跳转至

文章背景与核心概要

现代部署于精准农业等专业领域的机器人系统，通常需要操作员具备高级技术技能才能高效运行。为了弥合这一可用性鸿沟，研究人员开发了利用大语言模型（LLM）的任务规划器，能够将自然语言描述转化为可执行的农业任务规划。

然而，LLM 天生容易受到自然语言中歧义的影响。为解决这一问题，本文引入了一种扩展的规划架构，该架构包含由线性时序逻辑（LTL）驱动的多重反馈循环，从而确保生成的计划严格满足用户的规范。为了最大限度地减少偏好偏差，系统采用了两个不同的商业 LLM——一个专门负责规范制定，另一个负责验证。通过大量的实验，作者证明了这种方法的有效性，并强调了将自动化验证集成到自主机器人流水线中的能力与局限性。

As You Trust: Mission Planning with Formal Verification using LLMs in Precision Agriculture

Summary

现代 robotic 系统 deployed 在 specialized 领域如 precision agriculture 中，often require advanced technical skills to operate effectively. To bridge this usability gap, researchers have developed mission planners that utilize Large Language Models (LLMs) to translate natural language descriptions into actionable agricultural mission plans.

Modern robotic systems deployed in specialized fields like precision agriculture often require advanced technical skills to operate effectively. To bridge this usability gap, researchers have developed mission planners that utilize Large Language Models (LLMs) to translate natural language descriptions into actionable agricultural mission plans.

然而，LLMs 是 inherently 容易受到 natural language 的 ambiguities 的影响。To resolve this, 本文 introduces an extended planning architecture featuring multiple feedback loops powered by Linear Temporal Logic (LTL). This ensures that the generated plans strictly meet user specifications. 为了 minimize bias, the system employs two distinct commercial LLMs—one dedicated to specification and the other to verification. Through extensive experimentation, 作者 demonstrate the effectiveness of this approach, highlighting both the capabilities and limitations of integrating automated verification into autonomous robotics pipelines.

However, LLMs are inherently susceptible to the ambiguities of natural language. To resolve this, this paper introduces an extended planning architecture featuring multiple feedback loops powered by Linear Temporal Logic (LTL). This ensures that the generated plans strictly meet user specifications. To minimize bias, the system employs two distinct commercial LLMs—one dedicated to specification and the other to verification. Through extensive experimentation, the authors demonstrate the effectiveness of this approach, highlighting both the capabilities and limitations of integrating automated verification into autonomous robotics pipelines.

Article Metadata

  • arXiv Identifier: arXiv:2606.18519 [cs.RO]
  • Primary Subject: Robotics (cs.RO)
  • Secondary Subject: Artificial Intelligence (cs.AI)
  • Authors: Marcos Abel Zuzuárregui, Stefano Carpin
  • Journal Reference: Published in Proceedings of the 2026 International Conference on Robotics and Automation (ICRA)
  • Submission History:
  • [v1] Tue, 16 Jun 2026
  • [v2] Wed, 5 Aug 2026 (current version)

Abstract

尽管 robotic systems 现在正在各个 industries 中被 commercialized 和 deployed，但 many of these systems are highly specialized 并且通常 require an advanced skill set 来操作并确保它们按照 instructions 执行。To mitigate this problem, 我们 recently introduced a mission planner 借助 LLMs 来在 precision agriculture 中根据 provided in natural language 的 mission descriptions 综合 mission plans。While the system demonstrates impressive performance, it 也 suffers from the inherent ambiguities of natural language. 在本文中，我们 extend our system 来 address this issue，引入了 multiple feedback loops 在 planning architecture 中，这些 loops 借助 linear temporal logic (LTL) 来确保 mission planning system 能够 meet the specifications formulated by the user，同时仍然使用 natural language。为了 mitigate potential bias, 这是通过 using two different commercial LLMs 来完成的，它们分别负责 specification 和 verification subtasks。通过 extensive experiments, 我们 highlight the strengths and limitations of integrating mission verification into a fully autonomous pipeline，particularly regarding an LLM's ability to generate valuable LTL formulas，并展示了我们的 proposed implementation 如何 address and solve 这些 challenges。

Though robotic systems are now being commercialized and deployed in various industries, many of these systems are highly specialized and often require an advanced skill set to operate and ensure they perform as instructed. To mitigate this problem, we recently introduced a mission planner leveraging LLMs to synthesize mission plans in precision agriculture based on mission descriptions provided in natural language. While the system demonstrates impressive performance, it also suffers from the inherent ambiguities of natural language. In this paper, we extend our system to address this issue by introducing multiple feedback loops in the planning architecture that leverage linear temporal logic (LTL) to ensure the mission planning system meets the specifications formulated by the user while still using natural language. To mitigate potential bias, this is achieved by using two different commercial LLMs in charge of the specification and verification subtasks. Through extensive experiments, we highlight the strengths and limitations of integrating mission verification into a fully autonomous pipeline, particularly regarding an LLM's ability to generate valuable LTL formulas, and show how our proposed implementation addresses and solves these challenges.