Scaling Behavior Foundation Model for Humanoid Robots

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Weishuai Zeng, Kangning Yin, Xiaojie Niu, Shunlin Lu, Weixiang Zhong, Jiahe Chen, Feiyu Jia, Xiao Chen, Zirui Wang, Furui Xu, Ming Zhou, Kailin Li, Weinan Zhang, He Wang, Li Yi, Dahua Lin, Jiangmiao Pang, Jingbo Wang
• Venue: arXiv: cs.RO
• Published: 2026-07-16
• Collected: 2026-07-17
• URL: http://arxiv.org/abs/2607.15163v1
• PDF: https://arxiv.org/pdf/2607.15163v1
• DOI: UNKNOWN

Plain-Language Brief

Researchers have developed a new method for improving the performance of Behavior Foundation Models (BFMs) in controlling humanoid robots. The method involves coordinating three core components: motion tracking learning paradigm, strategic synergy between on-policy rollout quantity and reference motion diversity, and an expressive and scalable model architecture called Humanoid Transformer. This approach significantly reduces error rates in both simulation and real-world scenarios compared to existing humanoid controllers.

What the Document Says

• The paper discusses a new method for scaling Behavior Foundation Models (BFMs) for humanoid robot control.
• The method involves three core components: motion tracking learning paradigm, strategic synergy between on-policy rollout quantity and reference motion diversity, and the Humanoid Transformer model architecture.
• Through experiments in simulation and real-world deployment, the new approach yields significant improvements in control fidelity and task generalization, reducing Mean Per-Keypoint Position Error (MPKPE) by over 10% in local mode and 82% in global mode compared to existing humanoid controllers.

Key Concepts Mentioned

• Behavior Foundation Models (BFMs)
• Humanoid robots
• motion tracking learning paradigm
• on-policy rollout quantity
• reference motion diversity
• Humanoid Transformer
• Mean Per-Keypoint Position Error (MPKPE)

Claims Made by the Document

• The new method significantly improves the performance of BFMs for humanoid robot control.

Evidence Used

• Reduction in MPKPE by over 10% in local mode and 82% in global mode compared to existing humanoid controllers.

Methods / Approach

• Motion tracking learning paradigm
• Strategic synergy between on-policy rollout quantity and reference motion diversity
• Humanoid Transformer model architecture

Limitations Stated by the Authors

Unclear how the new method will perform in various other scenarios not tested.

Limitations Noted by the Agent

Agent interpretation — not from the source.

The paper does not provide information on the cost or feasibility of implementing this method in real-world applications.

Open Questions

• How will the new method perform in various other scenarios not tested?
• What is the cost and feasibility of implementing this method in real-world applications?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The paper focuses on demonstrating the performance gains of the new method in humanoid robot control, but does not provide detailed information about its implementation or potential limitations beyond the scope of the tested scenarios.

Raw Abstract / Excerpt

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.