Research map / Image generation / Autoregressive and latent generation
Compositional Diffusion with Guided search for Long-Horizon Planning
Utkarsh Aashu Mishra, David He, Yongxin Chen, Danfei Xu
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 iterative resampling and likelihood pruning (PR) to compositional sampling (ablation) | GSC · OGBench AntMaze Stitch Giant | success rate: 20 -> 85 (w/o PR 82) | Improved |
| add PR (ablation) | CDGS · OGBench PointMaze Stitch Giant | success rate: 78 -> 87 | Improved |
| add PR (ablation) | CDGS · OGBench HumanoidMaze Stitch Giant | success rate: 47 -> 55 | Improved |
| add PR on HumanoidMaze Giant versus CompDiffuser | CDGS · OGBench HumanoidMaze Stitch Giant | success rate: 64 (CD) -> 55 | Hurt |
| add PR (ablation) | CDGS · OGBench Scene Play | success rate: 36 -> 51 | Improved |
| add PR on medium-size mazes (ablation) | CDGS · OGBench AntMaze Medium | success rate: 97 -> 97 | No clear effect |
| add PR on AntSoccer (ablation) | CDGS · OGBench AntSoccer Arena (17D) | success rate: 67 -> 69 | No clear effect |
| add PR (ablation) | CDGS · TAMP Rearrangement Memory Task 1 | success rate: 0.11 -> 0.42 | Improved |
| add PR (ablation) | CDGS · TAMP Hook Reach Task 1 | success rate: 0.24 -> 0.64 | Improved |
| scale batch size B and resampling steps U (ablation) | CDGS · TAMP horizon 7 | success rate | Improved |
| add PR to panorama composition (ablation) | CDGS · Panorama | Intra-Style-L: 1.97 -> 1.38 (x10^-2) | Improved |
| add PR to long video composition | CDGS · Long video 350 frames | subject consistency: 89.51 -> 91.67 (GSC to CDGS) | Improved |
Closest papers by meaning
- Coarse-to-Fine Compositional Diffusion for Long-Horizon Planning (NeurIPS 2026 · similarity 0.93)
- Generative Trajectory Stitching through Diffusion Composition (NeurIPS 2025 · similarity 0.89)
- Generate Subgoal Images before Act: Unlocking the Chain-of-Thought Reasoning in Diffusion Model for Robot Manipulation with Multimodal Prompts (CVPR 2024 · similarity 0.88)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.88)
- Role Bias in Diffusion Models: Diagnosing and Mitigating through Intermediate Decomposition (NeurIPS 2025 · similarity 0.88)
- RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · 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.
