Research map / Reinforcement learning / Offline RL
Offline Reinforcement Learning via High-Fidelity Generative Behavior Modeling
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 |
|---|---|---|---|
| Use SfBC with a diffusion behavior model and in-sample planning instead of IQL | Offline RL · AntMaze-umaze Diverse | normalized score: 62.2 -> 85.3 ± 3.6 | Helped |
| Use SfBC instead of IQL | Offline RL · Bidirectional-Car Both-side | success rate: 15.7 -> 100.0 ± 0.0 | Helped |
| Replace diffusion behavior modeling with Gaussian modeling (ablation) | SfBC · AntMaze average | normalized score: 74.2 -> 69.4 | Hurt |
| Replace diffusion behavior modeling with VAE behavior modeling (ablation) | SfBC · AntMaze average | normalized score: 74.2 -> 59.7 | Hurt |
| Replace diffusion behavior modeling with Gaussian modeling (ablation) | SfBC · MuJoCo Locomotion average | normalized score: 75.6 -> 67.9 | Hurt |
| Replace diffusion behavior modeling with VAE behavior modeling (ablation) | SfBC · MuJoCo Locomotion average | normalized score: 75.6 -> 70.9 | Hurt |
| Add in-sample planning (ablation) | SfBC · AntMaze average | normalized score: 43.3 -> 74.2 | Helped |
| Add in-sample planning (ablation) | SfBC · MuJoCo Locomotion average | normalized score: 72.3 -> 75.6 | Mixed |
| Increase diffusion sampling steps from 5 to 15 (ablation) | SfBC · AntMaze average | normalized score: 5.5 -> 74.2 | Helped |
| Increase diffusion sampling steps from 15 to 25 (ablation) | SfBC · MuJoCo Locomotion average | normalized score: 75.6 -> 74.4 | No clear effect |
| Use a diffusion behavior model rather than a unimodal policy | Bidirectional-Car agent · Bidirectional-Car Both-side | success rate: 15.7 -> 100.0 ± 0.0 | Helped |
Closest papers by meaning
- Offline Reinforcement Learning with Generative Trajectory Policies (ICML 2026 · similarity 0.93)
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.92)
- Iteratively Refined Behavior Regularization for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- DeFlow: Decoupling Behavior-Prior Modeling and Value Maximization for Offline Policy Extraction (NeurIPS 2026 · similarity 0.90)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.90)
- Generative OOD-regularized Model-based Policy Optimization (NeurIPS 2026 · similarity 0.90)
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.