Saved in:
Bibliographic Details
Main Authors: Chen, Yanzhe, Ma, Kevin Yuchen, Lv, Qi, Lin, Yiqi, Bai, Zechen, Gao, Chen, Shou, Mike Zheng
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.07381
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914543130640384
author Chen, Yanzhe
Ma, Kevin Yuchen
Lv, Qi
Lin, Yiqi
Bai, Zechen
Gao, Chen
Shou, Mike Zheng
author_facet Chen, Yanzhe
Ma, Kevin Yuchen
Lv, Qi
Lin, Yiqi
Bai, Zechen
Gao, Chen
Shou, Mike Zheng
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2605_07381
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Escaping the Diversity Trap in Robotic Manipulation via Anchor-Centric Adaptation
Chen, Yanzhe
Ma, Kevin Yuchen
Lv, Qi
Lin, Yiqi
Bai, Zechen
Gao, Chen
Shou, Mike Zheng
Robotics
Artificial Intelligence
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.
title Escaping the Diversity Trap in Robotic Manipulation via Anchor-Centric Adaptation
topic Robotics
Artificial Intelligence
url https://arxiv.org/abs/2605.07381