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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| learn directly from 2D motion instead of reconstructed 3D (two-stage Sfv*) | Mimic2DM · Soccer Dribble | success rate (%): 47.8 -> 91.3 | Improved |
| learn directly from 2D motion instead of reconstructed 3D (two-stage Sfv*) | Mimic2DM · Animal | success rate (%): 50.0 -> 83.3 | Improved |
| randomize ball position at inference (ablation) | Mimic2DM · Soccer Dribble | success rate (%): 91.3 -> 83.2 | Hurt |
| add random 300 N force perturbations (ablation) | Mimic2DM · Soccer Dribble | success rate (%): 91.3 -> 72.6 | Hurt |
| corrupt initial states with Gaussian noise during training (ablation) | Mimic2DM · Soccer Dribble | success rate (%): 91.3 -> 93.4 | No clear effect |
| aggregate 2 views instead of 1 view at inference (ablation) | Mimic2DM · AIST++ test | 3D tracking error: 254.6 -> 164.9 | Improved |
| aggregate 3 views instead of 2 views at inference (ablation) | Mimic2DM · AIST++ test | 3D tracking error: 164.9 -> 161.5 | Improved |
| train with 2D data from 2 views vs 3D-supervised baseline | Mimic2DM · AIST++ test | 3D tracking error: 141 -> 164.9 | Hurt |
| use autoregressive 2D generator instead of diffusion generator | hierarchical controller · AIST++ | FID: 5.92 -> 2.44 | Improved |
| use autoregressive 2D generator instead of diffusion generator | hierarchical controller · Soccer Dribble | success rate: 0.13 -> 0.71 | Improved |
| train on diverse viewpoints instead of homogeneous viewpoints (ablation) | view-agnostic policy · lift-the-box imitation | qualitative naturalness | Improved |
| add adaptive state initialization (ablation) | single-view tracking training · initial state refinement | qualitative feasibility | Improved |
Closest papers by meaning
- InterMimic: Towards Universal Whole-Body Control for Physics-Based Human-Object Interactions (CVPR 2025 · similarity 0.88)
- 3DTrajMaster: Mastering 3D Trajectory for Multi-Entity Motion in Video Generation (ICLR 2025 · similarity 0.88)
- Object-centric 3D Motion Field for Robot Learning from Human Videos (NeurIPS 2025 · similarity 0.87)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.87)
- Move-in-2D: 2D-Conditioned Human Motion Generation (CVPR 2025 · similarity 0.87)
- VideoMDM: Towards 3D Human Motion Generation From 2D Supervision (NeurIPS 2026 · similarity 0.87)
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.
