Research map / Reinforcement learning / Offline RL
Accelerating Diffusion Planners in Offline RL via Reward-Aware Consistency Trajectory Distillation
Xintong Duan, Yutong He, Fahim Tajwar, Ruslan Salakhutdinov, J Zico Kolter, Jeff Schneider
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 reward-aware distillation of student with unconditioned teacher (vs unconditioned student) (ablation) | RACTD · Walker2d medium | normalized score: 93.3 -> 118.8 | Improved |
| add reward model to teacher on top of reward-aware student (ablation) | RACTD · Walker2d medium | normalized score: 118.8 -> 94.5 | Hurt |
| add reward model to teacher only (ablation) | unconditioned student · Walker2d medium | normalized score: 93.3 -> 97.0 | Improved |
| add reward loss (weight 0.0 -> 0.8) (ablation) | RACTD student · Hopper medium-replay | normalized score: 50.8 -> 108.3 | Improved |
| raise reward loss weight too high (0.8 -> 2.0) (ablation) | RACTD student · Hopper medium-replay | normalized score: 108.3 -> 17.0 | Hurt |
| add DSM loss (weight 0.0 -> 1.0) (ablation) | RACTD student · Hopper medium-replay | normalized score: 1.5 -> 108.3 | Improved |
| add reward guidance to consistency trajectory distillation (CTD -> RACTD) (ablation) | consistency trajectory distillation · Maze2d average | normalized score: 123.4 -> 133.4 | Improved |
| use 2 denoising steps instead of 1 (ablation) | RACTD student · Hopper medium-replay | normalized score: 109.5 -> 109.8 | No clear effect |
| condition on 4 past observations instead of 1 (c=16) (ablation) | RACTD · Walker2d medium-replay | normalized score: 105.2 -> 108.0 | Improved |
| condition on 4 past observations instead of 1 (c=8) (ablation) | RACTD · Walker2d medium-replay | normalized score: 104.3 -> 103.3 | No clear effect |
| distill diffusion planner to single step vs multi-step Diffuser | RACTD · Gym-MuJoCo average | normalized score: 77.5 -> 96.4 | Improved |
| one-step DDIM instead of 10-step DDIM (ablation) | DDIM · Maze2d average | normalized score: 130.2 -> -0.2 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MOReL — MOReL: Model-Based Offline Reinforcement Learning (NeurIPS 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Prior-Guided Diffusion Planning for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Task-Agnostic Pre-training and Task-Guided Fine-tuning for Versatile Diffusion Planner (ICML 2025 · similarity 0.90)
- Planning, Fast and Slow: Online Reinforcement Learning with Action-Free Offline Data via Multiscale Planners (ICML 2024 · similarity 0.89)
- Reasoning with Latent Diffusion in Offline Reinforcement Learning (ICLR 2024 · similarity 0.89)
- Reward-Consistent Dynamics Models are Strongly Generalizable for Offline Reinforcement Learning (ICLR 2024 · similarity 0.89)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
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.
