When Semantics Connect the Swarm: LLM-Driven Fuzzy Control for Cooperative Multi-Robot Underwater Coverage

TL;DR

This work addresses cooperative underwater coverage under partial observability and GPS-denied conditions by introducing a semantics-guided fuzzy control framework that couples LLM-driven semantic abstraction with an interpretable fuzzy controller and lightweight semantic communication. Raw multimodal data are distilled into compact semantic tokens describing obstacles, unexplored regions, and OOIs, which are then mapped to smooth steering and gait commands; robots exchange intent through semantic tokens to coordinate exploration while avoiding redundancy. Key contributions include (i) LLM-guided efficient semantic abstraction, (ii) an interpretable fuzzy control module with expert-informed rules and centroid defuzzification, and (iii) semantic communication for swarm coordination, validated in Webots simulations across reef-like environments. The findings highlight robust OOI-oriented navigation and improved cooperative coverage efficiency under limited sensing and communication, suggesting a scalable path toward interpretable, semantic-driven underwater swarms in GPS-denied, map-free settings.

Abstract

Underwater multi-robot cooperative coverage remains challenging due to partial observability, limited communication, environmental uncertainty, and the lack of access to global localization. To address these issues, this paper presents a semantics-guided fuzzy control framework that couples Large Language Models (LLMs) with interpretable control and lightweight coordination. Raw multimodal observations are compressed by the LLM into compact, human-interpretable semantic tokens that summarize obstacles, unexplored regions, and Objects Of Interest (OOIs) under uncertain perception. A fuzzy inference system with pre-defined membership functions then maps these tokens into smooth and stable steering and gait commands, enabling reliable navigation without relying on global positioning. Then, we further coordinate multiple robots by introducing semantic communication that shares intent and local context in linguistic form, enabling agreement on who explores where while avoiding redundant revisits. Extensive simulations in unknown reef-like environments show that, under limited sensing and communication, the proposed framework achieves robust OOI-oriented navigation and cooperative coverage with improved efficiency and adaptability, narrowing the gap between semantic cognition and distributed underwater control in GPS-denied, map-free conditions.

Figures (12)

• Figure 1: Schematic illustration of the proposed semantics-guided fuzzy control framework, comprising Semantic Abstraction, Fuzzy Inference & Control, and Cooperative Navigation & Coverage, which together enable intelligent perception, decision-making, and coordination among multiple robots in the underwater coverage task.
• Figure 2: Overview of the semantics-guided fuzzy control framework, which consists of three modules: (A) LLM-Guided Efficient Semantic Abstraction; (B) Fuzzy Inference System and Controller Design; (C) LLM-Driven Semantic Communication for Cooperative Swarms. The framework integrates LLM-guided semantic abstraction, fuzzy inference control, and semantic communication to form a closed-loop perception–reasoning–action cycle. Together, these modules enable interpretable, adaptive, and cooperative navigation for the multi-robot system under uncertain underwater conditions.
• Figure 3: Visualization of the underwater coverage task simulated in the Webots platform. (a) Top-down view. (b) Side view. (c) Robot's real-time camera feed.
• Figure 4: Illustration of the two-stage rendering process, where the simulation scenes are first generated in Webots and subsequently refined in Python to produce a high-fidelity visualization.
• Figure 5: Top-view visualization of the three simulated underwater environments—Grid World, E-Shape, and Disconnected Paths, which are used to evaluate coverage performance under different spatial structures. (a) Grid World. (b) E-Shape. (c) Disconnected Paths.