Scholarlink

<p>We present a language-driven, human-robot interface (HRI) for dynamic mission planning of autonomous surface vehicles (ASVs). The proposed Task2Mission (T2M) framework maps intuitive human language task commands into adaptive mission plans during autonomous remote operations. The framework incorporates a set of natural language task commands with their designed structures and leverages prompt engineering to generate a training dataset, which is used to fine-tune a compact sequence-to-sequence small language model (SLM), <Emphasis FontCategory="NonProportional">T5-small</Emphasis>. This model learns to translate natural language instructions into executable mission plans with high accuracy. We validate the framework through a series of simulation runs using water quality data collected from a physical ASV, demonstrating robust performance across diverse task scenarios. Our results show that T2M achieves reliable parsing and execution of task-to-mission mappings while supporting interactive and efficient mission visualization. This work highlights the potential of language-driven mission planning for ASVs, opening new avenues for adaptive, real-time monitoring of aquatic environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Task2Mission: language-driven task-to-mission mapping for adaptive water quality monitoring

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Sadman Islam,
  • Tauhidul Alam,
  • Sufi N. Quader

摘要

We present a language-driven, human-robot interface (HRI) for dynamic mission planning of autonomous surface vehicles (ASVs). The proposed Task2Mission (T2M) framework maps intuitive human language task commands into adaptive mission plans during autonomous remote operations. The framework incorporates a set of natural language task commands with their designed structures and leverages prompt engineering to generate a training dataset, which is used to fine-tune a compact sequence-to-sequence small language model (SLM), T5-small. This model learns to translate natural language instructions into executable mission plans with high accuracy. We validate the framework through a series of simulation runs using water quality data collected from a physical ASV, demonstrating robust performance across diverse task scenarios. Our results show that T2M achieves reliable parsing and execution of task-to-mission mappings while supporting interactive and efficient mission visualization. This work highlights the potential of language-driven mission planning for ASVs, opening new avenues for adaptive, real-time monitoring of aquatic environments.