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

2026-05-08

AI Summary

This work addresses the challenge of policy adaptation from limited real-world robot demonstration data, where conventional single-pass diverse sampling is prone to estimation noise and falls into a “diversity trap,” degrading adaptation performance. The authors propose the Anchor-Centric Adaptation (ACA) framework, which formally characterizes the trade-off between coverage and data density—revealing the root cause of this trap. ACA employs a two-stage strategy: first, it repeatedly samples around core anchor points to construct a stable policy backbone; then, it selectively expands data collection toward high-risk boundary regions for efficient allocation. Integrated with teacher-forcing error mining, constrained residual updates, and staged anchor learning, ACA significantly improves task success rates and reliability under the same data budget, outperforming existing diverse sampling approaches.

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.

Problem

• robotic manipulation
• embodiment gap
• data-efficient adaptation
• diversity trap
• coverage-density trade-off

Innovation

• Anchor-Centric Adaptation
• Coverage-Density Trade-off
• Vision-Language-Action Models
• Embodiment Gap
• Robot Policy Adaptation