Research map / Reinforcement learning / Policy gradient and actor-critic
Revisiting Domain Randomization via Relaxed State-Adversarial Policy Optimization
Yun-Hsuan Lien, Ping-Chun Hsieh, Yu-Shuen Wang
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 |
|---|---|---|---|
| relaxed state-adversarial training (RAPPO) instead of worst-case state adversary (SCPPO) | PPO · Hopper-v2, Walker2d-v2, HalfCheetah-v2 at larger attack strengths | episode reward under state attacks | Improved |
| relaxed state-adversarial training (RAPPO) instead of worst-case state adversary (SCPPO) | PPO · Ant-v2 and Humanoid-v2 under nominal and attacked settings | episode reward under state attacks | No clear effect |
| relaxed state-adversarial training (RAPPO) vs. MRPO domain randomization | PPO · environment size and gravity perturbations (0.6 to 1.4) | mean reward difference vs MRPO | Improved |
| relaxed state-adversarial training under combined attacks | PPO · HalfCheetah-v2 and Humanoid-v2 combined environment + state attacks | reward: HalfCheetah 3124 vs PPO 1568; Humanoid 2644 vs SCPPO 1035 | Improved |
| adaptive online relaxation parameter alpha vs fixed alpha=0.5 or linear schedules (ablation) | RAPPO · HalfCheetah-v2, Hopper-v2, Walker2d-v2, Humanoid-v2 state attacks | episode reward under attack: HalfCheetah sigma=0.025: 846 vs 406 (fixed alpha) | Improved |
| relaxed state-adversarial extension applied to SAPPO (observation adversary) | SAPPO · multiple MuJoCo environments under strong attacks | episode reward | Improved |
| increasing the relaxation/mixing effect (larger alpha) vs worst-case only | RAPPO · HalfCheetah-v2 nominal | reward | No clear effect |
Closest papers by meaning
- Statistical Guarantees for Offline Domain Randomization (ICLR 2026 · similarity 0.91)
- Domain Randomization via Entropy Maximization (ICLR 2024 · similarity 0.89)
- Diffusion Guided Adversarial State Perturbations in Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- Adversarial Diffusion for Robust Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Flow-based Domain Randomization for Learning and Sequencing Robotic Skills (ICML 2025 · similarity 0.85)
- Learning to Trust Bellman Updates: Selective State-Adaptive Regularization for Offline RL (ICML 2025 · similarity 0.85)
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.
