跳到正文
arXiv cs.CL· Woongkyu Lee, Jungwook Choi·· 10 小时前AI 评分38

CANDID:面向本地 Computer-Use Agent 的候选驱动干预框架,提升执行可靠性

CANDID: Candidate-Driven Intervention for Disciplined Execution in Local Computer-Use Agents

AI 导读

CANDID 是一个执行控制框架,通过并行测试时扩展在易失败的稀疏节点对本地 computer-use agent 进行执行级干预,避免文本反思带来的认知负担。它引入循环恢复机制过滤重复行为,并采用需全体一致的保守 DONE 门控策略来终止任务。在 OSWorld 上,CANDID 的成功率显著优于基于反馈的替代方案,且执行开销极小。

正文 · 原文

View PDF HTML (experimental)

Abstract:We present CANDID (Candidate-Driven Intervention for Disciplined Execution), an execution-control framework designed to enhance the reliability of local computer-use agents (CUAs) in resource-constrained settings. While local CUAs feature robust baseline proficiency in single-step action prediction, they frequently fail at critical junctions due to rigid action loops or premature DONE decisions. Rather than imposing imprecise cognitive loads through textual reflection, CANDID repurposes parallel test-time scaling into a disciplined, execution-level intervention at sparse, failure-prone points. We introduce a loop-recovery mechanism to filter candidates away from repeated behaviors and a conservative, unanimous DONE gating policy that requires absolute consensus to terminate. Experiments on OSWorld demonstrate that CANDID's structural control significantly improves success rates, outperforming feedback-based alternatives with minimal execution overhead.
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.24977 [cs.CL]
  (or arXiv:2611.24977v1 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2611.24977

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Jungwook Choi [view email]
[v1] Tue, 17 Nov 2026 15:39:11 UTC (3,911 KB)

来源:arXiv cs.CL · arxiv.org