arXiv cs.AI· Tiehan Cui, Peipei Liu, Yanxu Mao, Gaosheng Wang, Mingzhe Xing, Datao You·· 12 小时前AI 评分31
NEXUS：联合学习安全约束与动作模型，实现安全稳健的具身规划
NEXUS: Learning Safety Constraints and Action Models for Safe and Robust Embodied Planning
AI 导读
NEXUS 是一个模块化神经符号框架，同时维护 PDDL 领域模型与 LTL 安全规范，用执行失败修复动作模式、将检测到的危险转化为可复用 LTL 约束。在 SafeAgentBench 评测集上，NEXUS 在安全任务完成与不安全任务拒绝之间取得最佳整体权衡，并在越狱攻击下保持低违规率。消融实验证实 PDDL 领域精化与 LTL 约束积累具有互补收益。
正文 · 原文
Abstract:Large language models (LLMs) enable embodied agents to plan from naturallanguage instructions, yet their actions may be physically infeasible or hazardous. Existing neuro-symbolic learners primarily refine execution models, while formal safety methods acquire or enforce safety constraints. This separation limits the reuse of both execution corrections and safety knowledge across tasks. We present NEXUS, a modular neuro-symbolic framework that jointly maintains a PDDL domain model and an LTL safety specification. Execution failures are used to repair action schemas, whereas detected hazards are converted into reusable LTL constraints through hierarchical semantic and translation sampling with equivalence consensus. Accumulated constraints are enforced through LTL progression during symbolic search; comparing constrained and unconstrained search outcomes distinguishes physical unreachability from safety-induced refusal. Experiments on SafeAgentBench evaluation set show that NEXUS achieves the best overall trade-off between safetask completion and unsafe-task refusal, while maintaining low violation rates under jailbreak attacks. Ablations confirm the complementary benefits of PDDLdomain refinement and LTL-constraint accumulation.
| Subjects: | Artificial Intelligence (cs.AI); Robotics (cs.RO) |
| Cite as: | arXiv:2605.09387 [cs.AI] |
| (or arXiv:2605.09387v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.09387 arXiv-issued DOI via DataCite |
Submission history
From: Tiehan Cui [view email]
[v1]
Sun, 10 May 2026 07:25:38 UTC (2,810 KB)
[v2]
Fri, 16 Oct 2026 05:59:50 UTC (3,852 KB)
来源：arXiv cs.AI · arxiv.org