Chunk-Boundary Artifact in Action-Chunked Generative Policies: A Noise-Sensitive Failure Mechanism

2026-03-12

AI Summary

This work addresses the long-standing but poorly understood issue of task failure in pretrained action chunking policies due to discontinuities at chunk boundaries—commonly referred to as “boundary artifacts.” Rather than treating these artifacts as mere byproducts of execution, the study reframes the problem as a latent-noise-driven, analyzable, and intervenable failure mechanism. Through permutation tests, latent noise perturbations, identification of noise-space directions, and trajectory-level steering techniques, the authors systematically establish the causal role of latent noise in generating boundary artifacts. Experiments demonstrate that targeted noise interventions can substantially shift success–failure distributions in challenging tasks. Even in near-optimal settings, such interventions elicit measurable negative causal effects, confirming the validity and generality of the proposed mechanism.

Technology Category

• Knowledge Representation and Reasoning: Action, Change, and Causality
• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Action chunking has become a central design choice for generative visuomotor policies, yet the execution discontinuities that arise at chunk boundaries remain poorly understood. In a frozen pretrained action-chunked policy, we identify chunk-boundary artifact as a noise-sensitive failure mechanism. First, artifact is strongly associated with task failure (p < 1e-4, permutation test) and emerges during the rollout rather than only as a post-hoc symptom. Second, under a fixed observation context, changing only latent noise systematically modulates artifact magnitude. Third, by identifying artifact-related directions in noise space and applying trajectory-level steering, we reliably alter artifact magnitude across all evaluated tasks. In hard-task settings with remaining outcome headroom, the success/failure distribution shifts accordingly; on near-ceiling tasks, positive gains are compressed by policy saturation, while the negative causal effect remains visible. Overall, we recast boundary discontinuity from an unavoidable execution nuisance into an analyzable, noise-dominated, and intervenable failure mechanism.

Problem

• chunk-boundary artifact
• action chunking
• generative visuomotor policies
• execution discontinuities
• noise sensitivity

Innovation

• action chunking
• chunk-boundary artifact
• noise sensitivity
• latent noise steering
• generative visuomotor policies