Research map / Video / Video generation
Robotic Manipulation by Imitating Generated Videos Without Physical Demonstrations
Shivansh Patel, Shraddhaa Mohan, Hanlin Mai, Unnat Jain, Svetlana Lazebnik, Yunzhu Li
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 |
|---|---|---|---|
| filter generated videos with a GPT-4o check before execution (ablation) | RIGVid (Kling v1.6) · Sweeping trash | success rate: 20% -> 70% | Improved |
| filter generated videos with a VLM (ablation) | RIGVid (Kling v1.6) · Placing a spatula | success rate: 50% -> 90% | Improved |
| filter generated videos with a VLM (ablation) | RIGVid (Kling v1.6) · Pouring water | success rate: 80% -> 100% | Improved |
| use better video generator (Sora -> Kling v1.5 -> Kling v1.6) (ablation) | RIGVid · four real-world tasks | success rate: Sora 0% success; Kling v1.6 highest | Improved |
| replace real human demonstration videos with filtered generated videos (ablation) | RIGVid · four real-world tasks | success rate | No clear effect |
| use 6D object pose trajectories instead of VLM keypoint constraints (ReKep) | RIGVid · four real-world tasks | success rate: 50% -> 85% | Improved |
| use 6D pose tracking instead of point tracks (Gen2Act) (ablation) | RIGVid · four real-world tasks | success rate: 67.5% -> 85.0% | Improved |
| use 6D pose tracking instead of optical flow (AVDC), feature fields (4D-DPM) or goal-image tracks (Track2Act) (ablation) | RIGVid · four real-world tasks | success rate: 32.5% / 35.0% / 7.5% -> 85.0% | Improved |
| score videos with VBench++ video-text or I2V subject consistency instead of VLM filter (ablation) | video filtering · four tasks | Pearson correlation with human judgment: 0.34 / 0.37 vs 0.84 | Hurt |
| use monocular generated depth instead of real RGBD depth (ablation) | RIGVid · real videos | success rate: 100% -> 85% | Hurt |
| use MegaPose RGB-only tracking instead of FoundationPose (ablation) | RIGVid · pouring trajectories | rotational RMS jitter: 14.31 -> 37.47 degrees | Hurt |
| use mesh-free BundleSDF tracking instead of FoundationPose (ablation) | RIGVid · pouring | success rate: 90% success; ~30 minutes per video | No clear effect |
Closest papers by meaning
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.89)
- VideoVLA: Video Generators Can Be Generalizable Robot Manipulators (NeurIPS 2025 · similarity 0.89)
- Zero-Shot Robotic Manipulation with Pre-Trained Image-Editing Diffusion Models (ICLR 2024 · similarity 0.88)
- Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation (ICLR 2024 · similarity 0.88)
- OSVI-WM: One-Shot Visual Imitation for Unseen Tasks using World-Model-Guided Trajectory Generation (NeurIPS 2025 · similarity 0.88)
- Learning to Act from Actionless Videos through Dense Correspondences (ICLR 2024 · 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.
