Research map / Reinforcement learning / Imitation learning
Dense Policy: Bidirectional Autoregressive Learning of Actions
Yue Su, Xinyu Zhan, Hongjie Fang, Han Xue, Hao-Shu Fang, Yong-Lu Li, Cewu Lu, Lixin Yang
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 |
|---|---|---|---|
| replace diffusion action head with Dense Policy action head (3D) | DP3 · Adroit/DexArt/MetaWorld | success rate (%): 53 -> 72 | Improved |
| replace diffusion action head with Dense Policy action head (2D) | Diffusion Policy · Adroit/DexArt/MetaWorld | success rate (%): 25 -> 52 | Improved |
| use bidirectional prediction instead of next-token or next-chunk autoregression (ablation) | Dense Policy · Door, Bin Picking, Shelf Place, Box Close | learning efficiency and success rate | Improved |
| 3D Dense Policy vs RISE on Pour Balls complete transfer | RISE · Pour Balls (real, 3D) | complete (%): 25 -> 60 | Improved |
| 3D Dense Policy vs RISE on Pour Balls poured at least one ball | RISE · Pour Balls (real, 3D) | poured (%): 95 -> 85 | Hurt |
| 3D Dense Policy vs RISE on Flower Arrangement | RISE · Flower Arrangement (real, 3D) | success rate (%): 50 -> 70 | Improved |
| 2D Dense Policy vs Diffusion Policy on Put Bread | Diffusion Policy · Put Bread (real, 2D) | success rate (%): 40 -> 55 | Improved |
| 2D Dense Policy vs Diffusion Policy on Open Drawer | Diffusion Policy · Open Drawer (real, 2D) | success rate (%): 20 -> 20 | No clear effect |
| 2D Dense Policy vs Diffusion Policy on Pour Balls | Diffusion Policy · Pour Balls (real, 2D) | balls poured out of 10: 2.35 -> 3.30 | Improved |
| 2D Dense Policy vs ACT on Put Bread | ACT · Put Bread (real, 2D) | success rate (%): 35 -> 55 | Improved |
| Dense Policy action head vs Diffusion Policy action head, inference speed | Diffusion Policy · action head inference | inference time: nearly ten times faster | Improved |
Closest papers by meaning
- Chain-of-Action: Trajectory Autoregressive Modeling for Robotic Manipulation (NeurIPS 2025 · similarity 0.91)
- FreqPolicy: Frequency Autoregressive Visuomotor Policy with Continuous Tokens (NeurIPS 2025 · similarity 0.90)
- Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution (NeurIPS 2025 · similarity 0.89)
- Learning to Act from Actionless Videos through Dense Correspondences (ICLR 2024 · similarity 0.88)
- FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency (NeurIPS 2025 · similarity 0.88)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 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.
