Maritime Mission Planning for Unmanned Surface Vessel using Large Language Model

TL;DR

The paper addresses USV mission planning under dynamic sea conditions, where traditional static planners fail to adapt to obstacles and changing environments. It introduces an LLM-guided framework that converts high-level goals into executable symbolic plans and uses real-time low-level feedback to replan. The architecture comprises a Prompt Generator, a GPT-4-based Symbolic Plan Generator, and a Navigation/Control module that employs sensing, path planning, and PID control to achieve robust execution. Validation via the MBZIRC Maritime Simulator demonstrates improved plan adaptability, reduced operational risk, and effective task completion in docking-terminal inspection scenarios. Future work envisions Vision-Language Model integration and more sophisticated control to further enhance autonomy.

Abstract

Unmanned Surface Vessels (USVs) are essential for various maritime operations. USV mission planning approach offers autonomous solutions for monitoring, surveillance, and logistics. Existing approaches, which are based on static methods, struggle to adapt to dynamic environments, leading to suboptimal performance, higher costs, and increased risk of failure. This paper introduces a novel mission planning framework that uses Large Language Models (LLMs), such as GPT-4, to address these challenges. LLMs are proficient at understanding natural language commands, executing symbolic reasoning, and flexibly adjusting to changing situations. Our approach integrates LLMs into maritime mission planning to bridge the gap between high-level human instructions and executable plans, allowing real-time adaptation to environmental changes and unforeseen obstacles. In addition, feedback from low-level controllers is utilized to refine symbolic mission plans, ensuring robustness and adaptability. This framework improves the robustness and effectiveness of USV operations by integrating the power of symbolic planning with the reasoning abilities of LLMs. In addition, it simplifies the mission specification, allowing operators to focus on high-level objectives without requiring complex programming. The simulation results validate the proposed approach, demonstrating its ability to optimize mission execution while seamlessly adapting to dynamic maritime conditions.

Figures (7)

• Figure 1: The LLM-guided mission planning framework consists of three main modules: the Prompt Generator, which formulates the mission goal and system prompt; the Symbolic Plan Generator, powered by GPT-4, to create an adaptive symbolic plan; and the Navigation and Control Module, which handles sensing, path planning, and precise execution using PID control. Real-time visual outputs provide feedback on perception and task performance.
• Figure 2: Description of the LLM prompt for the mission of inspecting port terminals in a maritime environment. The prompt specifies USV dimension, , its current location, the environment description such as dimensions and the locations of the docking stations, and the mission goal. Furthermore, it also contains the set of symbolic actions that a USV can perform during the mission and the example output in JSON formate with reasoning.
• Figure 3: Output of the LLM for the inspection mission plan, presenting a sequence of symbolic actions: moving to docking stations and recording data. The reasoning highlights the rationale behind the plan, which emphasize on minimizing the total distance traveled by following the shortest path sequence to ensure efficient and effective mission execution.
• Figure 4: Simulation setup designed to validate the proposed mission planning framework. The environment includes four docking stations labeled $T_1$, $T_2$, $T_3$, and $T_4$, with the position of the USV represented by a white circle. The sequence of snapshots in Figures 1 to 6 illustrates the execution of the mission, which involves inspecting Docking_Station-1 and Docking_Station-2. The simulation highlights the USV's ability to autonomously navigate between docking stations, execute inspection tasks, and adapt to the mission plan in a structured and efficient manner. Code: https://github.com/Muhayyuddin/llm-guided-mission-planning
• Figure 5: Comparison of planned and executed trajectories for Mission-1, highlighting the symbolic plan's waypoints and the USV's actual path as navigated by the control system.