GreenWAM Launches Global Foundation Model for Robotics
robotics
| Source: HF Papers | Original article
GreenWAM is a universal robot world model that predicts scene changes from actions and selects actions, unifying fragmented robot data formats.
Sber Robotics Center has unveiled GreenWAM, a “world foundation model” that aims to combine two traditionally separate functions in robotics – predicting how a scene will evolve after an action and selecting the action itself. The five‑person team behind the project trained the model to operate across 20 distinct robot families, ranging from simple bimanual arms to dexterous humanoids, by mapping each embodiment onto a shared joint‑coordinate space. Weights for four benchmark setups are already available on Hugging Face, allowing researchers to experiment without rebuilding the model from scratch.
The release matters because most robot software pipelines still treat world modelling and policy generation as isolated components, each tied to proprietary kinematic formats. GreenWAM’s unified approach promises to cut down the engineering overhead of switching between simulators and controllers, and to enable knowledge transfer between disparate robot platforms. By speaking a common “movement language,” the model could accelerate the development of versatile robotic systems and lower the barrier for smaller labs to experiment with advanced control strategies.
The next steps will reveal how the community adopts the model. Watch for performance results on standard robotics benchmarks, integration with existing simulation environments, and any real‑world deployments that test the dual‑role claim outside the lab. Further releases of pretrained weights or extensions to additional robot families would also indicate whether GreenWAM can become a true foundation layer for the broader robotics ecosystem.
Sources
Back to AIPULSEN