The preprint introduces BFM-Zero, a promptable behavioral foundation model for humanoid robots. The authors argue that such a model could unify diverse control tasks under a single generalist policy, rather than training separate policies for each task. The approach is based on unsupervised reinforcement learning, which suggests it aims to learn reusable behaviors without dense, task-specific reward signals.

The abstract notes that existing approaches are "either exclusively deployed on…" but the text is truncated at that point in the available source. As a result, the specific limitations of prior work and the full details of BFM-Zero's evaluation are not accessible from this excerpt. The paper is available on arXiv under identifier 2511.04131.