Search papers, labs, and topics across Lattice.
This paper introduces ReactiveBFM, a real-time closed-loop planning-control framework that enhances humanoid motion planning by mitigating exposure bias through a scheduled prefix sampling curriculum. By employing an asynchronous replanning mechanism and trajectory chunking, the system achieves fluid execution and intricate whole-body coordination, even under environmental perturbations. Deployed on the Unitree G1 humanoid, ReactiveBFM demonstrates a remarkable 93.1% success rate in sim-to-sim benchmarking, outperforming traditional cascaded open-loop methods by 28.6%.
ReactiveBFM enables humanoids to achieve zero-shot moving target reaching with unprecedented agility and real-time adaptability.
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%.
