Search papers, labs, and topics across Lattice.
This study investigates the effects of task-vector subtraction on multitask vision-language-action (VLA) policies in closed-loop robot control, revealing that the behavioral locality of these interventions is fragile. The authors identify three distinct regimes of performance degradation across ten LIBERO-Goal skills, highlighting that while some skills can be suppressed, others exhibit resistance or complete failure. Notably, the findings indicate that even minor edits can significantly impair unrelated tasks, emphasizing the complexity of maintaining control in edited models.
Task-vector subtraction in VLA policies can lead to unexpected control failures, with edits harming unrelated tasks and revealing the fragility of behavioral locality.
Task-vector arithmetic offers a closed-form way to modify a model, yet its behavioral locality remains unclear in closed-loop robot control. We present a target-and-control audit of per-skill task-vector subtraction from multitask vision-language-action (VLA) policies. Across all ten LIBERO-Goal skills, subtraction produces three qualitatively different regimes: target-control separation for five skills, resistance for three, and global collapse for two. On held-out initial states, the five suppressible targets remain at 0% success; however, mean baseline-normalized control retention is only 52%, and each target-suppressing edit materially harms at least one nominally unrelated control. Additional Goal panels show separation across tested policies with continuous-regression, discrete-token, and flow-matching action heads, whereas we observe no clean separation on Spatial and control collapse on the tested Object and Long-horizon panels. Mean task-vector cosine does not account for this variation. A matched-norm control identifies a local sign asymmetry around one Goal anchor, while multi-vector outcomes vary with anchor and scale. Retain-aware gradient baselines provide data-dependent comparators but require removal-time data and optimization; subtraction is data- and gradient-free only at edit time, assuming precomputed expert deltas. Finally, a single-skill relearning probe is consistent with behavioral masking, not certified unlearning. These results characterize task-vector subtraction as a fast but brittle intervention and underscore the need for closed-loop target-and-control evaluation when assessing locality in embodied model editing.
