Researchers have released GreenWAM, described as a world foundation model for robotics that serves two roles at once: an action-conditioned world model that predicts how a robot's actions change a scene, and a policy that selects those actions. The work addresses a common split in the field, where models are typically built and evaluated for only one of these roles, and where each robot records its actions in its own format.
The recipe behind GreenWAM has three parts. First, a shared anatomical action space that describes 20 robot families — from bimanual arms to dexterous humanoids — in common joint coordinates. Second, joint training across those families in visual, action-conditioned and world-action modes. Third, policies that split an action expert off the generator.
As a world model, GreenWAM reports top-tier trajectory, depth and 3D accuracy on WorldArena 1.0, ahead of every world-action model and video generator in the comparison. As a policy, it places among the top entries of the RoboTwin 2.0 leaderboard, achieves the highest zero-shot LIBERO-plus success among the compared methods, and is competitive with dedicated vision-language-action and world-action policies on CALVIN and LIBERO.
The authors further adapted GreenWAM to a real humanoid, where it succeeded more often than policies adapted from π0.5 and Cosmos 3 Edge, and distilled it into a single-step policy for real-time control.
Controlled experiments produced three findings: robot specialization improves world prediction and control together; an action expert helps when split off at the start of adaptation; and keeping world-model modes during policy adaptation preserves robustness to scene changes. Weights are published on Hugging Face under SberRoboticsCenter/GreenWAM.