Doppel: one open video world model for eight robot embodiments

CMU, Johns Hopkins and Toyota Research open-sourced Doppel, a factored action-conditioned video world model that unifies eight robot embodiments by first 3D-rendering the robot body from the camera view, then conditioning a shared video model on the render plus actions.

2026-11-19 ~ 2026-11-19 · 2 related posts