Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Safe Reinforcement Learning via Curriculum Induction
Matteo Turchetta, Andrey Kolobov, Shital Shah, Andreas Krause, Alekh Agarwal
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 |
|---|---|---|---|
| Use a curriculum optimized across students instead of no intervention | CISR on Frozen Lake · Frozen Lake | Success: 0.768±0.028 -> 0.960±0.004 | Improved |
| Use a curriculum optimized across students instead of no intervention | CISR on Frozen Lake · Frozen Lake | Training failures: 3075.6±492.1 -> 0±0 | Improved |
| Use an optimized Narrow-to-Wide curriculum instead of no intervention | CISR on Lunar Lander · Lunar Lander, noiseless 2-layer student | Training failures: 1251.0±33.72 -> 1.7±0.13 | Improved |
| Use an optimized Narrow-to-Wide curriculum instead of no intervention | CISR on Lunar Lander · Lunar Lander, noiseless 2-layer student | Return: 225.1±7.4 -> 236.1±0.9 | No clear effect |
| Use an optimized curriculum instead of no intervention | CISR on Lunar Lander · Lunar Lander, noisy 2-layer student | Training failures: 1695.8±31.04 -> 2.6±0.23 | Improved |
| Transfer the optimized curriculum to a 1-layer student | CISR on Lunar Lander · Lunar Lander, noiseless 1-layer student | Training failures: 1175.0±80.56 -> 1.9±0.13 | Improved |
| Use Narrow rather than the optimized Narrow-to-Wide curriculum (ablation) | CISR on Lunar Lander · Lunar Lander, noiseless 2-layer student | Return: 187.2±9.0 vs 236.1±0.9 | Hurt |
| Use Wide rather than the optimized Narrow-to-Wide curriculum (ablation) | CISR on Lunar Lander · Lunar Lander, noiseless 2-layer student | Return: 199.5±8.3 vs 236.1±0.9 | Hurt |
| Use HR rather than an optimized curriculum (ablation) | CISR on Frozen Lake · Frozen Lake | Return: -0.222±0.013 vs 5.368±0.025 | Hurt |
| Use SR1 rather than an optimized curriculum (ablation) | CISR on Frozen Lake · Frozen Lake | Return: 4.839±0.065 vs 5.368±0.025 | Hurt |
Closest papers by meaning
- Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis (AAAI 2024 · similarity 0.88)
- Safety Representations for Safer Policy Learning (ICLR 2025 · similarity 0.87)
- Real-DRL: Teach and Learn in Reality (NeurIPS 2025 · similarity 0.87)
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Probabilistic Shielding for Safe Reinforcement Learning (AAAI 2025 · similarity 0.87)
- Sampling-Based Safe Reinforcement Learning (NeurIPS 2026 · 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.
