AI Paper Atlas

Research map / Reinforcement learning / Offline RL

NeoRL: A Near Real-World Benchmark for Offline Reinforcement Learning

Rong-Jun Qin, Xingyuan Zhang, Songyi Gao, Xiong-Hui Chen, Zewen Li, Weinan Zhang, Yang Yu

NeurIPS 2022 · Reinforcement learning › Offline RL · approach: Simple practical offline RL · 24 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
compare against deterministic policy instead of behavior policyCQL · NeoRL 51 taskswin ratio: 94.12% -> 88.24%Hurt
compare against deterministic policy instead of behavior policyPLAS · NeoRL 51 taskswin ratio: 72.55% -> 60.78%Hurt
use CQL instead of BCBC · NeoRL 51 tasksaverage rank: 4.63 -> 2.45Improved
use PLAS or BCQ instead of BCBC · NeoRL 51 tasksaverage rank: 4.63 -> 4.61 (PLAS), 5.63 (BCQ)No clear effect
use MOPO instead of BCBC · NeoRL 51 tasksaverage rank: 4.63 -> 7.08Hurt
add KL constraint to MB-PPO (ablation)MB-PPO · NeoRL 51 tasksaverage rank: 6.29 -> 3.90Improved
use model-based methods instead of model-freeoffline RL · NeoRLepisode returnMixed
increase dataset size or qualityoffline RL algorithms · NeoRLepisode returnMixed
select policies with FQE instead of uniform random selectionFQE · Walker2d-v3 policy selectionTop-1 mean scoreHurt
select policies by FQE rankFQE · Walker2d-v3RC score: -.005 +- .222No 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