Research map / Vision-language and multimodal / Robot action and navigation
EC-Flow: Enabling Versatile Robotic Manipulation from Action-Unlabeled Videos via Embodiment-Centric Flow
Yixiang Chen, Peiyan Li, Yan Huang, Jiabing Yang, Kehan Chen, Liang Wang
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 |
|---|---|---|---|
| add goal image prediction auxiliary task (ablation) | EC-Flow · Meta-World (9 tasks) | average success rate: 0.636 -> 0.720 | Improved |
| add point filtering before action calculation (ablation) | EC-Flow · Meta-World (9 tasks) | average success rate: 0.582 -> 0.720 | Improved |
| track full-arm embodiment points instead of only end-effector points (ablation) | EC-Flow · Meta-World (9 tasks) | average success rate: 0.604 -> 0.720 | Improved |
| predict flow end-to-end instead of video prediction plus off-the-shelf tracking (ablation) | EC-Flow · Meta-World (9 tasks) | average success rate: 0.667 -> 0.720 | Improved |
| embodiment-centric flow instead of object-centric flow | EC-Flow vs Track2Act · Meta-World (9 tasks) | average success rate: 0.556 -> 0.720 | Improved |
| embodiment-centric flow instead of object-centric flow | EC-Flow vs Track2Act · Real-world manipulation (7 tasks) | successful trials out of 70: 21/70 -> 54/70 | Improved |
| full-joint optimization instead of end-effector-only regression in action calculation (ablation) | EC-Flow · Meta-World and real world | success rate: 66.7% -> 72.0% (sim), 70.0% -> 77.1% (real) | Improved |
| full-joint optimization instead of end-effector-only regression in action calculation (ablation) | EC-Flow · Meta-World and real world | inference latency: 0.21 s -> 0.37 s | Hurt |
| add 5 human videos to 2 robot demos | EC-Flow · Meta-World door-open and door-close | success rate: 46% -> 52% | Improved |
| add 50 human videos to 2 robot demos | EC-Flow · Meta-World door-open and door-close | success rate: 46% -> 70% | Improved |
| human videos alone without robot demos | EC-Flow · Meta-World door-open and door-close | success rate: 0% vs 46% | Hurt |
| use fewer sampled points N_p (ablation) | EC-Flow · Meta-World (9 tasks) | average success rate | Hurt |
Closest papers by meaning
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.89)
- VidBot: Learning Generalizable 3D Actions from In-the-Wild 2D Human Videos for Zero-Shot Robotic Manipulation (CVPR 2025 · similarity 0.89)
- FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025 · similarity 0.88)
- Object-centric 3D Motion Field for Robot Learning from Human Videos (NeurIPS 2025 · similarity 0.88)
- FlowHOI: Flow-based Semantics-Grounded Generation of Hand-Object Interactions for Dexterous Robot Manipulation (NeurIPS 2026 · similarity 0.88)
- Meta-Controller: Few-Shot Imitation of Unseen Embodiments and Tasks in Continuous Control (NeurIPS 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.
