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

2026-06-29

AI Summary

Existing foundation models for humanoid robot behaviors struggle to adapt to environmental changes and lack whole-body coordination and real-time responsiveness. This work proposes a real-time closed-loop planning-and-control framework that mitigates exposure bias through a scheduled prefix-sampling mechanism and incorporates curriculum learning to train error-recovery behaviors. To address the frequency mismatch between planning and control, the approach introduces asynchronous replanning combined with a spatiotemporal trajectory chunking and fusion strategy, enabling highly agile, temporally coherent zero-shot grasping of moving targets. Evaluated on the Unitree G1 platform, the system supports diverse text-driven closed-loop actions and achieves a 93.1% success rate under simulated disturbances, representing a 28.6% improvement over an open-loop cascaded baseline.

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%.

Problem

• Behavior Foundation Models
• reactive whole-body control
• exposure bias
• closed-loop motion planning
• humanoid locomotion

Innovation

• ReactiveBFM
• closed-loop planning
• exposure bias mitigation
• asynchronous replanning
• whole-body coordination