Building Behavioral Foundation Models (BFMs) for humanoid robots has the potential to unify diverse control tasks under a single, promptable generalist policy. However, existing approaches are either exclusively deployed on simulated humanoid characters, or specialized to specific tasks such as tracking. We propose BFM-Zero, a framework that learns an effective shared latent representation that embeds motions, goals, and rewards into a common space, enabling a single policy to be prompted for multiple downstream tasks without retraining. This well-structured latent space in BFM-Zero enables versatile and robust whole-body skills on a Unitree G1 humanoid in the real world, via diverse inference methods, including zero-shot motion tracking, goal reaching, and reward optimization, and few-shot optimization-based adaptation. Unlike prior on-policy reinforcement learning (RL) frameworks, BFM-Zero builds upon recent advancements in unsupervised RL and Forward-Backward (FB) models, which offer an objective-centric, explainable, and smooth latent representation of whole-body motions. We further extend BFM-Zero with critical reward shaping, domain randomization, and history-dependent asymmetric learning to bridge the sim-to-real gap. Those key design choices are quantitatively ablated in simulation. A first-of-its-kind model, BFM-Zero establishes a step toward scalable, promptable behavioral foundation models for whole-body humanoid control.

An overview of the BFM-Zero: After the pre-training stage, BFM-Zero forms a latent space that can be used for zero-shot inference and few-shot adaptation.

How it works: Unsupervised RL and Zero-shot inference

Click image for real-world goal pose, click ▶ for natural transition, and drag the progress bar to view frame by frame.

We also test diverse initial on the ground pose to a standing pose. Follow the gallery to see the full description of the performance.

We do not have any labeled(task-related) rewards in training. The reward inference (optimization) happens in the test time. Users can prompt in any type of reward function w.r.t. the robot states, and the policy zero-shot output the optimized skills without retraining. ($R$ below is the rewards.)

Note: Right/Left is relative to the robot; the reward functions are for illustration purposes, in the inference time, we also have some soft constraints and regularization terms.

We demonstrate the robustness and flexibility of BFM-Zero: The policy enables the humanoid robot to recover gracefully from various disturbances, including heavy pushes, torso kicks, ground pulls, and leg kicks.

We demonstrate BFM-Zero's few-shot adaptation capability, the smooth structure of our latent space enables efficient search-based optimization in simulation to discover a better latent instead of directly using zero-shot inference within a short time.

When the robot carries a 4kg payload on its torso, we can perform adaptation to let the robot keep single-leg standing for longer time.

The structured nature of the learned space enables smooth interpolation between latent representations. We can leverage Spherical Linear Interpolation to generate intermediate latent vectors along the geodesic arc between the two end-points.

We can see simple interpolation shows meaningful semantic-level changes.