Humanoid robots often face a timing mismatch: high-level planners emit sparse, low-rate task-space trajectories, while whole-body controllers must run at high frequency. The authors of this paper propose an asynchronous upper-body task-space tracking framework that directly addresses this gap. A student policy is initialized via teacher-student distillation and conditioned on the full cached future trajectory plus an execution-time index, allowing it to act even when reference commands arrive irregularly.
To keep the policy from drifting over time, the training uses a sliding-window global reward that penalizes frame drift without requiring explicit frame estimation. For task-specific post-training, an MPC module fills in sparse references to produce floating-base and upper-body guidance, while action-level and forward-kinematics-level self-guidance constrain policy drift.
Experiments in simulation and on a Unitree G1 humanoid show improved tracking under low update rates, stronger performance than synchronous and decoupled baselines, and safer adaptation to out-of-distribution motions. Since this is a single source, there are no conflicting findings to compare; the results are reported as the authors' own claims.