Research map / Reinforcement learning / Offline RL
RORL: Robust Offline Reinforcement Learning via Conservative Smoothing
Rui Yang, Chenjia Bai, Xiaoteng Ma, Zhaoran Wang, Chongjie Zhang, Lei Han
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 conservative smoothing for perturbed states and uncertainty-based OOD value penalties | SAC-10, producing RORL · D4RL Gym tasks | normalized average return: 50.8 -> 85.7 | Improved |
| Add conservative smoothing and OOD value penalties | PBRL · D4RL Gym tasks | normalized average return: 74.4 -> 85.7 | Improved |
| Add conservative smoothing and OOD value penalties | EDAC-10 · D4RL Gym tasks | normalized average return: 71.2 -> 85.7 | Improved |
| Add conservative smoothing constraints | CQL · walker-medium-v2 under adversarial observation perturbations | normalized return | Improved |
| Remove the OOD loss (ablation) | RORL · walker2d-medium-v2 under adversarial observation attacks | normalized return | Hurt |
| Remove policy smoothing (ablation) | RORL · walker2d-medium-v2 under adversarial observation attacks | normalized return | Hurt |
| Remove Q smoothing (ablation) | RORL · walker2d-medium-v2 under adversarial observation attacks | normalized return | Hurt |
| Use adversarially perturbed observations during evaluation | RORL versus EDAC and SAC-10 · HalfCheetah-medium-v2, Walker2d-medium-v2, and Hopper-medium-v2 | normalized return | Improved |
| Use mixed-order rather than vanilla zero-order optimization for attacks | Adversarial evaluation of ensemble-based offline RL · HalfCheetah-medium-v2, Walker2d-medium-v2, and Hopper-medium-v2 | normalized return | Hurt |
Later papers that compared with this method
5 later accepted papers report RORL as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Variational OOD State Correction for Offline Reinforcement Learning (AAAI 2026 · better on 2 of 3 shared results)
- Policy-Driven World Model Adaptation for Robust Offline Model-based Reinforcement Learning (ICML 2026 · better on 1 of 1 shared result)
- Efficient Offline Reinforcement Learning via Peer-Influenced Constraint (ICLR 2026 · better on 1 of 1 shared result)
- Exploration and Anti-Exploration with Distributional Random Network Distillation (ICML 2024 · better on 2 of 3 shared results)
- Anti-Exploration by Random Network Distillation (ICML 2023 · better on 0 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
- NeoRL: A Near Real-World Benchmark for Offline Reinforcement Learning (NeurIPS 2022 · similarity 0.92)
- Optimal Conservative Offline RL with General Function Approximation via Augmented Lagrangian (ICLR 2023 · similarity 0.91)
- Towards Robust Offline Reinforcement Learning under Diverse Data Corruption (ICLR 2024 · similarity 0.91)
- COMBO: Conservative Offline Model-Based Policy Optimization (NeurIPS 2021 · similarity 0.91)
- A Minimalist Approach to Offline Reinforcement Learning (NeurIPS 2021 · similarity 0.91)
- OCEAN-MBRL: Offline Conservative Exploration for Model-Based Offline Reinforcement Learning (AAAI 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.
