AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

WMPO: World Model-based Policy Optimization for Vision-Language-Action Models

Fangqi Zhu, YAN Zhengyang, Zicong Hong, Quanxin Shou, Xiao Ma, Song Guo

ICLR 2026 · Vision-language and multimodal › Robot action and navigation · approach: Vision-language-action foundation models · 69 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
optimize policy in world model (WMPO) instead of GRPO, P=128OpenVLA-OFT · Mimicgen meansuccess rate (%): 33.2 -> 47.1Improved
WMPO instead of DPO, P=128OpenVLA-OFT · Mimicgen meansuccess rate (%): 37.3 -> 47.1Improved
WMPO instead of DPO, P=1280OpenVLA-OFT · Mimicgen meansuccess rate (%): 42.4 -> 57.6Improved
increase rollout budget 128 to 1280 for WMPOWMPO · Mimicgen meansuccess rate (%): 47.1 -> 57.6Improved
increase rollout budget 128 to 1280 for DPODPO · Mimicgen meansuccess rate (%): 37.3 -> 42.4Improved
increase rollout budget 128 to 1280 for GRPOGRPO · Mimicgen meansuccess rate (%): 33.2 -> 37.1Improved
GRPO online RL with P=128 on CoffeeGRPO · Mimicgen Coffeesuccess rate (%): 43.8 -> 38.3Hurt
apply WMPO to base policy under disruptionsOpenVLA-OFT · Disruption meansuccess rate (%): 23.7 -> 29.6Improved
apply DPO to base policy under disruptionsOpenVLA-OFT · Disruption meansuccess rate (%): 23.7 -> 19.5Hurt
apply GRPO to base policy under disruptionsOpenVLA-OFT · Disruption meansuccess rate (%): 23.7 -> 24.7No clear effect
apply WMPO on real robotOpenVLA-OFT · Real-world insert square into sticksuccess rate (%): 53% -> 70%Improved
apply DPO on real robotOpenVLA-OFT · Real-world insert square into sticksuccess rate (%): 53% -> 60%Improved

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