AI Paper Atlas

Research map / Reinforcement learning / Offline RL

EXPO: Stable Reinforcement Learning with Expressive Policies

Perry Dong, Qiyang Li, Dorsa Sadigh, Chelsea Finn

ICLR 2026 · Reinforcement learning › Offline RL · approach: Offline RL sample complexity theory · 42 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
use value-maximizing actions in the TD backup (not only for sampling) (ablation)EXPO · Robomimic Can and Squaresample efficiency and successImproved
add Gaussian action edits on top of base policy samples (ablation)EXPO · pen-binary-v0normalized returnImproved
add Gaussian action edits on top of base policy samples (ablation)EXPO · Robomimic Squaresuccess rateImproved
use better offline data (higher imitation-learning success) (ablation)EXPO · Robomimic Square (subsampled demos)success at 1M stepsImproved
add entropy backup with soft sampling for narrow datasets (ablation)EXPO · Square with weak offline datasuccess rateImproved
drop the offline dataset after pretraining, warm-start with pretrained policy rollouts (ablation)EXPO · Robomimic Lift and Cansample efficiencyNo clear effect
vary number of action samples N and edit scale beta (ablation)EXPO · Robomimic Lift and SquareperformanceMixed
fine-tune with EXPO versus offline-RL baselines with expressive policies (IDQL, DAC, QSM, DIPO)diffusion policy · 12 sparse-reward taskssample efficiency: 2-3xImproved

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