Research map / Reinforcement learning / Offline RL
A Closer Look at Offline RL Agents
Yuwei Fu, Di Wu, Benoit Boulet
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 |
|---|---|---|---|
| replace AWR actor loss with SAC loss plus behavior log-likelihood term (RIQL) | IQL · D4RL total over 9 tasks | total normalized return: 690.04 -> 759.05 | Improved |
| RIQL vs IQL | IQL · hopper-medium-v2 | normalized return: 64.64 -> 91.58 | No clear effect |
| RIQL vs IQL | IQL · halfcheetah-medium-v2 | normalized return: 47.43 -> 55.93 | Improved |
| RIQL vs IQL | IQL · walker2d-medium-replay-v2 | normalized return: 73.53 -> 81.46 | Improved |
| augment IQL with model-generated data (MIQL) without sample selection | IQL · D4RL total over 9 tasks | total normalized return: 690.04 -> 334.16 | Hurt |
| augment TD3+BC with model-generated data (MTD3+BC) without sample selection | TD3+BC · D4RL total over 9 tasks | total normalized return: 676.82 -> 488.99 | Hurt |
| add uncertainty-based sample selection (USS) (ablation) | MIQL · D4RL total over 9 tasks | total normalized return: 334.16 -> 699.98 | Improved |
| add uncertainty-based sample selection (USS) (ablation) | MTD3+BC · D4RL total over 9 tasks | total normalized return: 488.99 -> 702.29 | Improved |
| add USS (ablation) | MIQL · hopper-medium-v2 | normalized return: 8.89 -> 80.79 | Improved |
| add USS (ablation) | MTD3+BC · walker2d-medium-v2 | normalized return: 85.33 -> 84.80 | No clear effect |
| add USS (ablation) | MIQL · walker2d-medium-expert-v2 | normalized return: 87.96 -> 108.25 | Improved |
| compare USS agents with model-free counterparts | MIQL-USS · D4RL tasks | normalized return: 690.04 -> 699.98 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- COMBO — COMBO: Conservative Offline Model-Based Policy Optimization (NeurIPS 2021) · better on 1 of 2 shared results
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
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.93)
- A Minimalist Approach to Offline Reinforcement Learning (NeurIPS 2021 · similarity 0.92)
- Is Value Learning Really the Main Bottleneck in Offline RL? (NeurIPS 2024 · similarity 0.92)
- ODRL: A Benchmark for Off-Dynamics Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- The Generalization Gap in Offline Reinforcement Learning (ICLR 2024 · similarity 0.91)
- Bayesian Design Principles for Offline-to-Online Reinforcement Learning (ICML 2024 · similarity 0.91)
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.
