MOPO: uncertainty-penalized learned MDPs for offline policy optimization

What changed

Modified rewards in a learned dynamics model by an uncertainty penalty and optimized policies in the penalized MDP, connecting conservative model use to conditional return bounds.

Scientific scope

Modified rewards in a learned dynamics model by an uncertainty penalty and optimized policies in the penalized MDP, connecting conservative model use to conditional return bounds. The lower-bound analysis requires an admissible model-error estimator and stated value-class assumptions, not merely a learned variance heuristic. Exact surrogate optimization differs from finite deep RL updates. This soft reward-penalty mechanism is distinct from MOReL's unknown-region absorbing state.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. The lower-bound analysis requires an admissible model-error estimator and stated value-class assumptions, not merely a learned variance heuristic. Exact surrogate optimization differs from finite deep RL updates. This soft reward-penalty mechanism is distinct from MOReL's unknown-region absorbing state.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• MOPO: Model-based Offline Policy Optimization

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction