Autonomous robotic bridging using distributed swarm control without inter-agent communication

Introduction

• Autonomous powered floating bridges or rafts could enable unpredictable, dispersed river crossings, improve crew survivability, and reduce logistics footprint compared with the Improved Ribbon Bridge.The Army seeks autonomous robotic bridging for the 2040 battlespace, where current gap-crossing technologies do not account for near-peer adversaries or lethal contested logistics.Source: §1 · p. 1
• The simulation demonstrates a 2x3 rafting formation reached without GPS or inter-agent communication, while GPS and RF can be incorporated in non-contested settings.Prior work demonstrated the swarm technology on UAVs and UGVs using different control strategies.Source: Figure 1 · p. 1, §1 · p. 2
• SCARAB combines distributed nonlinear formation control based on and new boat models with multimodal sensing and localization for USVs and docking targets.UTARI developed the system with Hydronalix, Inc.Source: §1 · pp. 1–2

2. BACKGROUND

• Existing formation methods may require GPS or peer communication, while distance- or bearing-based control can converge to an incorrect local-minimum formation.Consensus and distributed-pose approaches avoid global sensing but still require peer communication for pose estimation, orientation synchronization, or local-frame alignment.Source: §2 · p. 2
• Barycentric coordinates are presented as eliminating limitations of existing methods and enabling operation without GPS or RF communications, while allowing their optional use in civilian relief operations.The passage relates restricting RF communications and dispensing with GPS in contested environments to improved robustness and survivability.Source: §2 · p. 2
• The proposed distributed formation strategy is described as convergent and robust across vehicle dynamics, without global position, common heading, inter-agent communication, or a complete sensing graph.It is reported to remain stable under significant disturbances and input saturation, and incorporates a TRL 4+ distributed collision-avoidance algorithm with stability guarantees demonstrated on ground robots and drones.Source: §2 · p. 2
• Fathian et al. developed barycentric-coordinate-based control for planar formations, using local relative positions to support distributed control of heterogeneous agents without global position or centralized coordination.Their work addressed differing dynamic models and robustness to modeling uncertainty and disturbances, with demonstrations involving ground and aerial vehicles; the desired formation was shown achievable under broad conditions on dynamics and interaction structures.Source: §2 · p. 2
• Fathian et al. [2] showed swarm formations of up to nine units in simulation and with ground robots, with robustness to disturbances and trajectory changes for collision avoidance.The formulation was described as applicable when idealized dynamics and precise models cannot be assumed.Source: §2 · p. 3
• AprilTags were localized at more than 30 meters with less than 5% error in day or night, and can provide six-degree-of-freedom pose information for agents or docking targets.Other possible localization options include template-based pose estimation and scenic landmarks [4].Source: §2 · p. 3

3.1. Simulation software and multi-modal sensing of agents and docking targets

• The simulation uses VRX/Gazebo with a WAM-V model adapted to the Hydronalix Reckless frame, and AprilTag vision provides localization in GPS-degraded or denied settings.VRX/Gazebo models surface vehicles, hydrodynamic forces, currents, waves, and thrusters; the Reckless model’s propulsion and dynamics were modified, with further tuning remaining future work.Source: §3.1 · p. 3, Figure 2 · p. 3, Figure 3 · p. 4
• AprilTag centroid and LIDAR cube estimates diverged by less than 5%, while localization fused GPS and reversed AprilTag positions using an exponentially weighted moving average.When multiple tags are visible, the cube center is estimated by combining tag positions with known offsets and spherical linear interpolation.Source: Figure 4 · p. 4, §3.1 · pp. 4–5

3.2. Vehicle control model

• The USV controller separates heading and speed regulation, using formation-generated reference commands and sensor feedback to track desired behavior under disturbances.Two podded thrusters actuate the vehicle; the control architecture uses IMU orientation feedback for heading and GPS-derived position and velocity for speed, though GPS is not required when unavailable.Source: §3.2 · p. 5, Figure 6 · p. 5

3.3. Swarm formation control

• The extended swarm controller uses onboard vision for relative positions, needs neither inter-agent communication nor absolute position or shared orientation, and includes active collision avoidance.Convergence is evaluated by convergence time and mean formation error; the proof-of-concept tolerance is 1 m for 10 consecutive simulation steps.Source: §3.3 · pp. 5–6

• Six Hydronalix Reckless USVs were simulated in Gazebo using ROS 2 Jazzy on an Intel i7, testing 2x3 raft and X formations at 1 m spacing in the controller’s first maritime validation.Source: §4 · p. 6
• The 2x3 raft converged in et = 492 time steps (49.2 seconds simulation time; 5 mins 49 seconds wall time), while X converged in et = 332 (33 seconds; 3 mins 59 seconds wall time).Source: §4 · p. 6, Figure 8 · p. 6, Figure 9 · p. 7
• Five bays starting in random positions converged to an X pattern in an alternate swarm control result.Source: §4 · p. 6, Figure 9 · p. 7

5. CONCLUSION

• Floating bays can assemble autonomously through distributed swarm control and multimodal perception, with optional RF or acoustic communication.The conclusion presents this as SCARAB’s autonomous robotic bridging capability.Source: §5 · p. 7
• Robustness extensions include terrain avoidance, adverse-condition vision, and damaged-unit control; prior work demonstrated potential-fields obstacle handling and day/night infrared target detection on embedded low-SWaP-C hardware, while is a possible moving-target background-modeling approach.Source: §5 · p. 7
• A non-responsive unit could be detected after a waiting period, such as 5 minutes, and extra bays could substitute to achieve the desired formation.For example, nine bays could be deployed when only six are needed, allowing bays 8 and 9 to replace non-working bays 5 and 6.Source: §5 · p. 7