Escaping the Diversity Trap in Robotic Manipulation via Anchor-Centric Adaptation

TL;DR

This paper introduces Anchor-Centric Adaptation (ACA), a novel framework for improving robotic manipulation by balancing demonstration diversity and estimation noise, leading to better task success rates within limited data budgets.

Contribution

The paper formalizes the Coverage-Density Trade-off in robot adaptation and proposes ACA, a two-stage method that stabilizes policies and selectively expands coverage for enhanced reliability.

Findings

ACA outperforms standard diverse sampling strategies in real-robot tasks.

Formalization of the Coverage-Density Trade-off guides optimal demonstration allocation.

ACA improves task success rates under limited data budgets.

Abstract

While Vision-Language-Action (VLA) models offer broad general capabilities, deploying them on specific hardware requires real-world adaptation to bridge the embodiment gap. Since robot demonstrations are costly, this adaptation must often occur under a strict data budget. In this work, we identify a critical diversity trap: the standard heuristic of "maximizing coverage" by collecting diverse, single-shot demonstrations can be self-defeating due to non-vanishing estimation noise. We formalize this phenomenon as a Coverage--Density Trade-off. By decomposing the policy error into estimation (density) and extrapolation (coverage) terms, we characterize an interior optimal allocation of unique conditions for a fixed budget. Guided by this analysis, we propose Anchor-Centric Adaptation (ACA), a two-stage framework that first stabilizes a policy skeleton through repeated demonstrations at core anchors, then selectively expands coverage to high-risk boundaries via teacher-forced error mining and constrained residual updates. Real-robot experiments validate our trade-off framework and demonstrate that ACA significantly improves task reliability and success rates over standard diverse sampling strategies under the same budget.