ReactiveBFM: Reactive Closed-Loop Motion Planning Towards
Universal Humanoid Whole-Body Control

1The Chinese University of Hong Kong 2Shanghai AI Laboratory
*Core Contributors (Random Order) Corresponding Authors
All motions are generated fully online, with no reference motions.

Online Replanning to Reach Targets

Real-World Deployment

Sim-to-Sim Transfer

  • Real-time whole-body replanning — references updated on the fly.
  • Robust real-world execution — reliably reaches a moving target.
  • Zero-shot generalization — trained on static targets, yet generalizes to the dynamic setting.

Stability and Robustness

Continuously recover from violent disturbances.

Highlight

😈 Surviving a Gang-Up 1

😈 Surviving a Gang-Up 2

Recover from diverse disturbances.

Torso Perturbation Recovery

Command-Faithful Recovery

Taichi Motion Recovery

Whole-Body Perturbation Recovery

We show that ReactiveBFM is highly robust to diverse and violent disturbances preventing task completion.

Streaming Interactive Control

Multi-Round Streaming Text + Real-Time Whole-Body Replanning

Stylized Walking

Kungfu

Behaviour Gallery

Unified Framework, Universal Behaviours

Kungfu Sword Training

Butterfly Kick

Continuous Anti-clockwise Spin

Tai Chi

Kungfu Front Kick

Superman-Like Walking

Abstract

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%.

Citation

If you find this project useful, please cite:

@article{yourproject2026humanoid,
  title={Project Title},
  author={Author One and Author Two and Author Three and Author Four},
  journal={arXiv preprint arXiv:XXXX.XXXXX},
  year={2026}
}