Underwater Swarm Robotics

• Underwater swarm robotics is a field where multiple autonomous underwater vehicles collaboratively perform tasks using bio-inspired distributed algorithms in complex aquatic environments.
• Advanced communication modalities, including acoustic, optical, and magnetic induction methods, are adapted to overcome bandwidth, latency, and energy challenges underwater.
• Integrated localization, sensing, and hierarchical control frameworks enable precise formation, mapping, and task execution, validated through field experiments and modular hardware designs.

Underwater swarm robotics refers to the study, design, and deployment of multiple autonomous underwater vehicles (AUVs) or robots that coordinate via distributed algorithms, local interactions, and constrained underwater communication modalities to collectively achieve navigation, exploration, sensing, mapping, or manipulation tasks in aquatic environments. This domain draws extensive inspiration from natural marine swarms (e.g., fish schools, barnacle colonies) and incorporates principles from control theory, optimization, artificial intelligence, and bio-inspired computation. The field addresses challenges unique to underwater operation, including limited bandwidth, latency and unreliability of acoustic/optical communication, energy constraints, complex hydrodynamics, and the absence of global positioning.

1. Bio-Inspired Coordination Algorithms for Underwater Swarms

Bio-inspired approaches constitute a foundational paradigm for underwater swarm robotics, with prominent algorithms such as Artificial Fish Swarm Algorithm (AFSA), Whale Optimisation Algorithm (WOA), Coral Reef Optimisation (CRO), and Marine Predators Algorithm (MPA) underpinning distributed coordination in formation control, coverage, sampling, and task allocation (Ramesh et al., 18 Jan 2026). These methods share key elements:

• AFSA relies on localized visual/neighbourhood perception and alternates among “preying,” “swarming,” and “following” rules, using proximity, relative fitness, and simple stepwise updates. Its message requirements are moderate and its adaptability to localization uncertainty is high.
• WOA implements spiral and encircling behaviors, exploiting intermittent global best information and spiral-based search. Its communication dependency is very low; agents only infrequently share global best positions.
• CRO encodes reef spatial competition, broadcast spawning, brooding, and depredation to drive optimization and allocation, maintaining exploration via settlement and replacement.
• MPA alternates between Brownian and Lévy motion, with environmental phase switches inspired by predator-prey dynamics and Fish Aggregating Device (FAD) phenomena. It achieves efficient exploration through Lévy flights and maintains scalability with minimal message passing.

A multi-dimensional classification [see Table below] situates these algorithms on axes of communication dependency, adaptability, energy efficiency, and scalability:

Algorithm	Comm. Dependency	Environmental Adaptability	Energy Efficiency	Scalability
AFSA	Moderate	High	Moderate	High
WOA	Low	Moderate	High	High
CRO	Moderate	Moderate	Moderate	Medium
MPA	Low	High	High	High

Key trade-offs are identified: AFSA/MPA offer superior local adaptability but slightly slower convergence; WOA and MPA minimize communication overhead, critical for low-bandwidth underwater networks (Ramesh et al., 18 Jan 2026).

2. Underwater Communication Modalities and Constraints

Underwater swarm robotics operates under severe channel limitations:

• Acoustic communication is the primary long-range link (up to 20 km), but suffers from low bandwidth (10 bps – 100 kbps), substantial latency (≈0.67 s/km), multipath, and Doppler shifts. Acoustic absorption increases with frequency per Thorp’s formula, requiring power increases of 4–10× for doubled range (Ramesh et al., 18 Jan 2026).
• Optical (blue-green) links offer Mbps–Gbps in clear water but are limited to <100 m and highly sensitive to turbidity.
• Magnetic induction and RF operate over only meter-scale ranges but enable very low-latency, high-throughput exchange in near-field coordination.

Advanced stack architectures adopt hybrid strategies, adapting link type to range, energy budget and data priority. Adaptive OFDM, TDMA—including acoustic slotting and guard bands to avoid inter-robot packet collision (Huang et al., 10 Nov 2025)—and bio-inspired distributed protocols augment performance. Bio-inspired mimicry and artificial stigmergy (pheromone gradients, acoustic camouflage) support covert, efficient swarm signaling (Ramesh et al., 18 Jan 2026).

3. Localization, Sensing, and State Estimation

Swarm-level spatial understanding is achieved in the absence of GPS and dense sensor networks by integrating acoustic ranging, inertial measurements, and local vision:

• Minimal-data localization employs inter-robot acoustic ranging coupled with motion (speed or heading) estimation, solved via global nonlinear least squares to recover the geometric swarm configuration. Ambiguities from symmetries are resolved with K≥3 time steps, yielding sub-decimeter accuracy for swarms up to N=10 under realistic noise and missing data (dell'Erba, 2020).
• Open-source platforms such as Raspi²USBL demonstrate that a passive acoustic inverted ultra-short baseline system can provide sub-meter, sub-degree positioning accuracy for multi-AUV swarms up to 1.3 km, leveraging synchronized OCXO clocks, matched-filter beamforming, and a multi-agent TDMA acoustic protocol for assignment and collision avoidance (Huang et al., 10 Nov 2025).
• Visual sensing and deep learning have been validated in micro-robotic swarms, where global and local vision data—when fused with onboard noisy coordinate data—yield highly accurate pose estimation and spatial reassembly (IoU up to 87.97%), vital for multi-robot mapping and monitoring in hazardous aquatic environments (Chen et al., 4 Mar 2025).

Effective localization and state fusion underpin formation control, cooperative behavior, and higher-order task execution.

4. Distributed and Hierarchical Control Frameworks

Control strategies range from centralized teleoperation to fully distributed, adaptive autonomy:

• Hierarchical architectures enable multi-level delegation: a human diver issues high-level commands through underwater gesture recognition (UGR), intercepted by a master AUV, which then relays commands to follower agents via acoustic/tethered links. Integration with digital twins supports real-time monitoring, parameter tuning, and operator-in-the-loop task redirection (Aldhaheri et al., 2024).
• Reactive, vision-based formation control in 3D underwater environments is realized through leader–follower strategies employing local perception, zone-based speed laws, and decentralized actuation (e.g., fin control), yielding convergence to various geometric patterns despite hydrodynamic inertia and perception dropout—without the need for RF communication or external localization (Ni et al., 2024).
• Learning-based multi-agent control: Dynamic-switching-enabled multi-agent reinforcement learning frameworks (DSBM/HSARL) enable robust multi-target tracking and adaptive group coordination under realistic sonar and hydrodynamic models, with hierarchical SDN architectures (USV-GC, LC-AUVs, ET-AUVs) decoupling control and data planes for scalable operation and significantly improved convergence/accuracy beyond conventional baselines (Wang et al., 2024).
• Semantics-driven fuzzy control leverages LLMs to abstract multimodal sensing into human-interpretable tokens, mapping these to fuzzy steering/gait via interpretable inference rules. Semantic intent sharing reduces communication payload and enables distributed agreement, facilitating robust and efficient exploration/coverage in map-free, GPS-denied scenarios (Xu et al., 2 Nov 2025).

5. Hardware Platforms, Modularity, and System Design

Robust and flexible hardware is critical for swarm feasibility:

• Modular platforms: The ModCube system enables self-assembly of 6-DoF cubic AUV modules with eight-thruster omnidirectional drive, electromagnet-based docking, and scalable morphologies. Rigorous characterization via Dirichlet/Willmore energy, per-thruster load variance, and empirical tank experiments (e.g., spiral tracking, cooperative lifting) demonstrate advantages over conventional monolithic ROVs, with robustness and adaptability for a range of tasks (Zheng et al., 2024).
• Persistence and settlement: NOAH (Nauplius Optimisation for Autonomous Hydrodynamics) introduces irreversible anchoring mechanisms inspired by barnacle colony morphogenesis. Agents progress from drift-adaptive exploration to settlement, after which they become sensing “colonies” and communication relays, scaling energy efficiency and providing task continuity under current-drifted environments (Ramesh et al., 17 Oct 2025).
• Micro-robotic swarms have been validated with microprocessor-based, low-cost, expendable robots for long-term visual monitoring, highlighting the feasibility of large-scale deployments in hazardous or confined environments (Chen et al., 4 Mar 2025).

System-level architectures typically comprise multi-core CPUs or SoCs running local control, multi-modal communication stacks, inertial/visual/acoustic localization, and adaptive power-management controllers (Ramesh et al., 18 Jan 2026).

6. Field Demonstrations, Benchmarking, and Open Research Directions

Empirical validation spans small-tank experiments to open-sea localization trials:

• 3D formation experiments with BlueSwarm show convergence and stability figures: steady-state lateral errors of 0.4 BL, depth ≤1 BL, and low vision-failure spikes (~1%) (Ni et al., 2024).
• Self-assembling modular robots succeed in high-precision tracking, robust docking (>90% success), and cooperative manipulation, with real-world morphological benchmarks supporting simulation-based inferences (Zheng et al., 2024).
• Swarm-level acoustic localization validated at sub-meter, sub-degree accuracy over 1.3 km, supporting formation control and state estimation in dense multi-agent deployments (Huang et al., 10 Nov 2025).
• Benchmarking frameworks: Bio-inspired optimization algorithms are systematically classified along axes of communication, adaptability, energy, and scalability; Table 2 in (Ramesh et al., 18 Jan 2026) tabulates ratings for AFSA, WOA, CRO, MPA.

Key open directions include: integration of 3D, non-stationary flow models into swarm control; hybrid communication protocols with environmental adaptation (acoustic/optical/MI); development of standardized benchmark suites for reproducibility; scalable, open-source modular hardware; and multi-week, real-world field deployments exposing swarms to ocean variability in current, temperature, and mission complexity (Ramesh et al., 18 Jan 2026).

7. Toward Real-World Underwater Swarm Autonomy

Current research converges toward marine-grounded adaptation—biomimetic control and morphological strategies rooted in fish, whale, and coral behavior; adaptive cross-layer communication integrating acoustic, optical, and MI links; and scalable, modular, open-source hardware platforms ranging from micro-AUVs to multi-meter systems (Ramesh et al., 17 Oct 2025, Zheng et al., 2024, Ramesh et al., 18 Jan 2026). Key challenges remain in bridging simulation–hardware gaps, establishing standard metrics and datasets, and achieving robust integration of perception, localization, and control. Field-validated performance, resilience to node losses and environmental disturbances, and human-swarm collaborative mechanisms (e.g., gesture-based UGR to human-in-the-loop digital twin integration) define current frontiers (Aldhaheri et al., 2024).

Continued interdisciplinary collaboration between optimization, control, sensing, robotics, and ocean engineering communities is essential for transitioning underwater swarms from controlled tank setups and simulation to scalable, robust, and adaptive systems operating in the open ocean.