Just Action Models unify two robot-control research paths
A paper by Yonghao Yu, Lang Huang, Runyi Li, and Toshihiko Yamasaki proposes Just Action Model (JAM), a framework combining a pretrained visual backbone, an action expert, an action loss, and a foresight loss. The reported JAM policies outperform matched FastWAM and ImageWAM counterparts on LIBERO and LIBERO-Plus, while achieving similar or slightly higher success rates on RoboTwin. The paper also reports that cached foresight targets reduce estimated online training FLOPs per sample per update by 73% relative to FastWAM.
The work offers a concrete way to compare vision-language-action and world-action policies by separating backbone initialization, future-observation objectives, and inference machinery, according to the paper.
Computer Science > Computer Vision and Pattern Recognition
arXiv:2610.37219 (cs)
[Submitted on 27 Oct 2026]
Title:Vision-Language-Action or World-Action? Just Action Models
Authors:Yonghao Yu, Lang Huang, Runyi Li, Toshihiko Yamasaki
View a PDF of the paper titled Vision-Language-Action or World-Action? Just Action Models, by Yonghao Yu and 3 other authors
#Just Action Model#FastWAM#ImageWAM#LIBERO#LIBERO-Plus#RoboTwin#Yonghao Yu#Lang Huang#Runyi Li#Toshihiko Yamasaki