论文HF Daily Papers · arXiv 2611.262393 天前
SAFI框架解决VLA模型执行中安全停止难题,接管后任务完成率降至1.3%
本文由 AI 根据公开内容整理,可能有误,请以原文为准:Teaching Vision-Language-Action Models to Stop Safely(HF Daily Papers · arXiv 2611.26239)
核心内容
针对视觉-语言-动作(VLA)模型在执行过程中难以安全停止禁止行为及控制物理响应的挑战,研究团队提出状态感知回退模仿(SAFI)方法。评估显示,仅抑制初始状态的任务策略在中途接管时仍可能完成任务或导致手持物体跌落。SAFI通过构建回退目标与联合策略适配两阶段训练,使OpenVLA-OFT模型在LIBERO和BadLIBERO测试集中,将接管后的聚合目标成功率降至1.3%,且未记录到任何物理事故或动作饱和现象。
SAFI技术方案包含回退目标构建与联合策略适配两个核心阶段。前者在禁止任务演示中分配静止机械臂指令与一致的夹爪目标,后者将这些目标与保留任务的原始动作联合学习,使模型在中间执行状态获得显式替代响应。实验验证了该方法在确保安全停机的同时,维持了机器人对非禁止任务的高成功率,解决了传统拒绝机制在动态物理交互中的安全性盲区,为具身智能体的运行时安全控制提供了可量化的工程范式。
关键看点
- 安全痛点:VLA模型中途接管禁止任务时仍可能完成操作或致物体跌落,单纯任务未完成并非可靠安全指标
- SAFI机制:通过回退目标构建分配静止臂指令与一致夹爪目标,结合联合策略适配学习中间态显式替代响应
- 量化成效:OpenVLA-OFT在LIBERO任务中接管后聚合成功率降至1.3%,实现零物理事故与零动作饱和
- 技能保持:在引入安全停止机制的同时,模型对保留任务的成功率维持高水平,未出现灾难性遗忘
工程与产业观察
该研究揭示了VLA模型运行时安全控制的工程缺口,即静态拒绝无法覆盖动态物理交互风险。SAFI通过将安全停止建模为可学习的回退策略而非简单截断,为具身智能体提供了兼顾安全性与功能完整性的训练范式,对工业及服务机器人的合规部署具有直接参考价值。
英文摘要 (Abstract)
Vision-language-action (VLA) models enable robots to perform diverse tasks from visual observations and language instructions. While prior work has explored task suppression and refusal, stopping prohibited behavior once execution is underway and controlling the resulting physical response remain challenging. Our evaluation reveals that policies which suppress a task from its initial state can still complete it when taking over partway through execution. Even when the task remains unfinished, the robot may drop a held object, making non-completion an incomplete measure of safety. To address these challenges, we introduce State-Aware Fallback Imitation (SAFI) for tasks where holding provides an appropriate alternative response. SAFI comprises two stages, (1) fallback target construction, which assigns stationary arm commands and consistent gripper targets throughout prohibited-task demonstrations, and (2) joint policy adaptation, which learns these targets alongside the original actions of retained tasks. This supervision teaches an explicit alternative response at intermediate execution states while maintaining the robot's other skills. Extensive experiments with OpenVLA-OFT across LIBERO and BadLIBERO tasks evaluate prohibited completion, physical incidents, and retained-task performance. On the evaluated LIBERO tasks, SAFI reduces aggregate target success after takeover to 1.3%, with no recorded physical incidents or action saturation, while preserving high retained-task success.
Hugging Face Daily Papers 热度:3 upvotes 作者:Zheyuan Liu, Chuhan Wang, Jinduo Guo, Yue Huang, Xiangchi Yuan, Ningshan Ma, Meng Jiang
外部来源
本页事实对应的公开页面,以原文为准。(点击将离开本站)