Research map / Reinforcement learning / Policy gradient and actor-critic
GAS: Enhancing Reward-Cost Balance of Generative Model-assisted Offline Safe RL
Zifan LIU, Xinran Li, Shibo Chen, Jun Zhang
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 expectile regression (alpha>0.5) to the goal functions (ablation) | GAS · Bullet-Safety-Gym | reward/cost | Improved |
| add temporal segmented return augmentation (TSRA) (ablation) | GAS · Bullet-Safety-Gym | reward/cost | Improved |
| add transition-level return relabeling (ablation) | GAS · Bullet-Safety-Gym (20% threshold) | cost under varied reward targets | Improved |
| increase relabel level delta (ablation) | GAS · Bullet-Safety-Gym (20% threshold) | robustness to imbalanced targets | Improved |
| raise expectile alpha to 0.99 (ablation) | GAS · DroneRun | safety | Hurt |
| vary expectile alpha from 0.6 to 0.9 (ablation) | GAS · Bullet-Safety-Gym | reward/cost | No clear effect |
| use very large dataset reshape threshold q% (ablation) | GAS · DroneCircle | reward/cost | Hurt |
| vary dataset sampling probability epsilon (ablation) | GAS · DroneCircle | final reward/cost | No clear effect |
| increase CDT memory length K from 1 to 10 (ablation) | CDT · CarCircle | reward/cost | No clear effect |
| replace CDT with GAS | CDT · Bullet-Safety-Gym (tight thresholds, 8-task avg) | normalized cost: 1.12 -> 0.67 | Improved |
| replace CDT with GAS | CDT · Bullet-Safety-Gym (loose thresholds, 8-task avg) | normalized reward: 0.80 -> 0.86 | Improved |
| replace CDT transformer with MLP-based GAS | CDT · CarCircle | training time: 11h32min -> 8h20min | Improved |
Closest papers by meaning
- Constraint-Adaptive Policy Switching for Offline Safe Reinforcement Learning (AAAI 2025 · similarity 0.88)
- Online Optimization for Offline Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Offline Safe Reinforcement Learning Using Trajectory Classification (AAAI 2025 · similarity 0.88)
- Boundary-to-Region Supervision for Offline Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.87)
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.
