AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

WorldGym: World Model as An Environment for Policy Evaluation

Julian Hector Quevedo, Ansh Kumar Sharma, Yixiang Sun, Varad Suryavanshi, Percy Liang, Sherry Yang

ICLR 2026 · Reinforcement learning › Visual and representation RL · approach: Vision-based robot RL

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
train world model on larger Bridge V2 data instead of Bridge V1 subset (ablation)WorldGym world model · Bridge validation videosLPIPS: 0.131 -> 0.073Improved
set diffusion prediction horizon equal to policy chunk size (h=4 vs h=1) (ablation)WorldGym inference · 40-frame rollouts on A100inference time (s): 93 -> 33Improved
evaluate policies in WorldGym instead of real worldRT-1-X, Octo, OpenVLA · OpenVLA Bridge eval suitemean success rate (%): 70.6 -> 67.4 (OpenVLA)No clear effect
add random OOD distractor objects to initial framesOpenVLA · OpenVLA Bridge eval suite (WorldGym)mean success rate (%): 67.4 -> 39.4Hurt
add random OOD distractor objects to initial framesOcto · OpenVLA Bridge eval suite (WorldGym)mean success rate (%): 23.82 -> 4.1Hurt
add random OOD distractor objects to initial framesRT-1-X · OpenVLA Bridge eval suite (WorldGym)mean success rate (%): 15.5 -> 7.6Hurt
recolor carrot red so it differs from orange targetOpenVLA · Bridge OOD: pick up orangegrasp correctnessImproved
add laptop displaying a carrot image as distractorOpenVLA · Bridge OOD: pick up carrotwrong-object grasp rate: 15% of trials grab the screen carrotHurt
use checkpoints with more training steps for UniPi and DexVLA policiesUniPi, DexVLA · Bridge V2 policies in WorldGymWorldGym success rateImproved

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