Research map / Reinforcement learning / Imitation learning
Style-Agnostic Reinforcement Learning
Juyong Lee, Seokjun Ahn, Jaesik Park
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 the adversarial style perturbation loss (ablation) | SAR · Procgen Starpilot | test score: 27.44 -> 28.92 | Improved |
| add value similarity regularization G_V (ablation) | SAR · Procgen Starpilot | score std: ±7.79 -> ±4.60 | Improved |
| add value similarity regularization G_V (ablation) | SAR · Procgen Starpilot | mean test score: 29.28 -> 28.92 | No clear effect |
| add adversarial style perturbation on top of style mixing (ablation) | MixStyle · Procgen Starpilot | test score: 26.81 -> 28.92 | Improved |
| add random translation augmentation | SAR · Procgen Starpilot | test score: 28.92 -> 30.76 | Improved |
| add random color cutout augmentation | SAR · Procgen Starpilot | test score: 28.92 -> 33.72 | Improved |
| replace PPO with SAR (style mixing plus adversarial perturbation) | PPO · Procgen (8 tasks) | average rank: 4.9 -> 2.9 | Improved |
| replace SAC with SAR | SAC · Distracting Control Suite (4 tasks) | average rank: 2.125 -> 1.25 | Improved |
| replace SAC with SAR under zero noise | SAC · Distracting Control Suite walker:walk (zero noise) | return: 373 -> 325 | Hurt |
| delay the adversarial loss until after 10M steps | SAR · Procgen (8 tasks, 50M steps) | test score: 30.7±5.3 vs 27.6±7.9 | Mixed |
| delay the adversarial loss until after 25M steps | SAR · Procgen (8 tasks, 50M steps) | test score: 27.6±7.9 -> 34.0±13.9 | Mixed |
| delay the adversarial loss until after 300k steps | SAR · Distracting Control Suite (zero noise and distraction) | return: walker zero noise 325 -> 420; moderate 139 -> 113 | Mixed |
Closest papers by meaning
- Fourier Guided Adaptive Adversarial Augmentation for Generalization in Visual Reinforcement Learning (AAAI 2025 · similarity 0.87)
- VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- C-GAIL: Stabilizing Generative Adversarial Imitation Learning with Control Theory (NeurIPS 2024 · similarity 0.86)
- Diffusion Guided Adversarial State Perturbations in Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- SiT: Symmetry-invariant Transformers for Generalisation in Reinforcement Learning (ICML 2024 · similarity 0.84)
- Zero Shot Generalization of Vision-Based RL Without Data Augmentation (ICML 2025 · similarity 0.84)
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.
