanFlow [@geng2025meanflow]
Diffusion的生成模型是基于一步一步的方式生成的，这显然不够快，对flow matching中的核心ODE公式
$$\frac{dx_t}{dt} = v(x_t, t)$$
这里提一下在看这篇文章时突然产生的对微分算子$d$的理解，$x_t$表示的是某一时刻的具体位置，我们想知道这一时刻的瞬时速度，要根据速度的计算公式进行计算，速度的计算公式是位移时间比，但$x_t$表示的是某一时刻的具体位置，不是一段时间的位移，因此我们使用微分算子表示在$x_t$附近取很小的一段位移除以很小的一段时间，这里一定注意加入微分算子前后的含义变化，对$t$来说加入微分算子前表示的是具体的时刻$t$，而加入后表示的是一段时间$t$。如果取的足够小，通过这种方式计算出来的一小段内的速度就约等于瞬时速度，这是微分算子的作用。
回到ODE公式本身，可以看出这个公式其实计算的是每一时刻的瞬时速度，而原本的flow matching需要多步采样就来自这个时刻的瞬时性，我们得到瞬时速度后是通过一种近似采样的方式向前移动，是把这个瞬时速度视为一段时间内的平均速度，这也是为什么一开始的模型采样步数多效果就好，因为采样步数太少会导致这种估计非常不准确，使得路径偏移严重。为了解决这种问题，Rectified Flow可译为重整流直接将路径假设为一条直线，注意此处的直线是一种假设，我们只是希望模型能够学会按照这条直线去走，为了能够让模型学会尽可能的按照直线去走，我们在计算损失时让模型预测的速度尽可能接近直线路径的速率，另一方面还通过重整修正，让训练好的但预测路径还不那么直的模型继续学习更直的路线（通过让第一次训练好的模型从任意起点出发得到一个目标点后学习二者间的直线来重整）。
尽管如此，实际实验中发现，模型学习的路线仍然不可能是我们预期中的直线，因此，这种假设可能本身存在一定的问题，真实的分布转移路径可能并不直，那是否有办法让模型可以学习到真实的平均速度而不是基于路径直线这一假设的平均速度呢？
回到平均速度的定义，平均速度是位移除以时间，那么任意两个时间点之间的平均速度是
$$\bar{v}(z_t, r, t) = \frac{1}{t-r} \int_r^t v(z_\tau, \tau)d\tau$$
我们训练模型去拟合这个平均速度，但写出损失函数后可以发现，我们并不知道这个平均速度的真实值（因为没有假设条件了，因此路径可能是任意的）。但MeanFlow提出，可以通过构建平均速度和瞬时速度的关系来获得真实的平均速度。推导如下
$$\begin{aligned} & \bar{v}(z_t, r, t) = \frac{1}{t-r} \int_r^t v(z_\tau, \tau) d\tau \\ \Leftrightarrow \quad & (t - r)\bar{v}(z_t, r, t) = \int_r^t v(z_\tau, \tau) d\tau \\ \Leftrightarrow \quad & \frac{d}{dt} \left[ (t - r)\bar{v}(z_t, r, t) \right] = \frac{d}{dt} \int_r^t v(z_\tau, \tau) d\tau \\ \Leftrightarrow \quad & \bar{v}(z_t, r, t) + (t - r)\frac{d}{dt}\bar{v}(z_t, r, t) = v(z_t, t) \\ \Leftrightarrow \quad & \bar{v}(z_t, r, t) = v(z_t, t) - (t - r)\frac{d}{dt}\bar{v}(z_t, r, t) \end{aligned} \tag{1}$$
其中
$$\frac{d}{dt}\bar{v}(z_t, r, t) = \frac{d\bar{v}}{dz} \cdot \frac{dz}{dt} + \frac{d\bar{v}}{dr} \cdot \frac{dr}{dt} + \frac{d\bar{v}}{dt} \cdot \frac{dt}{dt} = \frac{d\bar{v}}{dz} \cdot v(z_t, t) + \frac{d\bar{v}}{dt}$$
$$= jvp(\bar{v}, (z, r, t), (v, 0, 1))$$
最后可得
$$\bar{v}(z_t, r, t) = v(z_t, t) - (t - r)jvp(\bar{v}, (z, r, t), (v, 0, 1)) \tag{2}$$
这个表达式是自举的，即右式中包含左式，左右相关。式中的瞬时速度$v$我们用条件速度$\epsilon-x_0$代替，结合如下的损失函数
$$L = \mathbb{E}\|\bar{v}_\theta(z_t, r, t) - sg(v(z_t, t) - (t - r)jvp(\bar{v}_\theta, (z, r, t), (v, 0, 1)))\|_2^2$$
我们可以发现，这个损失函数表示的是一种自我纠正，即我们已知起始和终止位置的情况下的条件瞬时速度始终是$v$，模型预测的平均速度是空间中的宏观流，表示在$z$位置从时间$r$到$t$，平均的流向和平均流向的变化率是什么样的，我们用这个特定样本的瞬时速度校正这个平均速度应该是什么样的，和模型预测的平均速度的差值，最终希望的是模型在这一点的平均速度能够逆推出该特定样本的特定瞬时速度。
我们此时会考虑这样一个问题，既然我们想要让模型预测平均速度，为何不直接让模型去拟合直线平均速度$\epsilon-x$，而是费劲进行表达式的推导呢。
原因在于上式中的瞬时速度，从模型最终推理的角度来说，应该是边缘速度场，我们现在考虑的是真实的生成过程中的瞬时速度和平均速度的关系，而边缘速度场不可能是平坦的，因此不能将平均速度场简单的认为和条件速度一致。训练算法如下
此处模型学会的平均速度不再是一个固定值，而是虽然变化但其变化能符合我们特定样本的条件瞬时速度的平均速度。
尽管如此，这部分理解起来仍然怪怪的，并不是十分有说服力。这大概也是后续improved meanflow中提出改进的原因。
可得到训练算法：
![](https://pic1.imgdb.cn/item/69ab8e9959f896a650d454ac.png)
其中和flow matching的主要区别在于加入了雅可比向量积的计算，引入了约16%的额外计算量，但得到的好处是生成时只需一步生成。
## iMF [@gengImprovedMeanFlows2025]
本篇文章是对MeanFlow的改进工作，MeanFlow的推导过程中对平均速度表达式中的瞬时速度都直接使用条件速度替代，实际上原表达式中的所有瞬时速度都应该是边际速度（边缘概率）而不应该是条件速度，在Flow Matching训练时使用条件速度的原因是条件速度的期望和边际速度二者的梯度是相同的，仅相差常数。
但在式2中，jvp内的瞬时速度如果使用条件速度，jvp表达式是一个复杂的非线性表达式，此处的瞬时速度本应该使用边际速度，但这里使用条件速度进行替代，而非线性表达式的特性导致该表达式的期望的梯度并不和使用边际速度时的梯度相同，这就导致训练不稳定。
因此可以对式1进行简单变形，得到瞬时速度和平均速度之间的关系，如下
$$\mathbf{v}(z_t) = \mathbf{u}(z_t) + (t - r)\frac{d}{dt}\mathbf{u}(z_t) $$
$$\mathbf{V}_\theta(z_t) \triangleq \mathbf{u}_\theta(z_t) + (t - r)JVP_{sg}(\mathbf{u}_\theta; \mathbf{v}_\theta) \quad (12)$$
这里$v$和$t$使用同一个网络预测，时刻记住meanflow中预测平均速度的网络$u$的输入是$t,r,z_t$三个参数。当$r=t$时，网络预测的就相当于瞬时速度，因此使用同一个网络预测两遍分别预测$v$和$t$，根据式12得到修正后的瞬时速度，让这个修正后的瞬时速度尽可能接近在直线路径假设下的条件速度$\epsilon-x$。
这里我们已经如果能让网络直接预测$v_\theta$为什么还要通过式12来计算这个瞬时速度呢，这里要结合我们的训练过程考虑，我们的最终目的仍然和MeanFlow一致，即让网络可以预测平均速度，这里我们看似是在计算瞬时速度的损失，实际上该瞬时速度是由式12的右式得来的，而且jvp不参与梯度计算即其中模型预测的瞬时速度$v_\theta$不会产生梯度更新，梯度更新仅发生在前面的$u_\theta(z_t)$中，因此我们实际上是在要求网络学会预测平均速度$u_\theta$。
通过这种改进，我们可以充分利用原本的Flow Matching的训练稳定性。
作者还发现，通过去掉adaLN-zero，但使用将同一个条件token重复多次并作为序列的一部分的形式来进行建模，可以达到和使用adaLN-zero一样的效果。
![](https://pic1.imgdb.cn/item/69c4de9d2fb41b9ea32f5f12.png)
但之前都是用于计算机视觉，我们将其用于机器人的模仿学习和动作生成，之前的类似研究是 Mean-Flow based One-Step Vision-Language-Action 为CVPR2026。但这篇文章只使用了meanflow，我们的improved meanflow改进了meanflow中存在的一些理论问题，同时获得更稳定的训练效果，除此以外我们增加了attnres用于解决模型训练过程中的梯度消失和深层网络特征非常相似的问题。attnres的理论来源于大模型训练过程Residual connections [12] with PreNorm [60] are standard in modern LLMs, yet they accumulate
all layer outputs with fixed unit weights. This uniform aggregation causes uncontrolled hidden-state
growth with depth, progressively diluting each layer’s contribution [27]. We propose Attention
Residuals (AttnRes), which replaces this fixed accumulation with softmax attention over preceding
layer outputs, allowing each layer to selectively aggregate earlier representations with learned, input-
dependent weights. To address the memory and communication overhead of attending over all
preceding layer outputs for large-scale model training, we introduce Block AttnRes, which partitions
layers into blocks and attends over block-level representations, reducing the memory footprint while
preserving most of the gains of full AttnRes. Combined with cache-based pipeline communication
and a two-phase computation strategy, Block AttnRes becomes a practical drop-in replacement for
standard residual connections with minimal overhead.
Scaling law experiments confirm that the improvement is consistent across model sizes, and ablations
validate the benefit of content-dependent depth-wise selection. We further integrate AttnRes into
the Kimi Linear architecture [69] (48B total / 3B activated parameters) and pre-train on 1.4T tokens,
where AttnRes mitigates PreNorm dilution, yielding more uniform output magnitudes and gradient
distribution across depth, and improves downstream performance across all evaluated tasks 。我的概述如下 这里我们换另外一种写法，它能让我们看出更深刻的东西。先记 $\boldsymbol{y}_t = \boldsymbol{f}_t(\boldsymbol{x}_{t-1})$，那么有 $\boldsymbol{x}_t = \boldsymbol{x}_{t-1} + \boldsymbol{y}_t$，约定 $\boldsymbol{y}_0 = \boldsymbol{x}_0$，那么易得 $\boldsymbol{x}_t = \boldsymbol{y}_0 + \boldsymbol{y}_1 + \cdots + \boldsymbol{y}_t$，于是它可以等价地写成
$$
\boldsymbol{y}_{t+1} = \boldsymbol{f}_{t+1}(\boldsymbol{y}_0 + \boldsymbol{y}_1 + \cdots + \boldsymbol{y}_t) \tag{2}
$$
即从 $\boldsymbol{y}$ 的视角看，Residuals是将 $\boldsymbol{y}_0, \boldsymbol{y}_1, \cdots, \boldsymbol{y}_t$ 等权求和作为 $\boldsymbol{f}_{t+1}$ 的输入来得到 $\boldsymbol{y}_{t+1}$，那么一个自然的推广就是换成加权求和：
$$
\boldsymbol{y}_{t+1} = \boldsymbol{f}_{t+1}\left(\sum_{s=0}^t a_{t+1,s} \boldsymbol{y}_s\right) \qquad \text{where} \quad a_{t,s} \ge 0, \quad \sum_{s=0}^t a_{t+1,s} = 1 \tag{3}
$$
此时我们可以发现Hyper-connection就是这种加权求和的一种表现形式。但Hyper-connection中的$H$都由$tanh$激活后连乘得来，这导致该值有爆炸或者坍缩的风险。
Deepseek的mHC通过引入对H的交替归一化使其满足双随机性，双随机性即矩阵的所有行列和都为1,而且双随机性具有乘法的封闭性，即双随机矩阵相乘还是双随机矩阵，这样使得H的连乘不会出现爆炸和坍缩的问题。
回过头去考虑式3,既然是加权求和，那么一个权重的设置方法就是使用注意力机制，让权重$a$等于注意力分数，于是就有了AttnRes（苏神的作品）。其形式数学上看起来并不困难（RoPE也是这样）。
$$
a_{t+1,s} \propto \exp(\boldsymbol{w}_{t+1} \cdot \boldsymbol{y}_s) \tag{6}
$$
其中 $\boldsymbol{w}_t$ 是一个可训练的向量参数，即直接以一个数据无关的静态向量为Q、而K、V都是 $\boldsymbol{y}_s$ 去做Softmax Attention，这便是第一版AttnRes。随后通过一些实验得到给K多加个RMSNorm的操作，能取得比较稳定的收益，这构成了终版的AttnRes形式
$$
a_{t+1,s} \propto \exp(\boldsymbol{w}_{t+1} \cdot \text{RMSNorm}(\boldsymbol{y}_s)) \tag{7}
$$
我们研究用于对比的模型是Diffusion Policy，和ACT，在我自己的roboimi仿真环境下 有sim_transfer和pocket_insert两个仿真环境，在这个两个仿真环境下我使用imf-attnres和diffusion policy进行了很多实验。之前的实验都是使用swanlab记录的，实验以socket-insert开头 在runs目录下 可以看到之前的实验记录
workspace/inputs/template.tex
+145
View File
@@ -0,0 +1,145 @@
\documentclass{article}
\usepackage{corl_2026} % Use this for the initial submission.
% \usepackage[final]{corl_2026} % Uncomment for the camera-ready ``final'' version.
% \usepackage[preprint]{corl_2026} % Uncomment for pre-prints (e.g., arxiv); This is like ``final'', but will remove the CORL footnote.
\title{Formatting Instructions for CoRL 2026 Camera-Ready}
% The \author macro works with any number of authors. There are two
% commands used to separate the names and addresses of multiple
% authors: \And and \AND.
%
% Using \And between authors leaves it to LaTeX to determine where to
% break the lines. Using \AND forces a line break at that point. So,
% if LaTeX puts 3 of 4 authors names on the first line, and the last
% on the second line, try using \AND instead of \And before the third
% author name.
% NOTE: authors will be visible only in the camera-ready and preprint versions (i.e., when using the option 'final' or 'preprint').
% For the initial submission the authors will be anonymized.
\author{
Jane E.~Doe\\
Department of Electrical Engineering and Computer Sciences\\
University of California Berkeley
United States\\
\texttt{janedoe@berkeley.edu} \\
%% examples of more authors
%% \And
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
%% \AND
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
%% \And
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
%% \And
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
}
\begin{document}
\maketitle
%===============================================================================
\begin{abstract}
The purpose of this document is to provide both the basic paper template and submission guidelines. Abstracts should be a single paragraph, between 4--6 sentences long, ideally. Gross violations will trigger corrections at the camera-ready phase.
\end{abstract}
% Two or three meaningful keywords should be added here
\keywords{CoRL, Robots, Learning}
%===============================================================================
\section{Introduction}
Submission to CoRL 2026 will be entirely electronic, via a web site (not email). Information about the submission process and \LaTeX{} templates are available on the conference web site at \url{https://corl.org/}. For camera ready submission, use the \texttt{final} option for the \texttt{\textbackslash usepackage} command.
%===============================================================================
\section{Citations}
\label{sec:citations}
Citations can be made using either \textbackslash citep\{\} or \textbackslash citet\{\}, depending from the appropriateness. To avoid the citation moving to the next line, it is often a good practice to replace the space before with a tilde (\~{}) character.
Example 1: ``CoRL is the best conference ever~\citep{Gauss1857}.''
Example 2: ``\citet{Lagrange1788} proved, both theoretically and numerically, that CoRL is the best conference ever.''
%===============================================================================
\section{Experimental Results}
\label{sec:result}
Nam dui ligula, fringilla a, euismod sodales, sollicitudin vel, wisi. Morbi auctor lorem non justo.
Nam lacus libero, pretium at, lobortis vitae, ultricies et, tellus. Donec aliquet, tortor sed accumsan
bibendum, erat ligula aliquet magna, vitae ornare odio metus a mi. Morbi ac orci et nisl hendrerit
mollis.
Suspendisse ut massa. Cras nec ante. Pellentesque a nulla. Cum sociis natoque penatibus
et magnis dis parturient montes, nascetur ridiculus mus. Aliquam tincidunt urna. Nulla ullamcorper
vestibulum turpis. Pellentesque cursus luctus mauris.
Nulla malesuada porttitor diam. Donec felis erat, congue non, volutpat at, tincidunt tristique, libero.
Vivamus viverra fermentum felis. Donec nonummy pellentesque ante. Phasellus adipiscing semper elit.
Proin fermentum massa ac quam. Sed diam turpis, molestie vitae, placerat a, molestie nec, leo.
Maecenas lacinia. Nam ipsum ligula, eleifend at, accumsan nec, suscipit a, ipsum. Morbi blandit
ligula feugiat magna. Nunc eleifend consequat lorem. Sed lacinia nulla vitae enim. Pellentesque tin-
cidunt purus vel magna. Integer non enim. Praesent euismod nunc eu purus. Donec bibendum quam
in tellus. Nullam cursus pulvinar lectus. Donec et mi. Nam vulputate metus eu enim. Vestibulum
pellentesque felis eu massa.
Quisque ullamcorper placerat ipsum. Cras nibh. Morbi vel justo vitae lacus tincidunt ultrices.
Lorem
ipsum dolor sit amet, consectetuer adipiscing elit. In hac habitasse platea dictumst. Integer tempus
convallis augue. Etiam facilisis. Nunc elementum fermentum wisi. Aenean placerat. Ut imperdiet,
enim sed gravida sollicitudin, felis odio placerat quam, ac pulvinar elit purus eget enim. Nunc vitae
tortor. Proin tempus nibh sit amet nisl. Vivamus quis tortor vitae risus porta vehicula.
%===============================================================================
\section{Conclusion}
\label{sec:conclusion}
Nam dui ligula, fringilla a, euismod sodales, sollicitudin vel, wisi. Morbi auctor lorem non justo.
Nam lacus libero, pretium at, lobortis vitae, ultricies et, tellus. Donec aliquet, tortor sed accumsan
bibendum, erat ligula aliquet magna, vitae ornare odio metus a mi. Morbi ac orci et nisl hendrerit
mollis. Suspendisse ut massa. Cras nec ante. Pellentesque a nulla. Cum sociis natoque penatibus
et magnis dis parturient montes, nascetur ridiculus mus. Aliquam tincidunt urna. Nulla ullamcorper
vestibulum turpis. Pellentesque cursus luctus mauris.
Nulla malesuada porttitor diam. Donec felis erat, congue non, volutpat at, tincidunt tristique, libero.
Vivamus viverra fermentum felis. Donec nonummy pellentesque ante. Phasellus adipiscing semper
elit. Proin fermentum massa ac quam. Sed diam turpis, molestie vitae, placerat a, molestie nec, leo.
Maecenas lacinia. Nam ipsum ligula, eleifend at, accumsan nec, suscipit a, ipsum. Morbi blandit
ligula feugiat magna. Nunc eleifend consequat lorem. Sed lacinia nulla vitae enim. Pellentesque tin-
cidunt purus vel magna. Integer non enim. Praesent euismod nunc eu purus. Donec bibendum quam
in tellus. Nullam cursus pulvinar lectus. Donec et mi. Nam vulputate metus eu enim. Vestibulum
pellentesque felis eu massa.
Quisque ullamcorper placerat ipsum. Cras nibh. Morbi vel justo vitae lacus tincidunt ultrices. Lorem
ipsum dolor sit amet, consectetuer adipiscing elit. In hac habitasse platea dictumst. Integer tempus
convallis augue. Etiam facilisis. Nunc elementum fermentum wisi. Aenean placerat. Ut imperdiet,
enim sed gravida sollicitudin, felis odio placerat quam, ac pulvinar elit purus eget enim. Nunc vitae
tortor. Proin tempus nibh sit amet nisl. Vivamus quis tortor vitae risus porta vehicula.
%===============================================================================
\clearpage
% The acknowledgments are automatically included only in the final and preprint versions of the paper.
\acknowledgments{If a paper is accepted, the final camera-ready version will (and probably should) include acknowledgments. All acknowledgments go at the end of the paper, including thanks to reviewers who gave useful comments, to colleagues who contributed to the ideas, and to funding agencies and corporate sponsors that provided financial support.}
%===============================================================================
% no \bibliographystyle is required, since the corl style is automatically used.
\bibliography{example} % .bib
\end{document}
workspace/inputs/template_unzipped/corl_2026_template_submission/corl_2026.sty
+472
View File
@@ -0,0 +1,472 @@
% File: corl_2026.sty
%
% Latex templates for the Conference on Robot Learning (CoRL)
%
% This template is heavily inspired by the NeurIPS, ICML, ICLR and IEEE Transactions latex templates.
% Hence we would like to thank: Roman Garnett and the previous mantainers of the NIPS style, Percy Liang and the previous mantainers of the ICML style, Hugo Larochelle for the ICLR style, and Michael Shell for the IEEE Transactions style.
%
% History:
% 2017/04/16 - First revision by Roberto Calandra (roberto.calandra@berkeley.edu).
% Main changes:
% - The abstract is more compact compared to NeurIPS/ICML
% - References are by default using natbib with squared numbers (e.g., [1])
% - DOI fields from the bibtex are automatically converted to hyperlinks
% to the corresponding page
% - acknowledgments are now a command, and the corresponding subsubsection is
% automatically included only in the final version
% 2017/06/12 - Modified to use corlabbrvnat.bst, which order the reference by order of appearance in the paper
% 2017/06/13 - fixed typo
% 2018/05/09 - Slightly modified for CoRL 2018 by Jun Morimoto (xmorimo@atr.jp)
% 2019/01/28 - Slightly modified for CoRL 2019 by Jun Nakanishi (jnakanis@meijo-u.ac.jp)
% 2020/02/02 - Slightly modified for CoRL 2020 by Cynthia Matuszek (cmat@umbc.edu)
% 2020/08/19 - Added preprint option by Roberto Calandra (rcalandra@fb.com)
% 2021/05/06 - Slightly modified for CoRL 2021 by Gerhard Neumann (gerhard.neumann@kit.edu)
% 2022/03/09 - Slightly modified for CoRL 2022 by Minas Liarokapis (minas.liarokapis@auckland.ac.nz)
% 2022/03/06 - Slightly modified for CoRL 2023 by Marc Toussaint (toussaint@tu-berlin.de)
% 2024/03/26 - Slightly modified for CoRL 2024 by David Held (dheld@andrew.cmu.edu)
% 2026/01/10 - Slightly modified for CoRL 2026 by Yoonchang Sung (yoonchang.sung@ntu.edu.sg)
%
% TODO: nohyperref is not working at the moment
%
\NeedsTeXFormat{LaTeX2e}
% Content to be changed from year to year
\ProvidesPackage{corl_2026}[2026/08/15 CORL2026 submission/preprint/camera-ready style file]
\newcommand{\@conferenceordinal}{10th}
\newcommand{\@conferenceyear}{2026}
\newcommand{\@conferencelocation}{Austin TX, USA}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Accepted options: [final,preprint,nonatbib,nohyperref]
% Declare the final option, which creates camera-ready copy
\newif\if@conferencefinal\@conferencefinalfalse
\DeclareOption{final}{
\@conferencefinaltrue
}
% Declare the preprint option, which creates a camera-ready copy without the corl footnote
\newif\if@preprinttype\@preprinttypefalse
\DeclareOption{preprint}{
\@preprinttypetrue
}
% The natbib package is loaded by default. Declaring the nonatbib option, does not load natbib in case of package clash (users can pass options to natbib via \PassOptionsToPackage)
\newif\if@natbib\@natbibtrue
\DeclareOption{nonatbib}{
\@natbibfalse
}
% The hyperref package is loaded by default. Declaring the nohyperref option, does not load the hyperref.
\DeclareOption{nohyperref}{%
\gdef\nohyperref{1}
}
% Activate the options
\ProcessOptions\relax
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Required packages:
\RequirePackage{lineno}
\RequirePackage{color}
% Load natbib unless told otherwise
\if@natbib
\RequirePackage[square,numbers]{natbib}
\bibliographystyle{corlabbrvnat}
\fi
% set page geometry
\RequirePackage{hyperref} % hyperlinks
\RequirePackage[verbose=true,letterpaper]{geometry}
\AtBeginDocument{
\newgeometry{
textheight=9in,
textwidth=5.5in,
top=1in,
headheight=12pt,
headsep=25pt,
footskip=30pt
}
\@ifpackageloaded{fullpage}
{\PackageWarning{corl_2026}{fullpage package not allowed! Overwriting formatting.}}
{}
}
\ifdefined\nohyperref\else\ifdefined\hypersetup
\definecolor{mydarkblue}{rgb}{0,0.08,0.45}
\hypersetup{ %
pdftitle={},
pdfauthor={},
pdfsubject={Proceedings of the \@conferenceordinal\/ Conference on Robot Learning (CoRL \@conferenceyear)},
pdfkeywords={},
pdfborder=0 0 0,
pdfpagemode=UseNone,
colorlinks=true,
linkcolor=mydarkblue,
citecolor=mydarkblue,
filecolor=mydarkblue,
urlcolor=mydarkblue,
pdfview=FitH}
\ifdefined\isaccepted \else
\hypersetup{pdfauthor={Anonymous Submission}}
\fi
\fi\fi
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% fonts
\renewcommand{\rmdefault}{ptm}
\renewcommand{\sfdefault}{phv}
% Create acknowledgments -- only if the option 'final' is activated
\providecommand{\acknowledgments}{}
\renewcommand{\acknowledgments}[1]{%
\if@conferencefinal%
\subsubsection*{Acknowledgments} #1
\fi
\if@preprinttype%
\subsubsection*{Acknowledgments} #1
\fi
}
% handle tweaks for camera-ready copy vs. submission copy
\if@conferencefinal
\newcommand{\@noticestring}{%
\@conferenceordinal\/ Conference on Robot Learning
(CoRL \@conferenceyear), \@conferencelocation.%
}
\else
\if@preprinttype
\newcommand{\@noticestring}{%
% Nothing here.
}
\else
\newcommand{\@noticestring}{%
Submitted to the \@conferenceordinal\/ Conference on Robot Learning (CoRL \@conferenceyear). Do not distribute.%
}
% line numbers for submission
\linenumbers
% fix incompatibilities between lineno and amsmath, if required, by
% transparently wrapping linenomath environments around amsmath
% environments
\AtBeginDocument{%
\@ifpackageloaded{amsmath}{%
\newcommand*\patchAmsMathEnvironmentForLineno[1]{%
\expandafter\let\csname old#1\expandafter\endcsname\csname #1\endcsname
\expandafter\let\csname oldend#1\expandafter\endcsname\csname end#1\endcsname
\renewenvironment{#1}%
{\linenomath\csname old#1\endcsname}%
{\csname oldend#1\endcsname\endlinenomath}%
}%
\newcommand*\patchBothAmsMathEnvironmentsForLineno[1]{%
\patchAmsMathEnvironmentForLineno{#1}%
\patchAmsMathEnvironmentForLineno{#1*}%
}%
\patchBothAmsMathEnvironmentsForLineno{equation}%
\patchBothAmsMathEnvironmentsForLineno{align}%
\patchBothAmsMathEnvironmentsForLineno{flalign}%
\patchBothAmsMathEnvironmentsForLineno{alignat}%
\patchBothAmsMathEnvironmentsForLineno{gather}%
\patchBothAmsMathEnvironmentsForLineno{multline}%
}{}
}
\fi
\fi
% The DOI will now automatically generate a URL link. Pretty cool!
% + Fix for hyperref and DOI (https://www.tug.org/pipermail/tex-live/2012-August/032161.html)
%-----------------------------
\makeatletter
\providecommand{\doi}[1]{%
\begingroup
\let\bibinfo\@secondoftwo
\urlstyle{rm}%
\href{http://dx.doi.org/#1}{%
doi:\discretionary{}{}{}%
\nolinkurl{#1}%
}%
\endgroup
}
% \makeatother
% %-----------------------------
\widowpenalty=10000
\clubpenalty=10000
\flushbottom
\sloppy
% font sizes with reduced leading
\renewcommand{\normalsize}{%
\@setfontsize\normalsize\@xpt\@xipt
\abovedisplayskip 7\p@ \@plus 2\p@ \@minus 5\p@
\abovedisplayshortskip \z@ \@plus 3\p@
\belowdisplayskip \abovedisplayskip
\belowdisplayshortskip 4\p@ \@plus 3\p@ \@minus 3\p@
}
\normalsize
\renewcommand{\small}{%
\@setfontsize\small\@ixpt\@xpt
\abovedisplayskip 6\p@ \@plus 1.5\p@ \@minus 4\p@
\abovedisplayshortskip \z@ \@plus 2\p@
\belowdisplayskip \abovedisplayskip
\belowdisplayshortskip 3\p@ \@plus 2\p@ \@minus 2\p@
}
\renewcommand{\footnotesize}{\@setfontsize\footnotesize\@ixpt\@xpt}
\renewcommand{\scriptsize}{\@setfontsize\scriptsize\@viipt\@viiipt}
\renewcommand{\tiny}{\@setfontsize\tiny\@vipt\@viipt}
\renewcommand{\large}{\@setfontsize\large\@xiipt{14}}
\renewcommand{\Large}{\@setfontsize\Large\@xivpt{16}}
\renewcommand{\LARGE}{\@setfontsize\LARGE\@xviipt{20}}
\renewcommand{\huge}{\@setfontsize\huge\@xxpt{23}}
\renewcommand{\Huge}{\@setfontsize\Huge\@xxvpt{28}}
% sections with less space
\providecommand{\section}{}
\renewcommand{\section}{%
\@startsection{section}{1}{\z@}%
{-2.0ex \@plus -0.5ex \@minus -0.2ex}%
{ 1.5ex \@plus 0.3ex \@minus 0.2ex}%
{\large\bf\raggedright}%
}
\providecommand{\subsection}{}
\renewcommand{\subsection}{%
\@startsection{subsection}{2}{\z@}%
{-1.8ex \@plus -0.5ex \@minus -0.2ex}%
{ 0.8ex \@plus 0.2ex}%
{\normalsize\bf\raggedright}%
}
\providecommand{\subsubsection}{}
\renewcommand{\subsubsection}{%
\@startsection{subsubsection}{3}{\z@}%
{-1.5ex \@plus -0.5ex \@minus -0.2ex}%
{ 0.5ex \@plus 0.2ex}%
{\normalsize\bf\raggedright}%
}
\providecommand{\paragraph}{}
\renewcommand{\paragraph}{%
\@startsection{paragraph}{4}{\z@}%
{1.5ex \@plus 0.5ex \@minus 0.2ex}%
{-1em}%
{\normalsize\bf}%
}
\providecommand{\subparagraph}{}
\renewcommand{\subparagraph}{%
\@startsection{subparagraph}{5}{\z@}%
{1.5ex \@plus 0.5ex \@minus 0.2ex}%
{-1em}%
{\normalsize\bf}%
}
\providecommand{\subsubsubsection}{}
\renewcommand{\subsubsubsection}{%
\vskip5pt{\noindent\normalsize\rm\raggedright}%
}
% float placement
\renewcommand{\topfraction }{0.85}
\renewcommand{\bottomfraction }{0.4}
\renewcommand{\textfraction }{0.1}
\renewcommand{\floatpagefraction}{0.7}
\newlength{\@nipsabovecaptionskip}\setlength{\@nipsabovecaptionskip}{7\p@}
\newlength{\@nipsbelowcaptionskip}\setlength{\@nipsbelowcaptionskip}{\z@}
\setlength{\abovecaptionskip}{\@nipsabovecaptionskip}
\setlength{\belowcaptionskip}{\@nipsbelowcaptionskip}
% swap above/belowcaptionskip lengths for tables
\renewenvironment{table}
{\setlength{\abovecaptionskip}{\@nipsbelowcaptionskip}%
\setlength{\belowcaptionskip}{\@nipsabovecaptionskip}%
\@float{table}}
{\end@float}
% footnote formatting
\setlength{\footnotesep }{6.65\p@}
\setlength{\skip\footins}{9\p@ \@plus 4\p@ \@minus 2\p@}
\renewcommand{\footnoterule}{\kern-3\p@ \hrule width 12pc \kern 2.6\p@}
\setcounter{footnote}{0}
% paragraph formatting
\setlength{\parindent}{\z@}
\setlength{\parskip }{5.5\p@}
% list formatting
\setlength{\topsep }{4\p@ \@plus 1\p@ \@minus 2\p@}
\setlength{\partopsep }{1\p@ \@plus 0.5\p@ \@minus 0.5\p@}
\setlength{\itemsep }{2\p@ \@plus 1\p@ \@minus 0.5\p@}
\setlength{\parsep }{2\p@ \@plus 1\p@ \@minus 0.5\p@}
\setlength{\leftmargin }{3pc}
\setlength{\leftmargini }{\leftmargin}
\setlength{\leftmarginii }{2em}
\setlength{\leftmarginiii}{1.5em}
\setlength{\leftmarginiv }{1.0em}
\setlength{\leftmarginv }{0.5em}
\def\@listi {\leftmargin\leftmargini}
\def\@listii {\leftmargin\leftmarginii
\labelwidth\leftmarginii
\advance\labelwidth-\labelsep
\topsep 2\p@ \@plus 1\p@ \@minus 0.5\p@
\parsep 1\p@ \@plus 0.5\p@ \@minus 0.5\p@
\itemsep \parsep}
\def\@listiii{\leftmargin\leftmarginiii
\labelwidth\leftmarginiii
\advance\labelwidth-\labelsep
\topsep 1\p@ \@plus 0.5\p@ \@minus 0.5\p@
\parsep \z@
\partopsep 0.5\p@ \@plus 0\p@ \@minus 0.5\p@
\itemsep \topsep}
\def\@listiv {\leftmargin\leftmarginiv
\labelwidth\leftmarginiv
\advance\labelwidth-\labelsep}
\def\@listv {\leftmargin\leftmarginv
\labelwidth\leftmarginv
\advance\labelwidth-\labelsep}
\def\@listvi {\leftmargin\leftmarginvi
\labelwidth\leftmarginvi
\advance\labelwidth-\labelsep}
% create title
\providecommand{\maketitle}{}
\renewcommand{\maketitle}{%
\par
\begingroup
\renewcommand{\thefootnote}{\fnsymbol{footnote}}
% for perfect author name centering
\renewcommand{\@makefnmark}{\hbox to \z@{$^{\@thefnmark}$\hss}}
% The footnote-mark was overlapping the footnote-text,
% added the following to fix this problem (MK)
\long\def\@makefntext##1{%
\parindent 1em\noindent
\hbox to 1.8em{\hss $\m@th ^{\@thefnmark}$}##1
}
\thispagestyle{empty}
\@maketitle
\@thanks
\@notice
\endgroup
\let\maketitle\relax
\let\thanks\relax
}
% rules for title box at top of first page
\newcommand{\@toptitlebar}{
\hrule height 4\p@
\vskip 0.25in
\vskip -\parskip%
}
\newcommand{\@bottomtitlebar}{
\vskip 0.29in
\vskip -\parskip
\hrule height 1\p@
\vskip 0.09in%
}
%% keywords as first class citizens
\def\keywords#1{%
% \ifdefined\isaccepted \else
% \par {\bf Keywords:} #1%
% \fi
% \ifdefined\nohyperref\else\ifdefined\hypersetup
% \hypersetup{pdfkeywords={#1}}
% \fi\fi
\ifdefined\isaccepted \else
\begin{quote}
\textbf{Keywords:} #1%
\end{quote}
\fi
\ifdefined\nohyperref\else\ifdefined\hypersetup
\hypersetup{pdfkeywords={#1}}
\fi\fi
}
% create title (includes both anonymized and non-anonymized versions)
\providecommand{\@maketitle}{}
\renewcommand{\@maketitle}{%
\vbox{%
\hsize\textwidth
\linewidth\hsize
\vskip 0.1in
% \@toptitlebar
\centering
{\LARGE\bf \@title\par}
% \@bottomtitlebar
\if@conferencefinal
\def\And{%
\end{tabular}\hfil\linebreak[0]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\def\AND{%
\end{tabular}\hfil\linebreak[4]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\@author\end{tabular}%
\else
\if@preprinttype
\def\And{%
\end{tabular}\hfil\linebreak[0]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\def\AND{%
\end{tabular}\hfil\linebreak[4]\hfil%
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\ignorespaces%
}
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}\@author\end{tabular}%
\else
\begin{tabular}[t]{c}\bf\rule{\z@}{24\p@}
Anonymous Author(s) \\
Affiliation \\
Address \\
\texttt{email} \\
\end{tabular}%
\fi
\fi
\vskip 0.3in \@minus 0.1in
}
}
% add conference notice to bottom of first page
\newcommand{\ftype@noticebox}{8}
\newcommand{\@notice}{%
% give a bit of extra room back to authors on first page
\enlargethispage{2\baselineskip}%
\@float{noticebox}[b]%
\footnotesize\@noticestring%
\end@float%
}
% abstract styling
\renewenvironment{abstract}%
{%
% \vskip 0.075in%
% \centerline%
% {\large\bf Abstract}%
% \vspace{0.5ex}%
\begin{quote}%
\textbf{Abstract:}%
}
{
\par%
\vskip 1ex%
% \ifdefined\keywords
% \textbf{Keywords:}%
% \@keywords%
% \else
% \fi
\end{quote}%
}
\endinput
workspace/inputs/template_unzipped/corl_2026_template_submission/corlabbrvnat.bst
+1442
View File
File diff suppressed because it is too large Load Diff
workspace/inputs/template_unzipped/corl_2026_template_submission/example.bib
+20
View File
@@ -0,0 +1,20 @@
% This file was created with JabRef 2.10.
% Encoding: UTF-8
@Article{Gauss1857,
Title = {Theory of the motion of the heavenly bodies moving about the sun in conic sections},
Author = {Carl Friedrich Gauss and Charles Henry Davis},
Journal = {Gauss's Theoria Motus},
Year = {1857},
Number = {1},
Pages = {5--23},
Volume = {76}
}
@Book{Lagrange1788,
title = {M{\'e}canique Analytique},
author = {Joseph-Louis Lagrange},
publisher = {Desaint, Paris},
year = {1788}
}
workspace/inputs/template_unzipped/corl_2026_template_submission/example.tex
+145
View File
@@ -0,0 +1,145 @@
\documentclass{article}
\usepackage{corl_2026} % Use this for the initial submission.
% \usepackage[final]{corl_2026} % Uncomment for the camera-ready ``final'' version.
% \usepackage[preprint]{corl_2026} % Uncomment for pre-prints (e.g., arxiv); This is like ``final'', but will remove the CORL footnote.
\title{Formatting Instructions for CoRL 2026 Camera-Ready}
% The \author macro works with any number of authors. There are two
% commands used to separate the names and addresses of multiple
% authors: \And and \AND.
%
% Using \And between authors leaves it to LaTeX to determine where to
% break the lines. Using \AND forces a line break at that point. So,
% if LaTeX puts 3 of 4 authors names on the first line, and the last
% on the second line, try using \AND instead of \And before the third
% author name.
% NOTE: authors will be visible only in the camera-ready and preprint versions (i.e., when using the option 'final' or 'preprint').
% For the initial submission the authors will be anonymized.
\author{
Jane E.~Doe\\
Department of Electrical Engineering and Computer Sciences\\
University of California Berkeley
United States\\
\texttt{janedoe@berkeley.edu} \\
%% examples of more authors
%% \And
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
%% \AND
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
%% \And
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
%% \And
%% Coauthor \\
%% Affiliation \\
%% Address \\
%% \texttt{email} \\
}
\begin{document}
\maketitle
%===============================================================================
\begin{abstract}
The purpose of this document is to provide both the basic paper template and submission guidelines. Abstracts should be a single paragraph, between 4--6 sentences long, ideally. Gross violations will trigger corrections at the camera-ready phase.
\end{abstract}
% Two or three meaningful keywords should be added here
\keywords{CoRL, Robots, Learning}
%===============================================================================
\section{Introduction}
Submission to CoRL 2026 will be entirely electronic, via a web site (not email). Information about the submission process and \LaTeX{} templates are available on the conference web site at \url{https://corl.org/}. For camera ready submission, use the \texttt{final} option for the \texttt{\textbackslash usepackage} command.
%===============================================================================
\section{Citations}
\label{sec:citations}
Citations can be made using either \textbackslash citep\{\} or \textbackslash citet\{\}, depending from the appropriateness. To avoid the citation moving to the next line, it is often a good practice to replace the space before with a tilde (\~{}) character.
Example 1: ``CoRL is the best conference ever~\citep{Gauss1857}.''
Example 2: ``\citet{Lagrange1788} proved, both theoretically and numerically, that CoRL is the best conference ever.''
%===============================================================================
\section{Experimental Results}
\label{sec:result}
Nam dui ligula, fringilla a, euismod sodales, sollicitudin vel, wisi. Morbi auctor lorem non justo.
Nam lacus libero, pretium at, lobortis vitae, ultricies et, tellus. Donec aliquet, tortor sed accumsan
bibendum, erat ligula aliquet magna, vitae ornare odio metus a mi. Morbi ac orci et nisl hendrerit
mollis.
Suspendisse ut massa. Cras nec ante. Pellentesque a nulla. Cum sociis natoque penatibus
et magnis dis parturient montes, nascetur ridiculus mus. Aliquam tincidunt urna. Nulla ullamcorper
vestibulum turpis. Pellentesque cursus luctus mauris.
Nulla malesuada porttitor diam. Donec felis erat, congue non, volutpat at, tincidunt tristique, libero.
Vivamus viverra fermentum felis. Donec nonummy pellentesque ante. Phasellus adipiscing semper elit.
Proin fermentum massa ac quam. Sed diam turpis, molestie vitae, placerat a, molestie nec, leo.
Maecenas lacinia. Nam ipsum ligula, eleifend at, accumsan nec, suscipit a, ipsum. Morbi blandit
ligula feugiat magna. Nunc eleifend consequat lorem. Sed lacinia nulla vitae enim. Pellentesque tin-
cidunt purus vel magna. Integer non enim. Praesent euismod nunc eu purus. Donec bibendum quam
in tellus. Nullam cursus pulvinar lectus. Donec et mi. Nam vulputate metus eu enim. Vestibulum
pellentesque felis eu massa.
Quisque ullamcorper placerat ipsum. Cras nibh. Morbi vel justo vitae lacus tincidunt ultrices.
Lorem
ipsum dolor sit amet, consectetuer adipiscing elit. In hac habitasse platea dictumst. Integer tempus
convallis augue. Etiam facilisis. Nunc elementum fermentum wisi. Aenean placerat. Ut imperdiet,
enim sed gravida sollicitudin, felis odio placerat quam, ac pulvinar elit purus eget enim. Nunc vitae
tortor. Proin tempus nibh sit amet nisl. Vivamus quis tortor vitae risus porta vehicula.
%===============================================================================
\section{Conclusion}
\label{sec:conclusion}
Nam dui ligula, fringilla a, euismod sodales, sollicitudin vel, wisi. Morbi auctor lorem non justo.
Nam lacus libero, pretium at, lobortis vitae, ultricies et, tellus. Donec aliquet, tortor sed accumsan
bibendum, erat ligula aliquet magna, vitae ornare odio metus a mi. Morbi ac orci et nisl hendrerit
mollis. Suspendisse ut massa. Cras nec ante. Pellentesque a nulla. Cum sociis natoque penatibus
et magnis dis parturient montes, nascetur ridiculus mus. Aliquam tincidunt urna. Nulla ullamcorper
vestibulum turpis. Pellentesque cursus luctus mauris.
Nulla malesuada porttitor diam. Donec felis erat, congue non, volutpat at, tincidunt tristique, libero.
Vivamus viverra fermentum felis. Donec nonummy pellentesque ante. Phasellus adipiscing semper
elit. Proin fermentum massa ac quam. Sed diam turpis, molestie vitae, placerat a, molestie nec, leo.
Maecenas lacinia. Nam ipsum ligula, eleifend at, accumsan nec, suscipit a, ipsum. Morbi blandit
ligula feugiat magna. Nunc eleifend consequat lorem. Sed lacinia nulla vitae enim. Pellentesque tin-
cidunt purus vel magna. Integer non enim. Praesent euismod nunc eu purus. Donec bibendum quam
in tellus. Nullam cursus pulvinar lectus. Donec et mi. Nam vulputate metus eu enim. Vestibulum
pellentesque felis eu massa.
Quisque ullamcorper placerat ipsum. Cras nibh. Morbi vel justo vitae lacus tincidunt ultrices. Lorem
ipsum dolor sit amet, consectetuer adipiscing elit. In hac habitasse platea dictumst. Integer tempus
convallis augue. Etiam facilisis. Nunc elementum fermentum wisi. Aenean placerat. Ut imperdiet,
enim sed gravida sollicitudin, felis odio placerat quam, ac pulvinar elit purus eget enim. Nunc vitae
tortor. Proin tempus nibh sit amet nisl. Vivamus quis tortor vitae risus porta vehicula.
%===============================================================================
\clearpage
% The acknowledgments are automatically included only in the final and preprint versions of the paper.
\acknowledgments{If a paper is accepted, the final camera-ready version will (and probably should) include acknowledgments. All acknowledgments go at the end of the paper, including thanks to reviewers who gave useful comments, to colleagues who contributed to the ideas, and to funding agencies and corporate sponsors that provided financial support.}
%===============================================================================
% no \bibliographystyle is required, since the corl style is automatically used.
\bibliography{example} % .bib
\end{document}