FailureMimic: Humanoid Motion Tracking under Actuator Failure via Part-Wise Latent Residuals

Ziteng Ji, Chuye Hong*, Yiyang Shao*, Koushil Sreenath
University of California, Berkeley
*Equal contribution (Alphabetical Order)
Paper arXiv Video Code (coming soon)

Abstract

Humanoid motion-tracking policies are typically developed under nominal actuation, yet actuators can weaken or fail over a robot's service life. A local actuator impairment can affect whole-body tracking and require coordinated compensation from intact body parts. We therefore ask whether organizing the control representation around body parts provides a useful inductive bias for fault-tolerant tracking. We present FailureMimic, a framework that distills a whole-body tracking policy into a part-wise latent motion model and adapts the frozen model with a residual policy that coordinates across body parts. The residual policy is not given the identity of the impaired joints. In simulation on a Unitree G1, FailureMimic achieves higher fall-free survival than monolithic and non-part-aware baselines under passive leg impairments and systematic pushes; its margin grows with impairment severity and persists at severities not encountered during training. The same structure also serves the opposite demand: because body parts stay separable, it tracks novel combinations of upper- and lower-body motions. On the physical G1, FailureMimic autonomously performs loco-manipulation with passive or locked joints and tracks stitched motions.

Overview of the three stages: a whole-body expert tracks reference motions and emits joint-position targets; the expert is distilled into a part-wise latent prior aligned with a training-time posterior; the prior and decoder are then frozen while a body-part Transformer maps the state and the sparse five-point command to a latent residual.

Overview of the three stages. Stage 1: a whole-body expert tracks reference motions g, emitting joint-position targets. Stage 2: the expert is distilled into a part-wise latent prior p(z | s), aligned by a divergence term with a training-time posterior. Stage 3: the prior and decoder are frozen, and a body-part Transformer maps the state and the sparse five-point command to a residual. Colors mark the five trunk–limb groups.

Part-wise architecture: the sparse five-point command, proprioceptive state, and attitude signals are tokenized by body part into five tokens; a Transformer encoder mixes them by multi-head self-attention, and zero-initialized per-part heads emit a latent residual that shifts the frozen prior mean before each slice is decoded by its own branch.

Part-wise architecture. The command, proprioceptive state, and attitude signals are tokenized by body part into five tokens; a Transformer encoder mixes them by self-attention, and zero-initialized per-part heads emit the latent residual. The residual shifts the frozen prior mean, and each latent slice is decoded by an independent branch driving only its own group's joints. Snowflakes mark frozen modules; the bolt marks the trained residual generator.

Loco-manipulation

Box carrying

Box carrying

Box carrying with right-wrist damping

Box carrying with right hip-roll damping

Box carrying with left-elbow locking

Box pushing

Box pushing

Box pushing with right ankle-roll damping

Stitched motion tracking

Basic motion tracking

Dance motion tracking

Damping Tracking

damping left ankle roll joint

damping right hip yaw joint

damping right ankle roll joint

damping right ankle pitch joint

Lock Tracking

lock left knee

lock left hip roll joint

BibTeX

@article{ji2027failuremimic,
  author    = {Ji, Ziteng and Hong, Chuye and Shao, Yiyang and Sreenath, Koushil},
  title     = {{FailureMimic}: Humanoid Motion Tracking under Actuator Failure via Part-Wise Latent Residuals},
  journal   = {In-submission},
  year      = {2027},
}