arXiv cs.AI· Tu Nguyen, Matthieu Zimmer, Vu Anh Vu, Ziyi Wang, Jannik Hammel Nielsen, Xuebing Zhou, Haitham Bou Ammar·· 10 天前AI 评分22
VICS-G：为 VLA 策略解码可行未来，安全动作不等于可行动作
A Safe Action Is Not Enough: Feasible-Future Decoding for Vision-Language-Action Policies
AI 导读
针对冻结的 VLA 策略可能选中局部安全但断绝后续安全完成路径的问题，研究者提出可行未来质量这一候选相关指标，并开发免训练、告警触发的重排序器 VICS-G。在六个 Safety-CHORES 设置中，VICS-G 将平均累计安全代价降低 1.9%-57.5%，成功率与策略采样差距在 2.5 个百分点以内，平均回合长度差距 0.82 步，且无需重新训练策略或在线 rollout。
正文 · 原文
Abstract:A safe action is not necessarily a viable one. A frozen vision-language-action (VLA) policy can favor a locally admissible move that leaves no policy-supported route to safe task completion. We call this the feasibility-likelihood gap: likelihood ranks the next move, while feasibility depends on the futures it leaves open.
To bring those futures into the decision, we derive the exact next-block marginal of the history-conditioned policy-environment trajectory law restricted to safe task completion. The derivation reveals a candidate-dependent feasible-future mass: its support records whether safe completion remains possible under the frozen continuation process, while its magnitude measures how much weighted safe-completion mass remains. Since exact evaluation is impractical online, we develop a selective finite-candidate approximation and establish conditions for recovering the best retained viable candidate.
Our alarm-triggered, training-free reranker VICS-G lowers mean cumulative safety cost by 1.9%-57.5% across six Safety-CHORES settings while remaining within 2.5 percentage points of policy sampling in success and 0.82 steps in mean episode length. Our approach offers a promising and practical path toward safer task completion, grounded in an exact policy-relative target yet requiring neither policy retraining nor online rollouts.
| Subjects: | Artificial Intelligence (cs.AI); Robotics (cs.RO) |
| Cite as: | arXiv:2610.05166 [cs.AI] |
| (or arXiv:2610.05166v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.05166 arXiv-issued DOI via DataCite |
Submission history
From: Tu Nguyen [view email]
[v1]
Sun, 4 Oct 2026 12:24:17 UTC (480 KB)
[v2]
Tue, 6 Oct 2026 10:17:30 UTC (480 KB)
来源：arXiv cs.AI · arxiv.org