Research map / Image generation / Autoregressive and latent generation
AdaptDiffuser: Diffusion Models as Adaptive Self-evolving Planners
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 reward-guided synthetic data generation and discriminator filtering to Diffuser | Diffuser · Maze2D U-Maze | normalized score: 113.9 -> 135.1 ± 5.8 | Helped |
| Add reward-guided synthetic data generation and discriminator filtering to Diffuser | Diffuser · Maze2D Medium | normalized score: 121.5 -> 129.9 ± 4.6 | Helped |
| Add reward-guided synthetic data generation and discriminator filtering to Diffuser | Diffuser · Maze2D Large | normalized score: 123.0 -> 167.9 ± 5.0 | Helped |
| Self-evolve the planner with synthetic-data fine-tuning | Diffuser · MuJoCo locomotion, 9 tasks | normalized average return: 88.9 -> 89.6 ± 0.8; 103.3 -> 111.6 ± 2.0; 106.9 -> 108.2 ± 0.8; 42.8 -> 44.2 ± 0.6; 74.3 -> 96.6 ± 2.7; 79.6 -> 84.4 ± 2 | Mixed |
| Add self-evolution using synthetic data for the new task | Diffuser · KUKA Pick-and-Place setup 1 | success score: 28.16 ± 2.0 -> 36.03 ± 2.1 | Helped |
| Add self-evolution using synthetic data for the new task | Diffuser · KUKA Pick-and-Place setup 2 | success score: 35.25 ± 1.4 -> 39.00 ± 1.3 | Helped |
| Add a second synthetic-data generation and fine-tuning phase (ablation) | AdaptDiffuser · MuJoCo Med-Expert, 3 tasks | normalized average return: 89.3 ± 0.6 -> 89.6 ± 0.8; 110.7 ± 3.2 -> 111.6 ± 2.0; 107.7 ± 0.9 -> 108.2 ± 0.8 | Helped |
| Add a second synthetic-data generation and fine-tuning phase (ablation) | AdaptDiffuser · MuJoCo Medium, 3 tasks | normalized average return: 43.8 ± 0.5 -> 44.2 ± 0.6; 95.4 ± 3.4 -> 96.6 ± 2.7; 83.2 ± 3.5 -> 84.4 ± 2.6 | Helped |
| Use 20% of expert data (ablation) | AdaptDiffuser · Maze2D Large | score: 167.9 -> 112.5 | Hurt |
| Use 50% of expert data (ablation) | AdaptDiffuser · Maze2D Large | score: 167.9 -> 123.8 | Hurt |
| Use 50% expert data and 25% training steps rather than full training (ablation) | AdaptDiffuser · Maze2D Large | score: 107.9 -> 123.8 | Helped |
Closest papers by meaning
- Task-Agnostic Pre-training and Task-Guided Fine-tuning for Versatile Diffusion Planner (ICML 2025 · similarity 0.88)
- Simple Hierarchical Planning with Diffusion (ICLR 2024 · similarity 0.88)
- Model-based Diffusion for Trajectory Optimization (NeurIPS 2024 · similarity 0.88)
- SafeDiffuser: Safe Planning with Diffusion Probabilistic Models (ICLR 2025 · similarity 0.88)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.88)
- GenPlan: Generative Sequence Models as Adaptive Planners (AAAI 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.