We show that ReactiveBFM is highly robust to diverse and violent disturbances preventing task completion.
While current Behavior Foundation Models (BFMs) provide robust control priors for humanoids, they only execute pre-defined reference motions. As a result, they are vulnerable to environmental shifts and incapable of reactive whole-body coordination. Naively cascading them with generative motion planners fails to achieve true reactivity, as inevitable tracking discrepancies induce fatal cumulative exposure bias. To bridge this gap, we propose ReactiveBFM, a real-time closed-loop planning-control framework. At its core, we effectively mitigate exposure bias via a scheduled prefix sampling curriculum, forcing the generative planner to actively learn error-recovery behaviors from imperfect physical states rather than ground-truth trajectories. Systematically, to reconcile the severe latency mismatch between auto-regressive planning and high-frequency tracking, we introduce an asynchronous replanning mechanism. Combined with trajectory chunking to temporally ensemble spatial references, our system guarantees spatio-temporally fluid execution without physical jitter. Deployed on the Unitree G1 humanoid, ReactiveBFM demonstrates unprecedented physical agility across a vast repertoire of text-conditioned closed-loop motions. Notably, ReactiveBFM achieves zero-shot moving target reaching, showcasing intricate whole-body coordination and on-the-fly replanning. In sim-to-sim benchmarking under severe perturbations, ReactiveBFM achieves a 93.1% success rate, significantly outperforming cascaded open-loop baselines by 28.6%.
We sincerely thank Yuxi Wei, Ke Fan, Tao Huang, Junli Ren, and Weiji Xie for their constructive advice and insightful discussions. We also thank Kinetix AI for their support with the HTC VIVE Ultimate Trackers.
We would like to acknowledge and recommend the following related works and resources:
If you find our work useful, please consider citing ReactiveBFM and the related papers mentioned above:
@article{chen2026reactivebfm,
title={ReactiveBFM: Reactive Closed-Loop Motion Planning Towards Universal Humanoid Whole-Body Control},
author={Chen, Xiao and Zeng, Weishuai and Niu, Xiaojie and Wang, Zirui and Li, Jianan and Wang, Huayi and Xu, Furui and Chen, Jiahe and Zhong, Weixiang and Ding, Lihe and Li, Kailin and Pang, Jiangmiao and Wang, Tai and Xue, Tianfan and Wang, Jingbo},
journal={arXiv preprint arXiv:2606.30362},
year={2026}
}@inproceedings{tevet2025closd,
title={{CL}o{SD}: Closing the Loop between Simulation and Diffusion for Multi-task Character Control},
author={Tevet, Guy and Raab, Sigal and Cohan, Setareh and Reda, Daniele and Luo, Zhengyi and Peng, Xue Bin and Bermano, Amit Haim and van de Panne, Michiel},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=pZISppZSTv}
}
@article{li2025mimic2dm,
title={Learning to Control Physically-simulated 3D Characters via Generating and Mimicking 2D Motions},
author={Li, Jianan and Chen, Xiao and Huang, Tao and Wong, Tien-Tsin},
journal={arXiv preprint arXiv:2512.08500},
year={2025}
}
@article{wang2025physhsi,
title={PhysHSI: Towards a Real-World Generalizable and Natural Humanoid-Scene Interaction System},
author={Wang, Huayi and Zhang, Wentao and Yu, Runyi and Huang, Tao and Ren, Junli and Jia, Feiyu and Wang, Zirui and Niu, Xiaojie and Chen, Xiao and Chen, Jiahe and Chen, Qifeng and Wang, Jingbo and Pang, Jiangmiao},
journal={arXiv preprint arXiv:2510.11072},
year={2025}
}
@article{wei2025humanoidunion,
title={Unveiling the Impact of Data and Model Scaling on High-Level Control for Humanoid Robots},
author={Wei, Yuxi and Wang, Zirui and Yin, Kangning and Hu, Yue and Wang, Jingbo and Chen, Siheng},
journal={arXiv preprint arXiv:2511.09241},
year={2025}
}
@misc{fan2025zerozeroshotmotiongeneration,
title={Go to Zero: Towards Zero-shot Motion Generation with Million-scale Data},
author={Fan, Ke and Lu, Shunlin and Dai, Minyue and Yu, Runyi and Xiao, Lixing and Dou, Zhiyang and Dong, Junting and Ma, Lizhuang and Wang, Jingbo},
year={2025},
eprint={2507.07095},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2507.07095}
}