robbyant-research / Zero-WAM

In‑Context World‑Action Modeling from Human Videos for Open‑Ended Task Generalization