AI Paper Atlas

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

Robust Reinforcement Learning for Continuous Control with Model Misspecification

Daniel J. Mankowitz, Nir Levine, Rae Jeong, Abbas Abdolmaleki, Jost Tobias Springenberg, Yuanyuan Shi, Jackie Kay, Todd Hester, Timothy Mann, Martin Riedmiller

ICLR 2020 · Reinforcement learning › Policy gradient and actor-critic · approach: Model-based RL with guarantees · 41 citations (OpenAlex)

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
Add robust worst-case transition evaluationE-MPO · Nine Mujoco domains with unseen perturbationsaverage rewardImproved
Add soft-robust average-transition evaluationE-MPO · Mujoco domains with unseen perturbationsaverage rewardMixed
Add robust worst-case transition evaluationE-MPO · Shadow handaverage rewardImproved
Use robust or soft-robust transition evaluationE-SVG · Cartpole and Pendulumaverage rewardImproved
Train the non-robust agent for three times more episodes (ablation)Non-robust agent · Cartpole Balance and Pendulum Swingupaverage rewardNo clear effect
Train with domain randomization over the limited uncertainty-set perturbations (ablation)Non-robust agent · Cartpole Balance and Pendulum Swingupaverage rewardHurt
Increase domain randomization to a larger perturbation set (ablation)Non-robust agent · Pendulum Swingupaverage rewardMixed
Modify the third uncertainty-set perturbation so it approaches the unseen evaluation dynamics (ablation)R-MPO · Pendulum Swingupaverage rewardImproved
Add entropy regularization to robust MPO (ablation)R-MPO · Mujoco domainsaverage rewardMixed
Use offline-data-trained transition models as the uncertainty set (ablation)R-MPO · Cartpole Swingup and Pendulum Swingupaverage rewardMixed
Increase offline data used to train uncertainty-set transition models (ablation)DDR-MPO · Cartpole Swingup and Pendulum Swingupaverage rewardMixed
Change the nominal model from the smallest to the median or largest perturbation (ablation)Robust and non-robust agents · Mujoco domains with held-out perturbationsaverage rewardMixed

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