Coarse-to-Fine Compositional Diffusion for Long-Horizon Planning
About
Diffusion models provide strong priors for generating structured data, but many tasks require outputs beyond the scale on which these models are typically trained. Compositional generation addresses this by composing overlapping local plans from a pretrained short-horizon prior into a long-horizon output. However, standard composition primarily enforces agreement between neighboring local plans, yielding local consistency without directly specifying the global structure of the full composition. As a result, locally compatible plans may still form an implausible route, task sequence, or temporal evolution. Existing methods improve global coherence by repeatedly propagating local consistency signals or by adding inference-time optimization, but these procedures become expensive as the number or dimensionality of local plans increases. We propose Coarse-to-Fine Compositional Diffusion (CoFi), an inference-time sampler that separates global structure formation from local detail refinement. CoFi first aligns local denoised estimates around a shared coarse structure, producing a global scaffold that captures the long-range task-level arrangement. It then diffuses this scaffold to an intermediate noise level and denoises it with the same pretrained local prior, restoring local fine structure while preserving the scaffold-induced global coherence. Across long-horizon robotic planning, panoramic image generation, and long video generation, CoFi not only improves both global coherence and local sample quality over prior compositional baselines, but also requires 2-8x fewer denoiser evaluations.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Robotic Planning | OGBench PointMaze Giant 48 (stitch) | Success Rate96 | 16 | |
| Robotic Planning | OGBench AntMaze Giant 48 (stitch) | Success Rate85 | 16 | |
| Robotic Planning | OGBench Scene 48 (play) | Success Rate0.63 | 16 | |
| Goal-oriented planning | OGBench PointMaze Medium Stitch v1 | Success Rate100 | 12 | |
| Robotic Planning | OGBench AntMaze-Stitch Large | Success Rate88 | 8 | |
| Robotic Planning | OGBench AntMaze-Stitch Medium | Success Rate97 | 8 | |
| Robotic Planning | OGBench Pointmaze Stitch Large | Success Rate100 | 8 | |
| Long Video Generation | VBench 273-frame | Subject Consistency94.11 | 5 | |
| Panoramic Image Generation | Panoramic image generation evaluation 512 x 4608 | LPIPS0.48 | 4 |