AI Paper Atlas

Research map / 3D vision / 3D generation

Learning to Control Physically-simulated 3D Characters via Generating and Mimicking 2D Motions

Jianan Li, Xiao Chen, Tao Huang, Tien-Tsin Wong

CVPR 2026 · 3D vision › 3D generation · approach: Learning physics from video · 3 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
learn directly from 2D motion instead of reconstructed 3D (two-stage Sfv*)Mimic2DM · Soccer Dribblesuccess rate (%): 47.8 -> 91.3Improved
learn directly from 2D motion instead of reconstructed 3D (two-stage Sfv*)Mimic2DM · Animalsuccess rate (%): 50.0 -> 83.3Improved
randomize ball position at inference (ablation)Mimic2DM · Soccer Dribblesuccess rate (%): 91.3 -> 83.2Hurt
add random 300 N force perturbations (ablation)Mimic2DM · Soccer Dribblesuccess rate (%): 91.3 -> 72.6Hurt
corrupt initial states with Gaussian noise during training (ablation)Mimic2DM · Soccer Dribblesuccess rate (%): 91.3 -> 93.4No clear effect
aggregate 2 views instead of 1 view at inference (ablation)Mimic2DM · AIST++ test3D tracking error: 254.6 -> 164.9Improved
aggregate 3 views instead of 2 views at inference (ablation)Mimic2DM · AIST++ test3D tracking error: 164.9 -> 161.5Improved
train with 2D data from 2 views vs 3D-supervised baselineMimic2DM · AIST++ test3D tracking error: 141 -> 164.9Hurt
use autoregressive 2D generator instead of diffusion generatorhierarchical controller · AIST++FID: 5.92 -> 2.44Improved
use autoregressive 2D generator instead of diffusion generatorhierarchical controller · Soccer Dribblesuccess rate: 0.13 -> 0.71Improved
train on diverse viewpoints instead of homogeneous viewpoints (ablation)view-agnostic policy · lift-the-box imitationqualitative naturalnessImproved
add adaptive state initialization (ablation)single-view tracking training · initial state refinementqualitative feasibilityImproved

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