Scaling Behavior Foundation Model for Humanoid Robots

Summary

Humanoid control requires natural whole-body coordination, precise real-time responses to control signals, and robust generalization across diverse environmental contexts, making it a cornerstone for generalist embodied agents.

Abstract Summary

Key Points

• Humanoid control requires natural whole-body coordination, precise real-time responses to control signals, and robust generalization across diverse environmental contexts, making it a cornerstone for generalist embodied agents.
• Behavior Foundation Models (BFMs) have recently emerged as a promising solution to address these challenges by leveraging large-scale behavioral data to achieve superior expressiveness, versatility and generalization.
• However, despite growing interest in scaling BFMs to further improve their capabilities, it remains unclear how key factors, including the learning paradigm, behavioral data and model architecture should be coordinated to enable effective scaling.
• In this work, we revisit the scaling recipe for BFMs and demonstrate that substantial performance gains can be achieved through the coordination of three core components: 1) the learning paradigm of motion tracking that reformulates diverse humanoid control problems as the reproduction of integrated whole-body behaviors in the global frame; 2) the strategic synergy between on-policy rollout quantity and reference motion diversity; and 3) the expressive and scalable model architecture termed Humanoid Transformer that facilitates the natural emergence of structured behavioral representations.
• Through extensive experiments in both simulation and real-world deployment, we demonstrate that our approach yields significant improvements in control fidelity and task generalization, reducing Mean Per-Keypoint Position Error (MPKPE) on the test set by over 10% in local mode and 82% in global mode compared with existing humanoid controllers.

Method

In this section, we first present our revised formulation and learning objective for BFM pretraining and establish motion tracking as its principled and scalable instantiation. We then detail our motion-tracking recipe, including the state representation, control interface, reward design, and domain randomization strategy.

Next, we formalize the notion of training data within our framework and elucidate the corresponding dataset construction pipeline to achieve a synergy between data quantity and diversity.

Experiments

In this section, we first investigate the scaling behavior of BFMs from the perspectives of training data and model architecture. Building on these findings, we then analyze the learned latent space to show that structured representations of behavioral intentions emerge naturally without auxiliary objectives and are further refined through scaling.

Finally, we benchmark the resulting BFMs against existing humanoid controllers to evaluate their effectiveness as humanoid foundation controllers. Environments.

Sources and Demonstrations

Figures are reproduced from Zeng et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.