Research map / Vision-language and multimodal / Robot action and navigation
CycleManip: Enabling Cycle-based Manipulation via Effective History Perception and Understanding
Yi-Lin Wei, Haoran Liao, Yuhao Lin, Pengyue Wang, Zhizhao Liang, Guiliang Liu, Wei-Shi Zheng
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 cost-aware history sampling (dense proprioception, sparse binary/exponential visual frames) (ablation) | DP3 · real-world cyclic tasks (block hammering) | success rate (%): 37.5 -> 62.5 | Improved |
| add cost-aware history sampling (ablation) | DP3 · real-world drum beating (bi-gripper) | success rate (%): 0 -> 60 | Improved |
| add progress-prediction multi-task learning (historical understanding) (ablation) | CycleManip w/o Task · real-world block hammering | success rate (%): 62.5 -> 93.75 | Improved |
| add progress-prediction multi-task learning (ablation) | CycleManip w/o Task · real-world table cleaning (bi-dexterous) | success rate (%): 40 -> 100 | Improved |
| add progress-prediction multi-task learning (ablation) | CycleManip w/o Task · real-world tire pumping (humanoid) | success rate (%): 20 -> 50 | Improved |
| add progress-prediction multi-task learning (ablation) | CycleManip w/o Task · real-world bottle shaking | cycle count deviation: 1.31 -> 0.375 | Improved |
| replace Pi-0 baseline with CycleManip history encoding plugged into Pi-0 | Pi-0 · CycleManip sim: bottle shaking | success rate (%): 19 -> 72 | Improved |
| plug CycleManip into Pi-0 | Pi-0 · CycleManip sim: dual-knife chopping | success rate (%): 1 -> 41 | Improved |
| apply CycleManip to general manipulation tasks | DP3 · RoboTwin 2.0 Place Bread Basket | success rate (%): 26 -> 61 | Improved |
| apply CycleManip to general manipulation tasks | DP3 · RoboTwin 2.0 Turn Switch | success rate (%): 46 -> 64 | Improved |
| add CycleManip history sampling and multi-task head | DP3 · CycleManip sim: carrot cutting | efficiency: Time_test 0.0893 -> 0.0953; Suc. 38 -> 86 | Mixed |
| use baselines with short observation windows (Pi-0 with 1-frame observation) | Pi-0 · CycleManip sim: dual-knife chopping | success rate (%): 1 | Hurt |
Closest papers by meaning
- Expanding Spatial and Temporal Context for Robotic Imitation Learning With Scene Graphs (CVPR 2026 · similarity 0.87)
- RoboMME: Benchmarking and Understanding Memory for Robotic Generalist Policies (ICML 2026 · similarity 0.87)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.86)
- RoboProcessBench: Benchmarking Process-Aware Understanding in Visual-Language Robotic Manipulation (NeurIPS 2026 · similarity 0.86)
- What's the Move? Hybrid Imitation Learning via Salient Points (ICLR 2025 · similarity 0.86)
- ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning (CVPR 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.
