Research map / 3D vision / Gaussian splatting
Video2Robo: 3DGS-based Synthetic Data from One Video Enables Scalable Robot Learning
Yinan Deng, Kejia Hu, Ye Chen, Jianyu Dou, Jiahui Wang, Jingyu Zhao, Haojia Ao, Yi Yang, Yufeng Yue
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 |
|---|---|---|---|
| use Video2Robo 3DGS tracking with CoTracker3 track loss instead of FoundationPose-O | 6D object localization · 6D localization, mean of 6 tasks | BOP Average Recall: 55.36 -> 78.41 | Improved |
| use Video2Robo tracking instead of MegaPose | 6D object localization · 6D localization, Stack | BOP Average Recall: 91.57 -> 90.90 | No clear effect |
| use 3DGS rendering and kinematic motion modeling instead of simulator-based MimicGen | data generation · 6 self-captured tasks | average time per sample (s): 17.23 -> 05.23 | Improved |
| use robot-object coupled motion instead of simulated physics for data generation | data generation · 6 self-captured tasks | generation success rate (%): 97.40 -> 100.00 | Improved |
| add texture, background, lighting and camera augmentation to MimicGen (ablation) | MimicGen · Simulation, various changes | success rate (%): 11.67 -> 29.67 | Improved |
| add appearance and view augmentation to MimicGen (ablation) | MimicGen · Simulation, only object poses change | success rate (%): 93.00 -> 25.67 | Hurt |
| train on Video2Robo data rather than MimicGen/SkillGen data | Diffusion Policy · Simulation, various changes | success rate (%): 44.67 -> 63.83 | Improved |
| train on Video2Robo data rather than MimicGen/SkillGen data | Diffusion Policy · Simulation, only object poses change | success rate (%): 93.00 -> 66.67 | Hurt |
| train on Video2Robo data instead of 20 real teleoperated demos | Diffusion Policy · Real world, various changes | success rate (%): 27.50 -> 58.33 | Improved |
| train on Video2Robo data instead of simulation baselines | Diffusion Policy · Real world, various changes | success rate (%): 5.83 -> 58.33 | Improved |
Closest papers by meaning
- RoboScape: Physics-informed Embodied World Model (NeurIPS 2025 · similarity 0.87)
- VidMan: Exploiting Implicit Dynamics from Video Diffusion Model for Effective Robot Manipulation (NeurIPS 2024 · similarity 0.86)
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.86)
- DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videos (NeurIPS 2025 · similarity 0.86)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.86)
- VidBot: Learning Generalizable 3D Actions from In-the-Wild 2D Human Videos for Zero-Shot Robotic Manipulation (CVPR 2025 · similarity 0.86)
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.
