arXiv cs.AI· Nurmukhambet Serikbayev, Abdul Basit, Muhammad Shafique·· 13 小时前AI 评分21
从关联到决策价值：VLA 语料选择中的配对指令依赖性检验
From Association to Decision Value: Testing Paired Instruction Dependence for VLA Corpus Selection
AI 导读
研究提出以“配对指令依赖性”作为机器人演示语料的上游诊断指标，在 136 个独立随机种子的四足导航语料上，用 56 个语料建模、80 个留出语料前瞻验证。
正文 · 原文
Abstract:Choosing among robot demonstration corpora is a ranking problem: an upstream diagnostic must preserve which candidates produce better downstream policies, not merely correlate with average performance. We study paired instruction dependence, the rate at which a demonstrator changes its first decision when only the goal changes. Across 136 independently seeded quadruped-navigation corpora, predictions developed on 56 corpora are evaluated prospectively on 80 held-out corpora. The audit is reliable and associated with downstream paired correctness, and lowers mean absolute prediction error when added to a correctness-aware baseline, but the primary proper-log-score increment is +0.059 nats (95% corpus-bootstrap CI [-0.027, 0.133]) and the prespecified top-20% comparison does not establish an incremental selection gain. An exploratory response decomposition suggests a structural source of this predictive-ranking mismatch: under the tested predictor specification, separating both-correct and both-wrong responses yields larger held-out gains than their unsigned switch statistic. Seed-to-seed agreement indicates that the held-out endpoint retains recoverable rank signal. These results show that counterfactual response structure can be reliable and predictive while providing limited incremental decision value in this setting when compressed into an unsigned statistic, motivating diagnostics that preserve outcome-relevant structure and are validated against the decisions they are intended to support.
| Comments: | 23 Pages, under review of ICLR2027 |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.29646 [cs.AI] |
| (or arXiv:2610.29646v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.29646 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Nurmukhambet Serikbayev [view email]
[v1]
Wed, 21 Oct 2026 12:27:55 UTC (6,433 KB)
来源：arXiv cs.AI · arxiv.org