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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| optimize policy in world model (WMPO) instead of GRPO, P=128 | OpenVLA-OFT · Mimicgen mean | success rate (%): 33.2 -> 47.1 | Improved |
| WMPO instead of DPO, P=128 | OpenVLA-OFT · Mimicgen mean | success rate (%): 37.3 -> 47.1 | Improved |
| WMPO instead of DPO, P=1280 | OpenVLA-OFT · Mimicgen mean | success rate (%): 42.4 -> 57.6 | Improved |
| increase rollout budget 128 to 1280 for WMPO | WMPO · Mimicgen mean | success rate (%): 47.1 -> 57.6 | Improved |
| increase rollout budget 128 to 1280 for DPO | DPO · Mimicgen mean | success rate (%): 37.3 -> 42.4 | Improved |
| increase rollout budget 128 to 1280 for GRPO | GRPO · Mimicgen mean | success rate (%): 33.2 -> 37.1 | Improved |
| GRPO online RL with P=128 on Coffee | GRPO · Mimicgen Coffee | success rate (%): 43.8 -> 38.3 | Hurt |
| apply WMPO to base policy under disruptions | OpenVLA-OFT · Disruption mean | success rate (%): 23.7 -> 29.6 | Improved |
| apply DPO to base policy under disruptions | OpenVLA-OFT · Disruption mean | success rate (%): 23.7 -> 19.5 | Hurt |
| apply GRPO to base policy under disruptions | OpenVLA-OFT · Disruption mean | success rate (%): 23.7 -> 24.7 | No clear effect |
| apply WMPO on real robot | OpenVLA-OFT · Real-world insert square into stick | success rate (%): 53% -> 70% | Improved |
| apply DPO on real robot | OpenVLA-OFT · Real-world insert square into stick | success rate (%): 53% -> 60% | Improved |
Closest papers by meaning
- SRPO: Self-Referential Policy Optimization for Vision-Language-Action Models (CVPR 2026 · similarity 0.94)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- Human-assisted Robotic Policy Refinement via Action Preference Optimization (NeurIPS 2025 · similarity 0.90)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
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.
