AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Off-Policy Safe Reinforcement Learning with Cost-Constrained Optimistic Exploration

Guopeng Li, Matthijs T. J. Spaan, Julian F. P. Kooij

ICLR 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Safety layers and shielding

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 changedWhereMeasuredEffect
cost-constrained exploration (COX) instead of ORAC-style exploration (ablation)TQC-based COX-Q · SMARTS Overtakingtimeouts at test: 887 -> 0Improved
COX exploration instead of ORAC-style exploration (ablation)TQC-based COX-Q · SMARTS Overtakingunsafe events in data collection: 3215 -> 1665Improved
COX exploration instead of ORAC-style exploration (ablation)TQC-based COX-Q · SMARTS Intersectioncollisions at test: 18 -> 12Improved
COX exploration instead of ORAC-style exploration (ablation)TQC-based COX-Q · SMARTS Overtakingcollisions at test: 97 -> 99No clear effect
COX-Q vs on-policy CPPOPIDCOX-Q · SMARTS Intersectioncollisions at test: 183 -> 12Improved
COX-Q vs CAL on data-collection unsafe eventsCOX-Q · SMARTS Intersectionunsafe events in data collection: 739 -> 1123Hurt
add TQC truncated quantile critics (ablation)COX-Q · Safe Velocity and Safe NavigationreturnImproved
add cost-constrained exploration on top of TQC (ablation)COX-Q · Safe Velocitytraining costImproved
add COX or ORAC exploration on top of TQC (ablation)COX-Q · Safe Navigationreturn and costNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy