Research map / Reinforcement learning / Imitation learning
Abstracting Robot Manipulation Skills via Mixture-of-Experts Diffusion Policies
Ce Hao, Xuanran Zhai, Yaohua Liu, Harold Soh
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 SMP instead of Sparse Diffusion Policy (FFN-MoE experts) | Sparse DP · RoboTwin-2 multi-task avg | success rate: 0.44 -> 0.54 | Improved |
| use SMP instead of scaling to RDT (10x parameters) | RDT · RoboTwin-2 multi-task avg | success rate: 0.48 -> 0.54 | Improved |
| scale DP to RDT (about 10x parameters) | DP · RoboTwin-2 multi-task avg | success rate: 0.29 -> 0.48 | Improved |
| adaptive expert activation (SMP activates about 30% of its parameters) | SMP · inference cost | inference time (ms): 107.3 vs RDT 183.1; active params 80.2M of 258.9M | Improved |
| increase expert activation budget (ablation) | SMP · reconstruction and latency | reconstruction quality vs latency | Mixed |
Closest papers by meaning
- Goal-Oriented Skill Abstraction for Offline Multi-Task Reinforcement Learning (ICML 2025 · similarity 0.89)
- Active Fine-Tuning of Multi-Task Policies (ICML 2025 · similarity 0.89)
- Two-Steps Diffusion Policy for Robotic Manipulation via Genetic Denoising (NeurIPS 2025 · similarity 0.88)
- RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation (ICLR 2025 · similarity 0.88)
- Learning Parameterized Skills from Demonstrations (NeurIPS 2025 · similarity 0.88)
- Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training (NeurIPS 2025 · similarity 0.88)
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.
