Research map / Reinforcement learning / Policy gradient and actor-critic
Safe Reinforcement Learning via Shielding
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 post-posed shielding with action ranking of size 3 and no penalty for corrected actions (ablation) | Q-learning · 15x9 grid world | average reward | Helped |
| Use action ranking of size 1 instead of size 3 (ablation) | Post-posed shielding with Q-learning · 15x9 grid world | average reward | Hurt |
| Add penalties for corrected actions instead of using no penalty (ablation) | Post-posed shielding with Q-learning · 15x9 grid world | average reward | Hurt |
| Add post-posed shielding with action ranking of size 3 and penalties for corrected actions | Q-learning · 9x9 grid world | learning speed | Helped |
| Use unshielded Q-learning with large penalties for unsafe actions instead of shielding (ablation) | Q-learning · 9x9 grid world | maximum reward | Hurt |
| Add shielding without punishment | Deep Q-Network with Boltzmann exploration · Self-driving car | learning speed | Helped |
| Add shielding without punishment | Deep Q-Network with Boltzmann exploration · Self-driving car | crashes | Helped |
| Add shielding enforcing the two safety properties | Deep reinforcement learning agent · Seaquest | performance | No clear effect |
| Add shielding enforcing the two safety properties | Deep reinforcement learning agent · Seaquest | safety violations | Helped |
| Add post-posed shielding | Q-learning and SARSA · Water tank | time to optimal policy | Helped |
| Add shielding | Tabular Q-learning · 9x9 and 15x9 grid worlds | negative rewards | Helped |
| Add shielding | Tabular Q-learning · 9x9 and 15x9 grid worlds | learning speed | Helped |
Closest papers by meaning
- Probabilistic Shielding for Safe Reinforcement Learning (AAAI 2025 · similarity 0.92)
- Explainably Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Latent Q-Barrier Shielding for Safe In-Context Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- Safety Representations for Safer Policy Learning (ICLR 2025 · similarity 0.88)
- Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis (AAAI 2024 · similarity 0.88)
- Pure-Past Action Masking (AAAI 2024 · similarity 0.88)
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.