Research map / Reinforcement learning / Policy gradient and actor-critic
Safe Reinforcement Learning From Pixels Using a Stochastic Latent Representation
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 |
|---|---|---|---|
| Use Safe SLAC instead of LAMBDA | pixel-based safe RL · PointGoal1 | training time: 10 hours and 14 minutes vs 13 hours and 20 minutes | Helped |
| Use inferred latent state rather than truncated history as actor input (ablation) | Safe SLAC · PointGoal2 | reward return | Helped |
| Increase safety Lagrange multiplier learning rate from 2*10^{-6} to 6*10^{-6} (ablation) | Safe SLAC · DoggoGoal1 | cost return: 2*10^{-6} to 6*10^{-6} | Helped |
| Increase safety Lagrange multiplier learning rate from 2*10^{-6} to 6*10^{-6} (ablation) | Safe SLAC · DoggoGoal1 | reward return: 2*10^{-6} to 6*10^{-6} | Hurt |
| Use image reconstruction noise variance 0.4 rather than low variance (ablation) | Safe SLAC · SafetyGym SG6 | performance: 0.4 | Helped |
| Use Safe SLAC with 2 million environment steps rather than the 1 million-step comparison | Safe SLAC · PointGoal2 | reward return: 2 million; 1 million; approximately 40% more samples | Helped |
| Use Safe SLAC rather than LAMBDA | pixel-based safe RL · PointGoal1 and CarGoal1 | reward return | Helped |
| Use a safety critic and cost-constrained policy training | Safe SLAC · SafetyGym SG6 | cost return | Mixed |
| Use Safe SLAC rather than sensor-observation baselines | pixel-based safe RL · PointPush1 | reward return | Hurt |
Closest papers by meaning
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.91)
- Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis (AAAI 2024 · similarity 0.90)
- Probabilistic Shielding for Safe Reinforcement Learning (AAAI 2025 · similarity 0.90)
- Safety Representations for Safer Policy Learning (ICLR 2025 · similarity 0.90)
- Long-Term Safe Reinforcement Learning with Binary Feedback (AAAI 2024 · similarity 0.89)
- Sampling-Based Safe Reinforcement Learning (NeurIPS 2026 · 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.