SALA Uses Latent Actions to Scale Embodied AI
AI-generated summary Researchers from Shanghai Jiao Tong University, Tsinghua University, and Sharpa Robotics published an arXiv preprint on November 12, 2026, introducing SALA as an embodied foundation model framework. The SALA architecture maps human demonstration videos and robot teleoperation traces into a shared latent action space to scale vision-language-action policies.