Research map / Reinforcement learning / Offline RL
Speedup Patch: Learning a Plug-and-Play Policy to Accelerate Embodied Manipulation
Zhichao Wu, Junyin Ye, Zhilong Zhang, Yihao Sun, Haoxin Lin, Jiaheng Luo, Haoxiang Ren, Lei Yuan, Yang Yu
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 SuP scheduler instead of fixed ds2 downsampling | ACT · BiGym average | success rate: 0.61 -> 0.67 | Improved |
| add SuP scheduler instead of fixed ds2 downsampling | DP · BiGym average | success rate: 0.40 -> 0.51 | Improved |
| add SuP scheduler instead of fixed ds2 downsampling | pi0.5 · Libero average | success rate: 0.928 -> 0.973 | Improved |
| add SuP scheduler instead of fixed ds2 downsampling | VLA-Adapter · Libero average | success rate: 0.862 -> 0.937 | Improved |
| use SuP instead of ds3 downsampling | pi0.5 · Real-world average | success rate: 0.356 -> 0.611 | Improved |
| use SuP instead of DemoSpeedup retraining | pi0.5 · Real-world average | speedup: 2.07x -> 2.17x | Improved |
| add SuP scheduler to base policy | pi0.5 · Libero average | success rate: 0.969 -> 0.973 | No clear effect |
| add IQL-trained scheduler instead of greedy MPC (ablation) | SuP · BiGym Sandwich Remove | success rate: 0.54 -> 0.59 | Improved |
| add IQL-trained scheduler instead of greedy MPC (eps 0.02) (ablation) | SuP · BiGym Sandwich Remove | success rate: 0.35 -> 0.46 | Improved |
| raise deviation threshold from 0.015 to 0.02 (ablation) | SuP · BiGym Sandwich Remove | success rate: 0.64 -> 0.46 | Hurt |
| lower deviation threshold from 0.015 to 0.01 (ablation) | SuP · BiGym Sandwich Remove | success rate: 0.64 -> 0.59 | Hurt |
| add gripper action compensation to ds2 downsampling (ablation) | pi0.5 · Libero average | success rate: 0.842 -> 0.928 | Improved |
Closest papers by meaning
- PACT: Self-Evolving Physical Safety Alignment for Diffusion Policies in Embodied Manipulation (ICML 2026 · similarity 0.90)
- GENMANIP: LLM-driven Simulation for Generalizable Instruction-Following Manipulation (CVPR 2025 · similarity 0.89)
- Implicit Drifting Policy: One-Step Action Generation via Conditional Expert Geometry (NeurIPS 2026 · similarity 0.88)
- DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation (NeurIPS 2025 · similarity 0.88)
- ManiSkill-HAB: A Benchmark for Low-Level Manipulation in Home Rearrangement Tasks (ICLR 2025 · similarity 0.88)
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning (ICML 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.
