Scaling Behavior Foundation Model for Humanoid Robots

2026-07-16

AI Summary

Humanoid robot control faces significant challenges in whole-body coordination, real-time responsiveness, and cross-scenario generalization, with existing approaches limited in scalability and universality. This work proposes a scalable behavioral foundation model that achieves breakthrough performance through three core innovations: a global-frame-based motion tracking learning paradigm, a co-designed strategy involving the number of policy rollouts and diversity of reference motions, and a novel Humanoid Transformer architecture. The study systematically uncovers a scalable pathway for behavioral foundation models, enabling structured behavioral representations to emerge naturally. Evaluated in both simulation and real-world deployment, the approach substantially improves performance, reducing MPKPE by over 10% on local motion patterns and by 82% on global motion patterns in the test set.

Abstract

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. These results establish BFM as a principled and effective foundation for scalable and general-purpose humanoid control.

Problem

• Behavior Foundation Models
• humanoid control
• scaling
• generalization
• whole-body coordination

Innovation

• Behavior Foundation Model
• Humanoid Transformer
• motion tracking
• scalable architecture
• whole-body coordination