Research map / Reinforcement learning / Policy gradient and actor-critic
Accelerating Safe Reinforcement Learning with Constraint-mismatched Baseline Policies
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 SPACE’s reward optimization, adaptable distance constraint to the baseline, and cost projection | Safe RL with baseline policies · Five control tasks | constraint violations: 10 times fewer constraint violations | Helped |
| Use SPACE instead of the best baseline | Safe RL with baseline policies · Bottleneck and car-racing | reward: at least 10% more reward | Helped |
| Use SPACE instead of the tested baselines | Safe RL with baseline policies · All tested cases | cost constraint satisfaction speed: at least 2 times faster | Helped |
| Use SPACE rather than the tested baseline algorithms | Safe RL with baseline policies · All tested cases | constraint satisfaction | Helped |
| Use dynamic rather than fixed h_D (ablation) | SPACE · Circle | reward: 2.3 times more reward | Helped |
| Fix h_D instead of updating it dynamically (ablation) | SPACE · Gather and Circle | reward | Hurt |
| Learn from a higher-reward but constraint-violating baseline (ablation) | SPACE · Gather and Circle | initial reward learning efficiency | Helped |
| Learn from a baseline inside the cost-constraint set (ablation) | SPACE · Gather and Circle | later reward learning efficiency | Helped |
| Use a human demonstration baseline as guidance | SPACE · Car-racing | reward | Helped |
| Use fixed imitation regularization rather than SPACE’s projection and adaptable distance constraint | f-CPO and f-PCPO · Tested tasks | reward | Hurt |
| Fade the imitation regularization weight rather than update the distance constraint with SPACE’s rule | d-CPO and d-PCPO · Tested tasks | reward | Hurt |
Closest papers by meaning
- Breaking Safety Paradox with Feasible Dual Policy Iteration (ICLR 2026 · similarity 0.92)
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.91)
- Safe Reinforcement Learning with Instantaneous Constraints: The Role of Aggressive Exploration (AAAI 2024 · similarity 0.91)
- Constraint-Adaptive Policy Switching for Offline Safe Reinforcement Learning (AAAI 2025 · similarity 0.90)
- Constrained Reinforcement Learning Under Model Mismatch (ICML 2024 · similarity 0.90)
- Safety Representations for Safer Policy Learning (ICLR 2025 · similarity 0.90)
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.