AI Paper Atlas

Research map / 3D vision / 3D generation

Bootstrap Dynamic-Aware 3D Visual Representation for Scalable Robot Learning

Qiwei Liang, Boyang Cai, Minghao Lai, Sitong Zhuang, Tao Lin, Yan Qin, Yixuan Ye, Jiaming Liang, Renjing Xu

CVPR 2026 · 3D vision › 3D generation · approach: Learning 3D reconstruction from images · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use a diffusion transformer forward dynamics model instead of a deterministic Transformer (ablation)AFRO · Real-world tasksmean success rate (%): 76.25 -> 84.00Improved
add feature differencing input to the inverse dynamics model (ablation)AFRO · Real-world tasksmean success rate (%): 75.00 -> 84.00Improved
add inverse-consistency supervision (ablation)AFRO · Real-world tasksmean success rate (%): 73.75 -> 84.00Improved
use VICReg loss instead of plain MSE on features (ablation)AFRO · Real-world tasksmean success rate (%): 57.50 -> 84.00Improved
use latent action dimension 16 rather than 64 or 128 (ablation)AFRO · Real-world tasksmean success rate (%): 80.00 / 76.25 -> 84.00Improved
use frame interval k=4 rather than k=1 or k=16 (ablation)AFRO · Real-world tasksmean success rate (%): 81.25 / 77.50 -> 84.00Improved
pretrain on multiple task domains instead of in-domain only (ablation)AFRO · MetaWorld and Adroitmean success rate (%): 68.0% -> 82.8%Improved
pretrain on large out-of-domain RH20T data before fine-tuningAFRO · Real-world tasksmean success rate (%): 70.0% -> 84%Improved
pretrain on out-of-domain data for dynamics baselines such as DynaMo-3D and FVPDynaMo-3D, FVP · Real-world tasksmean success rate (%): DynaMo-3D 48 -> 58; FVP 34 -> 46Improved
add dynamics modeling to 3D pretraining via DynaMo-3Dstatic 3D pretraining · MetaWorldmean success rate (%): PointMAE 63.9 vs DynaMo-3D 64.9; FVP 54.3Mixed
use AFRO instead of baselines for unseen object transferBell Pressing · unseen objectssuccess rate (%): AFRO 75 -> 60 vs FVP 40 -> 5Improved
use AFRO under cluttered scenesBell Pressing, B2B Alignment · cluttered scenessuccess rate (%): 75 -> 70; 70 -> 65Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy