Research map / Probabilistic and scientific modeling / Molecules and proteins
TapSampling: Inference-Time Sampling with a Task-Progress-Understanding Verifier for Robotic Manipulation
Sizhe Zhao, Shengping Zhang, Shuo Yang, Weiyu Zhao, Shuigen Wang, Xiangyang Ji
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 TapSampling verifier-based sampling | Diffusion Policy · CALVIN ABC->D | average success length: 2.41 -> 2.58 | Improved |
| add TapSampling | OpenVLA · CALVIN ABC->D | average success length: 3.30 -> 3.51 | Improved |
| add TapSampling | VPP · CALVIN ABC->D | average success length: 4.39 -> 4.46 | Improved |
| add TapSampling | pi0.5 · LIBERO-Long | average success rate: 96.8 -> 98.0 | Improved |
| add TapSampling | pi0 · real-world tasks | average success rate: 78.3 -> 83.3 | Improved |
| use Learned Posterior Sampling instead of Gaussian Sampling (ablation) | TapSampling on VPP · CALVIN ABC->D | average success length: 4.43 -> 4.46 | Improved |
| use Learned Posterior Sampling instead of Gaussian Sampling (ablation) | TapSampling on VPP · CALVIN ABC->D | 5th task success rate: 79.9 -> 81.1 | Improved |
| use Policy Sampling instead of Learned Posterior Sampling (ablation) | TapSampling on VPP · CALVIN ABC->D | average success length: 4.46 -> 4.50 | Improved |
| use Policy Sampling instead of Learned Posterior Sampling (ablation) | TapSampling on VPP · CALVIN ABC->D | latency (s): 0.488 -> 2.638 | Hurt |
| use Learned Posterior Sampling instead of Gaussian Sampling (ablation) | TapSampling · action sampling MMD gamma=2.0 | MMD: 0.098 -> 0.064 | Improved |
| allow retry attempts (oracle) on VPP (ablation) | VPP · CALVIN ABC->D | average success length: 4.39 -> 4.72 | Improved |
| add TapSampling | pi0.5 · LIBERO-Long | task 1 and 8 success rate: 98.0 -> 96.0 | Mixed |
Closest papers by meaning
- Bidirectional Decoding: Improving Action Chunking via Guided Test-Time Sampling (ICLR 2025 · similarity 0.90)
- Chain-of-Action: Trajectory Autoregressive Modeling for Robotic Manipulation (NeurIPS 2025 · similarity 0.89)
- Two-Steps Diffusion Policy for Robotic Manipulation via Genetic Denoising (NeurIPS 2025 · similarity 0.89)
- DynScene: Scalable Generation of Dynamic Robotic Manipulation Scenes for Embodied AI (CVPR 2025 · similarity 0.89)
- FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency (NeurIPS 2025 · similarity 0.88)
- Self-Improving Embodied Foundation Models (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.
