Research map / Reinforcement learning / Visual and representation RL
MoMaGen: Generating Demonstrations under Soft and Hard Constraints for Multi-Step Bimanual Mobile Manipulation
Chengshu Li, Mengdi Xu, Arpit Bahety, Hang Yin, Yunfan Jiang, Huang Huang, Josiah Wong, Sujay Garlanka, Cem Gokmen, Ruohan Zhang, Weiyu Liu, Jiajun Wu and 2 more
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 hard and soft visibility constraints (ablation) | MoMaGen · Tidy Table D0 object visibility | visibility ratio: 0.46 -> 0.86 | Improved |
| add hard and soft visibility constraints (ablation) | MoMaGen · Clean Frying Pan D1 object visibility | visibility ratio: 0.43 -> 0.80 | Improved |
| add hard and soft visibility constraints (ablation) | MoMaGen · Tidy Table D0 data generation | success rate: 0.74 -> 0.80 | Improved |
| add visibility constraints (ablation) | MoMaGen · Pick Cup D0 data generation | success rate: 0.97 -> 0.86 | Hurt |
| add hard visibility constraint (ablation) | MoMaGen w/o hard · Clean Frying Pan D0 data generation | success rate: 0.24 -> 0.51 | Improved |
| add soft visibility constraint (ablation) | MoMaGen w/o soft · Put Dishes Away D0 data generation | success rate: 0.50 -> 0.38 | Hurt |
| use MoMaGen instead of SkillMimicGen with base replay | SkillMimicGen · Tidy Table D0 object visibility | visibility ratio: 0.40 -> 0.86 | Improved |
| use MoMaGen instead of SkillMimicGen | SkillMimicGen · Pick Cup D0 data generation | success rate: 1.00 -> 0.86 | Hurt |
| add visibility constraints to data generation (ablation) | WB-VIMA policy · Tidy Table D0 policy success | success rate: 0.05 -> 0.40 | Improved |
| add visibility constraints to data generation (ablation) | WB-VIMA policy · Pick Cup D0 policy success | success rate: 0.65 -> 0.75 | Improved |
| train on MoMaGen D0+ generated data instead of D0 baseline data | WB-VIMA policy · Pick Cup D1 policy success | success rate: 0 -> 0.25 | Improved |
| pretrain on MoMaGen sim data before real fine-tune | pi0 · Real-world Pick Cup | success rate (%): 0 -> 60 | Improved |
Closest papers by meaning
- RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation (ICLR 2025 · similarity 0.87)
- MoManipVLA: Transferring Vision-language-action Models for General Mobile Manipulation (CVPR 2025 · similarity 0.87)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.86)
- HumanoidGen: Data Generation for Bimanual Dexterous Manipulation via LLM Reasoning (NeurIPS 2025 · similarity 0.86)
- DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation (NeurIPS 2025 · similarity 0.86)
- Data Scaling Laws in Imitation Learning for Robotic Manipulation (ICLR 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.
