Research map / Probabilistic and scientific modeling / Dynamical systems
Learning High-Frequency Continuous Action Chunks in Latent Space
Kunyun Wang, Yuhang Zheng, Yupeng Zheng, Jieru Zhao, Wenchao Ding
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 |
|---|---|---|---|
| train policy in VAE latent space instead of action space | OFT · Peel Cucumber (real robot, sync) | success rate: 28% -> 74% | Improved |
| train policy in VAE latent space instead of action space | OFT · Write Board (real robot, sync) | jerk: 5.238 -> 0.558 | Improved |
| train policy in VAE latent space instead of action space | DP · Peel Cucumber (real robot, sync) | jerk: 2.057 -> 0.412 | Improved |
| train policy in VAE latent space instead of action space | PI0.5 · Peel Cucumber (real robot, sync) | success rate: 78% -> 84% | Improved |
| train policy in VAE latent space instead of action space | DP · Peel Cucumber (real robot, sync) | success rate: 90% -> 90% | No clear effect |
| train in latent space instead of action space | PI0.5 · dataset-based Write Board | deviation xyz: 1.24 -> 1.32 | Hurt |
| train in latent space instead of action space | OFT · dataset-based Write Board | deviation xyz: 7.59 -> 1.47 | Improved |
| train low-frequency policy and interpolate instead of latent high-frequency policy (ablation) | DP · Wipe Vase (real robot, sync) | jerk: 0.645 -> 4.143 | Hurt |
| add Reuse-then-Refine to latent policy under asynchronous inference (ablation) | PI0.5 · Peel Cucumber (real robot, async) | jerk: 3.608 -> 1.601 | Improved |
| apply RT-C in latent space instead of RTR (ablation) | PI0.5 · Chunk continuity, dataset-based (async) | boundary gap xyz: 6.842 -> 8.478 | Hurt |
| add RT-C to original action-space PI0.5 (ablation) | PI0.5 · Peel Cucumber (real robot, async) | jerk: 4.124 -> 4.697 | Hurt |
| add RTR to latent DP for end-to-end latency (ablation) | DP · Peel Cucumber (end-to-end latency, async) | latency (s): 18.07 -> 14.59 | Improved |
Closest papers by meaning
- FreqPolicy: Frequency Autoregressive Visuomotor Policy with Continuous Tokens (NeurIPS 2025 · similarity 0.89)
- PACE: Phase-Aware Chunk Execution for Robot Policies with Action Chunking (NeurIPS 2026 · similarity 0.89)
- Bidirectional Decoding: Improving Action Chunking via Guided Test-Time Sampling (ICLR 2025 · similarity 0.88)
- Real-Time Execution of Action Chunking Flow Policies (NeurIPS 2025 · similarity 0.88)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.88)
- FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency (NeurIPS 2025 · 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.
