Research map / Reinforcement learning / Policy gradient and actor-critic
Evaluating Model-Free Reinforcement Learning toward Safety-Critical Tasks
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 |
|---|---|---|---|
| Combine penalty-based policy optimization with iterative safety projection (ablation) | USL · SpeedLimit | Episodic cost rate (%): 38.87\pm15.50 -> 0.63\pm0.55 | Helped |
| Add Stage 2 iterative safety projection (ablation) | USL · SpeedLimit | Episodic cost rate (%): 5.05\pm2.69 -> 0.63\pm0.55 | Helped |
| Add Stage 2 iterative safety projection (ablation) | USL · PathTracking | Episodic return: $242.32\pm 0.27$ -> $241.74\pm 0.80$ | Hurt |
| Add Stage 1 penalty-based optimization (ablation) | USL · PathTracking | Episodic return: $211.88\pm18.28$ -> $241.74\pm 0.80$ | Helped |
| Use USL rather than TD3 | TD3 · SpeedLimit | Episodic cost rate (%): $95.34\pm1.03$ -> $0.63\pm0.55$ | Helped |
| Use USL rather than TD3 | TD3 · Stabilization | Episodic cost rate (%): $12.89\pm4.08$ -> $0.00\pm0.00$ | Helped |
| Use USL rather than TD3 | TD3 · PathTracking | Episodic cost rate (%): $48.40\pm9.49$ -> $0.17\pm0.18$ | Helped |
| Increase the penalty factor to a sufficiently large value | USL · SpeedLimit | Episodic cost | No clear effect |
| Set K=0, removing the post-projection iterations (ablation) | USL · SpeedLimit | Cost return | Hurt |
Closest papers by meaning
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.90)
- Safety Representations for Safer Policy Learning (ICLR 2025 · similarity 0.90)
- Balance Reward and Safety Optimization for Safe Reinforcement Learning: A Perspective of Gradient Manipulation (AAAI 2024 · similarity 0.90)
- Safe Reinforcement Learning with Instantaneous Constraints: The Role of Aggressive Exploration (AAAI 2024 · similarity 0.90)
- Feasibility Consistent Representation Learning for Safe Reinforcement Learning (ICML 2024 · similarity 0.90)
- A Provable Approach for End-to-End Safe Reinforcement Learning (NeurIPS 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.