Research map / Reinforcement learning / Visual and representation RL
Safe Reinforcement Learning via Shielding under Partial Observability
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 a shield during learning | REINFORCE · Six domains | safety violations per episode: 3153 -> 0 | Helped |
| Deactivate the shield suddenly after initial shielded training (ablation) | REINFORCE · Six domains | safety violations per episode: 0 -> 1867 | Hurt |
| Gradually reduce the probability of applying the shield (ablation) | REINFORCE · Six domains | safety violations per episode: 1867 -> 27 | Helped |
| Add a state estimator providing belief-support inputs (ablation) | REINFORCE · Obstacle | convergence rate | Helped |
| Use a shield | REINFORCE · Six partially observable sparse-reward domains | average reward | Helped |
| Remove the shield | REINFORCE · Six partially observable domains | safety | Hurt |
| Make the domain fully observable (ablation) | REINFORCE · Six domains | convergence rate | Helped |
| Use a dense reward function (ablation) | REINFORCE · Six domains | final performance | Hurt |
| Use a shield | Random policy · Obstacle | steps before crash: approximately 12 -> approximately 47 | Helped |
| Use a shield | Random policy · Evade | chance of reaching the goal: approximately 1% -> 8% | Helped |
Closest papers by meaning
- Explainably Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.92)
- Probabilistic Shielding for Safe Reinforcement Learning (AAAI 2025 · similarity 0.91)
- Sampling-Based Safe Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- Latent Q-Barrier Shielding for Safe In-Context Reinforcement Learning (NeurIPS 2026 · similarity 0.89)
- Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis (AAAI 2024 · similarity 0.89)
- Dynamic Model Predictive Shielding for Provably Safe Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
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.