多智能体视觉语言导航的系统化形式化、基准与方法
原标题：Paper page - Systematic Multi-Agent Vision-and-Language Navigation: Formulation, Benchmark, and Method
AI 摘要
该论文首次将多智能体视觉语言导航（VLN）系统形式化为受约束的协调问题，其中每个任务由带依赖和资源约束（presence locks 与 holding chains）的子任务组成。作者构建了 MAVLN 基准，包含 145 个场景中的 11,724 个 episode，支持最多四个智能体在三种指令模式下协作，并配有约束感知指标。同时提出 TRISS 系统，结合基于 LLM 的子任务调度器、共享拓扑记忆和冲突感知执行机制。实验表明 TRISS 可作为综合基线，但在调度、规划和执行方面仍有较大提升空间。
正文节选
Systematic Multi-Agent Vision-and-Language Navigation: Formulation, Benchmark, and Method Abstract Vision-and-Language Navigation (VLN) has largely focused on a single agent following a single instruction, yet many real-world applications require teams of robots to tackle tasks beyond the capabilities of any individual agent. We present Systematic Multi-Agent Vision-and-Language Navigation, providing, to our knowledge, the first systematic formalization of multi-agent VLN as a constrained coordi