AI Paper Atlas

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

GEM: Generative Supervision Helps Embodied Intelligence

Ruowen Zhao, Bangguo Li, Zuyan Liu, Yinan Liang, junliang ye, FANGFU LIU, Diankun Wu, Zhengyi Wang, Xumin Yu, Yongming Rao, Han Hu, Jun Zhu

ECCV 2026 · Vision-language and multimodal › Robot action and navigation · approach: Embodied multimodal LLM agents · 7 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
add depth generative supervision during pretraining (ablation)Qwen3-VL-8B SFT · VSI-Benchaccuracy: 68.6 -> 70.6Improved
add depth generative supervision during pretraining (ablation)Qwen3-VL-2B SFT · VSI-Benchaccuracy: 60.0 -> 62.8Improved
add depth generative supervision during pretraining (ablation)Qwen3-VL-8B SFT · MMSI-Benchaccuracy: 32.8 -> 35.3Improved
add depth generative supervision during pretraining (ablation)Qwen3-VL-2B SFT · RoboSpatialaccuracy: 44.6 -> 47.4Improved
add depth generative supervision during pretraining (ablation)Qwen3-VL-8B SFT · RefSpatialaccuracy: 45.8 -> 44.4Hurt
add depth generative supervision during pretraining (ablation)Qwen3-VL-2B SFT · Where2Placeaccuracy: 53.0 -> 53.0No clear effect
add GEM-4M SFT plus depth supervision over base modelQwen3-VL-8B · VSI-Benchaccuracy: 57.9 -> 70.6Improved
use depth instead of RGB reconstruction as generative target (ablation)GEM · VSI-Benchaccuracy: 60.0 -> 63.0Improved
use progressive three-stage training instead of direct end-to-end (ablation)GEM · VSI-Benchaccuracy: 57.6 -> 63.0Improved
add depth generative supervision to VLA (ablation)Qwen3-VL-2B VLA · LIBEROaverage success rate: 94.9 -> 96.1Improved
add depth generative supervision to VLA (ablation)Qwen3-VL-2B VLA · Simpler WidowXaverage success rate: 61.5% -> 67.0%Improved
keep depth head active during VLA finetuning (ablation)GEM-VLA · real-world UR5 taskssuccess 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