feat(vla): add SmolVLA conditioning and experiment artifacts
This commit is contained in:
Logic
@@ -0,0 +1,118 @@
|
||||
\relax
|
||||
\bibstyle{corlabbrvnat}
|
||||
\providecommand\hyper@newdestlabel[2]{}
|
||||
\providecommand\HyField@AuxAddToFields[1]{}
|
||||
\providecommand\HyField@AuxAddToCoFields[2]{}
|
||||
\citation{chi2023diffusion}
|
||||
\citation{ho2020denoising}
|
||||
\citation{peebles2023scalable}
|
||||
\citation{lipman2023flow,liu2022flow}
|
||||
\citation{geng2025mean}
|
||||
\citation{geng2025improved}
|
||||
\citation{he2016deep,vaswani2017attention}
|
||||
\citation{kimi2026attention}
|
||||
\@writefile{toc}{\contentsline {section}{\numberline {1}Introduction}{1}{section.1}\protected@file@percent }
|
||||
\newlabel{sec:introduction}{{1}{1}{Introduction}{section.1}{}}
|
||||
\newlabel{sec:introduction@cref}{{[section][1][]1}{[1][1][]1}{}{}{}}
|
||||
\citation{zhao2023learning,shukor2025smolvla}
|
||||
\citation{chi2023diffusion}
|
||||
\citation{ho2020denoising}
|
||||
\citation{peebles2023scalable}
|
||||
\citation{lipman2023flow}
|
||||
\citation{liu2022flow}
|
||||
\citation{brohan2022rt,brohan2023rt}
|
||||
\citation{kim2024openvla}
|
||||
\citation{shukor2025smolvla,pertsch2025fast}
|
||||
\citation{zhao2023learning}
|
||||
\citation{geng2025mean}
|
||||
\def\@LN@column{1}
|
||||
\@writefile{toc}{\contentsline {section}{\numberline {2}Related Work}{2}{section.2}\protected@file@percent }
|
||||
\newlabel{sec:related_work}{{2}{2}{Related Work}{section.2}{}}
|
||||
\newlabel{sec:related_work@cref}{{[section][2][]2}{[1][2][]2}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {paragraph}{Diffusion and flow policies for robot action generation.}{2}{section*.1}\protected@file@percent }
|
||||
\@writefile{toc}{\contentsline {paragraph}{Vision-language-action models and action chunking.}{2}{section*.2}\protected@file@percent }
|
||||
\@writefile{toc}{\contentsline {paragraph}{Mean-flow training for one-step generation.}{2}{section*.3}\protected@file@percent }
|
||||
\citation{geng2025improved}
|
||||
\citation{kimi2026attention}
|
||||
\def\@LN@column{1}
|
||||
\@writefile{toc}{\contentsline {paragraph}{Residual aggregation and attention residuals.}{3}{section*.4}\protected@file@percent }
|
||||
\@writefile{toc}{\contentsline {section}{\numberline {3}Method}{3}{section.3}\protected@file@percent }
|
||||
\newlabel{sec:method}{{3}{3}{Method}{section.3}{}}
|
||||
\newlabel{sec:method@cref}{{[section][3][]3}{[1][3][]3}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {3.1}Policy formulation}{3}{subsection.3.1}\protected@file@percent }
|
||||
\newlabel{sec:policy_formulation}{{3.1}{3}{Policy formulation}{subsection.3.1}{}}
|
||||
\newlabel{sec:policy_formulation@cref}{{[subsection][1][3]3.1}{[1][3][]3}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {3.2}Improved Mean Flow action generation}{3}{subsection.3.2}\protected@file@percent }
|
||||
\newlabel{sec:imf_action_generation}{{3.2}{3}{Improved Mean Flow action generation}{subsection.3.2}{}}
|
||||
\newlabel{sec:imf_action_generation@cref}{{[subsection][2][3]3.2}{[1][3][]3}{}{}{}}
|
||||
\@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces Planned method overview figure. The current draft intentionally stores the Paper Banana generation prompt as a placeholder instead of generating an image.}}{4}{figure.1}\protected@file@percent }
|
||||
\newlabel{fig:method_overview}{{1}{4}{Planned method overview figure. The current draft intentionally stores the Paper Banana generation prompt as a placeholder instead of generating an image}{figure.1}{}}
|
||||
\newlabel{fig:method_overview@cref}{{[figure][1][]1}{[1][4][]4}{}{}{}}
|
||||
\@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces Planned technical component figure. The prompt is retained as a placeholder for later Paper Banana rendering.}}{4}{figure.2}\protected@file@percent }
|
||||
\newlabel{fig:imf_attnres_components}{{2}{4}{Planned technical component figure. The prompt is retained as a placeholder for later Paper Banana rendering}{figure.2}{}}
|
||||
\newlabel{fig:imf_attnres_components@cref}{{[figure][2][]2}{[1][4][]4}{}{}{}}
|
||||
\def\@LN@column{1}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {3.3}Attention Residual policy transformer}{4}{subsection.3.3}\protected@file@percent }
|
||||
\newlabel{sec:attnres_policy}{{3.3}{4}{Attention Residual policy transformer}{subsection.3.3}{}}
|
||||
\newlabel{sec:attnres_policy@cref}{{[subsection][3][3]3.3}{[1][4][]4}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {3.4}Inference and control loop}{4}{subsection.3.4}\protected@file@percent }
|
||||
\newlabel{sec:inference_loop}{{3.4}{4}{Inference and control loop}{subsection.3.4}{}}
|
||||
\newlabel{sec:inference_loop@cref}{{[subsection][4][3]3.4}{[1][4][]4}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {section}{\numberline {4}Experiments}{4}{section.4}\protected@file@percent }
|
||||
\newlabel{sec:experiments}{{4}{4}{Experiments}{section.4}{}}
|
||||
\newlabel{sec:experiments@cref}{{[section][4][]4}{[1][4][]4}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {4.1}Setup and metrics}{4}{subsection.4.1}\protected@file@percent }
|
||||
\newlabel{sec:setup_metrics}{{4.1}{4}{Setup and metrics}{subsection.4.1}{}}
|
||||
\newlabel{sec:setup_metrics@cref}{{[subsection][1][4]4.1}{[1][4][]4}{}{}{}}
|
||||
\@writefile{lot}{\contentsline {table}{\numberline {1}{\ignorespaces Socket peg insertion over 100 rollouts. Success-like uses the recorded \texttt {max\_reward > 4} count.}}{5}{table.1}\protected@file@percent }
|
||||
\newlabel{tab:socket_results}{{1}{5}{Socket peg insertion over 100 rollouts. Success-like uses the recorded \texttt {max\_reward > 4} count}{table.1}{}}
|
||||
\newlabel{tab:socket_results@cref}{{[table][1][]1}{[1][5][]5}{}{}{}}
|
||||
\@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces Planned socket reward-latency figure. The current draft contains the Paper Banana prompt placeholder instead of a rendered image.}}{5}{figure.3}\protected@file@percent }
|
||||
\newlabel{fig:socket_reward_latency}{{3}{5}{Planned socket reward-latency figure. The current draft contains the Paper Banana prompt placeholder instead of a rendered image}{figure.3}{}}
|
||||
\newlabel{fig:socket_reward_latency@cref}{{[figure][3][]3}{[1][5][]5}{}{}{}}
|
||||
\def\@LN@column{1}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {4.2}Socket peg insertion}{5}{subsection.4.2}\protected@file@percent }
|
||||
\newlabel{sec:socket_results}{{4.2}{5}{Socket peg insertion}{subsection.4.2}{}}
|
||||
\newlabel{sec:socket_results@cref}{{[subsection][2][4]4.2}{[1][5][]5}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {4.3}Object transfer and ablations}{5}{subsection.4.3}\protected@file@percent }
|
||||
\newlabel{sec:sim_transfer_results}{{4.3}{5}{Object transfer and ablations}{subsection.4.3}{}}
|
||||
\newlabel{sec:sim_transfer_results@cref}{{[subsection][3][4]4.3}{[1][5][]5}{}{}{}}
|
||||
\@writefile{lot}{\contentsline {table}{\numberline {2}{\ignorespaces Object transfer / sim\_transfer over 100 rollouts. Success-like uses the recorded \texttt {max\_reward >= 4} count.}}{6}{table.2}\protected@file@percent }
|
||||
\newlabel{tab:sim_transfer_results}{{2}{6}{Object transfer / sim\_transfer over 100 rollouts. Success-like uses the recorded \texttt {max\_reward >= 4} count}{table.2}{}}
|
||||
\newlabel{tab:sim_transfer_results@cref}{{[table][2][]2}{[1][5][]6}{}{}{}}
|
||||
\@writefile{lof}{\contentsline {figure}{\numberline {4}{\ignorespaces Planned sim\_transfer ablation figure. The prompt is retained for later Paper Banana rendering.}}{6}{figure.4}\protected@file@percent }
|
||||
\newlabel{fig:sim_transfer_ablation}{{4}{6}{Planned sim\_transfer ablation figure. The prompt is retained for later Paper Banana rendering}{figure.4}{}}
|
||||
\newlabel{fig:sim_transfer_ablation@cref}{{[figure][4][]4}{[1][6][]6}{}{}{}}
|
||||
\def\@LN@column{1}
|
||||
\@writefile{toc}{\contentsline {subsection}{\numberline {4.4}Derived speed-quality comparisons}{6}{subsection.4.4}\protected@file@percent }
|
||||
\newlabel{sec:derived_comparisons}{{4.4}{6}{Derived speed-quality comparisons}{subsection.4.4}{}}
|
||||
\newlabel{sec:derived_comparisons@cref}{{[subsection][4][4]4.4}{[1][6][]6}{}{}{}}
|
||||
\@writefile{toc}{\contentsline {section}{\numberline {5}Limitations}{6}{section.5}\protected@file@percent }
|
||||
\newlabel{sec:limitations}{{5}{6}{Limitations}{section.5}{}}
|
||||
\newlabel{sec:limitations@cref}{{[section][5][]5}{[1][6][]6}{}{}{}}
|
||||
\@writefile{lot}{\contentsline {table}{\numberline {3}{\ignorespaces Derived comparisons from the 100-rollout logs.}}{7}{table.3}\protected@file@percent }
|
||||
\newlabel{tab:derived_comparisons}{{3}{7}{Derived comparisons from the 100-rollout logs}{table.3}{}}
|
||||
\newlabel{tab:derived_comparisons@cref}{{[table][3][]3}{[1][6][]7}{}{}{}}
|
||||
\def\@LN@column{1}
|
||||
\@writefile{toc}{\contentsline {section}{\numberline {6}Conclusion}{7}{section.6}\protected@file@percent }
|
||||
\newlabel{sec:conclusion}{{6}{7}{Conclusion}{section.6}{}}
|
||||
\newlabel{sec:conclusion@cref}{{[section][6][]6}{[1][7][]7}{}{}{}}
|
||||
\bibdata{refs}
|
||||
\bibcite{chi2023diffusion}{{1}{2023}{{Chi et~al.}}{{Chi, Xu, Feng, Cousineau, Du, Burchfiel, Tedrake, and Song}}}
|
||||
\bibcite{ho2020denoising}{{2}{2020}{{Ho et~al.}}{{Ho, Jain, and Abbeel}}}
|
||||
\bibcite{peebles2023scalable}{{3}{2023}{{Peebles and Xie}}{{}}}
|
||||
\bibcite{lipman2023flow}{{4}{2023}{{Lipman et~al.}}{{Lipman, Chen, Ben-Hamu, Nickel, and Le}}}
|
||||
\bibcite{liu2022flow}{{5}{2022}{{Liu et~al.}}{{Liu, Gong, and Liu}}}
|
||||
\bibcite{geng2025mean}{{6}{2025{}}{{Geng et~al.}}{{Geng, Deng, Bai, Kolter, and He}}}
|
||||
\bibcite{geng2025improved}{{7}{2025{}}{{Geng et~al.}}{{Geng, Lu, Wu, Shechtman, Kolter, and He}}}
|
||||
\bibcite{he2016deep}{{8}{2016}{{He et~al.}}{{He, Zhang, Ren, and Sun}}}
|
||||
\bibcite{vaswani2017attention}{{9}{2017}{{Vaswani et~al.}}{{Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, and Polosukhin}}}
|
||||
\bibcite{kimi2026attention}{{10}{2026}{{Kimi Team}}{{}}}
|
||||
\bibcite{zhao2023learning}{{11}{2023}{{Zhao et~al.}}{{Zhao, Kumar, Levine, and Finn}}}
|
||||
\bibcite{shukor2025smolvla}{{12}{2025}{{Shukor et~al.}}{{Shukor, Aubakirova, Capuano, Kooijmans, Palma, Zouitine, Aractingi, Pascal, Russi, Marafioti, et~al.}}}
|
||||
\bibcite{brohan2022rt}{{13}{2022}{{Brohan et~al.}}{{Brohan, Brown, Carbajal, Chebotar, Dabis, Finn, Gopalakrishnan, Hausman, Herzog, Hsu, et~al.}}}
|
||||
\bibcite{brohan2023rt}{{14}{2023}{{Brohan et~al.}}{{Brohan, Brown, Carbajal, Chebotar, Chen, Choromanski, Ding, Driess, Dubey, Finn, et~al.}}}
|
||||
\bibcite{kim2024openvla}{{15}{2024}{{Kim et~al.}}{{Kim, Pertsch, Karamcheti, Xiao, Balakrishna, Nair, Rafailov, Foster, Lam, Sanketi, et~al.}}}
|
||||
\bibcite{pertsch2025fast}{{16}{2025}{{Pertsch et~al.}}{{Pertsch, Stachowicz, Ichter, Driess, Nair, Vuong, Mees, Finn, and Levine}}}
|
||||
\def\@LN@column{1}
|
||||
\gdef \@abspage@last{8}
|
||||
Reference in New Issue
Block a user