Research map / Reinforcement learning / Offline RL
Offline Reinforcement Learning with Reverse Model-based Imagination
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 reverse model-based imagination data augmentation | BCQ · D4RL sparse maze2d-umaze | normalized return: 41.1 -> 139.5 | Helped |
| add reverse model-based imagination data augmentation | BCQ · D4RL play antmaze-medium | normalized return: 0.0 -> 35.3 | Helped |
| add reverse model-based imagination data augmentation | BCQ · D4RL fixed antmaze-umaze | normalized return: 75.3 -> 68.7 | Hurt |
| use reverse instead of forward imagination (ablation) | ROMI-BCQ · D4RL maze2d umaze/large and antmaze medium/large | normalized return: sparse maze2d-umaze 8.1 -> 139.5; diverse antmaze-large 17.8 -> 41.2 | Helped |
| use reverse instead of forward imagination (ablation) | ROMI-BCQ · D4RL sparse maze2d-medium | normalized return: 93.6 -> 82.4 | Hurt |
| use reverse instead of forward imagination (ablation) | ROMI-BCQ · D4RL diverse antmaze-umaze | normalized return: 66.8 -> 61.2 | Hurt |
| use reverse instead of forward imagination (ablation) | ROMI-CQL · D4RL medium hopper | normalized return: 1.9 -> 72.3 | Helped |
| use reverse instead of forward imagination (ablation) | ROMI-CQL · D4RL medium-expert halfcheetah | normalized return: 61.2 -> 86.8 | Helped |
| use CVAE rollout policy instead of reverse behavior cloning (ablation) | ROMI-BCQ · D4RL diverse antmaze-large | normalized return: 14.2 -> 41.2 | Helped |
| use CVAE rollout policy instead of reverse behavior cloning (ablation) | ROMI-BCQ · D4RL play antmaze-medium | normalized return: 33.8 -> 35.3 | No clear effect |
| increase rollout length from 5 to 10-20 (ablation) | ROMI-BCQ · D4RL maze2d | normalized return: dense maze2d-large 124.0 -> 97.5 (len 20); sparse maze2d-medium 82.4 -> 104.5 | Mixed |
| increase rollout length from 5 to 10 (ablation) | ROMI-BCQ · D4RL antmaze | normalized return: play antmaze-large 20.2 -> 0; diverse antmaze-umaze 61.2 -> 42.8 | Hurt |
Closest papers by meaning
- Double Check Your State Before Trusting It: Confidence-Aware Bidirectional Offline Model-Based Imagination (NeurIPS 2022 · similarity 0.93)
- Compositional Conservatism: A Transductive Approach in Offline Reinforcement Learning (ICLR 2024 · similarity 0.93)
- Constrained Latent Action Policies for Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.91)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · 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.