Research map / Reinforcement learning / Visual and representation RL
Ethics-Aware Safe Reinforcement Learning for Rare-Event Risk Control in Interactive Urban Driving
Dianzhao Li, Ostap Okhrin
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 |
|---|---|---|---|
| add dynamic risk-sensitive PER (ablation) | EthicAR · Waymo held-out scenarios | constraint compliance on risky steps (cost limits 0.6-1): below 30% -> 90~95% | Improved |
| add LSTM to the SACLag networks (ablation) | EthicAR · Waymo held-out scenarios | constraint compliance on risky steps (cost limits 0.6-1): 60~70% -> 90~95% | Improved |
| add Lagrangian safety constraint instead of reward penalty (ablation) | EthicAR · Waymo held-out scenarios | constraint compliance on risky steps (cost limits 0.6-1): <15% -> 90~95% | Improved |
| add ethical cost training over reward-only standard mode (ablation) | LSTMSAC · Waymo held-out scenarios (other risk) | other risk: 0.398 -> 0.158 | Improved |
| train EthicAR with ethical cost instead of selfish cost (ablation) | EthicAR · Waymo held-out scenarios (other risk, cost limit 0.6) | other risk: 0.099 -> 0.071 | Improved |
| train EthicAR with ethical cost instead of selfish cost (ablation) | EthicAR · Waymo held-out scenarios (ego risk, cost limit 0.6) | ego risk: 0.041 -> 0.029 | Improved |
| tighten cost limit from 1 to 0.6 (ablation) | EthicAR (ethical mode) · Waymo held-out scenarios | other risk: 0.099 -> 0.071 | Improved |
| add selfish cost over reward-only standard mode (ablation) | LSTMSAC · Waymo held-out scenarios (other risk) | other risk: 0.398 -> 0.164 | Improved |
| remove PER from EthicAR (other risk at cost limit 0.6) (ablation) | EthicAR w/o PER · Waymo held-out scenarios (other risk, cost limit 0.6, ethical mode) | other risk: 0.071 -> 0.165 | Hurt |
| remove LSTM from EthicAR (other risk at cost limit 0.6) (ablation) | SACLAG · Waymo held-out scenarios (other risk, cost limit 0.6, ethical mode) | other risk: 0.071 -> 0.213 | Hurt |
| EthicAR ethical mode in worst-case TTC<2s and risk>0.25 cases (ablation) | EthicAR · Waymo held-out scenarios | critical TTC-risk cases | Improved |
Closest papers by meaning
- Safety through feedback in Constrained RL (NeurIPS 2024 · similarity 0.88)
- From Gambits to Assurances: Game-Theoretic Integration of Safety and Learning for Interactive Robotics (AAAI 2025 · similarity 0.87)
- DriveDPO: Policy Learning via Safety DPO For End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.87)
- SafeAuto: Knowledge-Enhanced Safe Autonomous Driving with Multimodal Foundation Models (ICML 2025 · similarity 0.87)
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Beyond Self-Play and Scale: A Behavior Benchmark for Generalization in Autonomous Driving (NeurIPS 2026 · similarity 0.86)
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.
