REACT: Real-time Entanglement-Aware Coverage Path Planning for Tethered Underwater Vehicles

Technical Summary: REACT – Real-time Entanglement-Aware Coverage Path Planning for Tethered Underwater Vehicles

Problem Statement
Remotely Operated Vehicles (ROVs) are essential for inspecting underwater infrastructure, often relying on tethers for power and communication. However, the tether introduces a critical operational risk: entanglement with underwater structures, flora, or fauna. While numerous Coverage Path Planning (CPP) algorithms exist for untethered systems, they fail to account for tether constraints, leading to trajectories that may cause the tether to loop around obstacles. This can restrict vehicle mobility, reduce operational range, and necessitate extensive post-mission disentanglement or the carriage of excessive tether lengths. Existing tether-aware planners are often limited to 2D environments, rely on computationally intensive offline methods unsuitable for real-time use, or lack integrated frameworks for 3D inspection tasks.

Methodology
The authors propose REACT (Real-time Entanglement-Aware Coverage Path Planning), a framework designed to generate safe, entanglement-free inspection paths in arbitrary 3D environments. The framework decouples the problem into two phases:

1. Offline Planning: A standard, off-the-shelf CPP (specifically FC-Planner) generates a nominal, distance-optimal waypoint sequence based on a Signed Distance Field (SDF) map derived from a point cloud of the environment. This phase ignores tether constraints to ensure full coverage.
2. Online Execution & Replanning: A modular online layer manages tether constraints in real-time.
   • Tether Modeling: The system employs a computationally efficient, geometry-based tether model. The continuous tether is discretized into nodes. Using the SDF map, the model simulates a "taut" tether configuration by iteratively performing shortcutting (replacing collision-free segments with straight lines) and pulling (moving nodes away from non-convex obstacle cavities) operations. This allows for rapid prediction of the tether's 3D shape.
   • Entanglement-Aware Planner: The planner operates in two modes:
     • Normal Mode: Follows nominal waypoints if the tether length is within the maximum limit ($L_{max}$).
     • Recovery Mode: If the tether length approaches or exceeds $L_{max}$, the planner triggers a disentanglement search. This algorithm performs a backward search along the tether path to find a "pivot point." It constructs a candidate path by reversing the tether from the pivot to the vehicle and planning a new shortest path from the pivot to the target waypoint. If a feasible path respecting $L_{max}$ is found, the vehicle follows this recovery trajectory.
   • Path Refinement: Generated recovery paths undergo refinement (centroid offsetting, stochastic local perturbation, and polynomial smoothing) to ensure obstacle clearance and dynamic feasibility before being tracked by a Model Predictive Control (MPC) controller.

Key Contributions
The paper outlines four primary contributions:

1. A computationally efficient, geometry-based tether model that accurately predicts entangled tether configurations in 3D using SDF maps.
2. An efficient online replanning strategy that proactively prevents entanglement by enforcing a maximum tether length constraint.
3. A complete, integrated framework combining CPP, SDF-based mapping, and MPC to enable safe, full-coverage asset inspections.
4. Validation of the framework through both simulation and real-world experiments, demonstrating safe navigation and full coverage where conventional methods fail.

Results

• Simulation: In a pipe inspection scenario, REACT achieved 99.91% coverage with a total mission time of 680 seconds. While the inspection phase took longer (546s vs. 429s) due to proactive avoidance maneuvers, the total mission time was 20% faster than the baseline CPP (855s). The baseline planner required extensive post-mission disentanglement time (426s) and violated tether constraints significantly (reaching 31.16m vs. the 10m limit). REACT maintained tether lengths near the limit (max 10.52m) and avoided entanglement entirely.
• Real-World Experiments: Conducted in a test basin with a BlueROV2, the baseline CPP failed completely, becoming physically stuck due to tether entanglement at 124 seconds, achieving only 80.9% coverage. In contrast, REACT successfully completed the mission with 95.6% coverage. When an imminent entanglement was detected at 96 seconds, the replanner activated, computed a new path (with a maximum computation overhead of 0.48s), and successfully disentangled the vehicle to continue the inspection.

Significance
The paper claims that REACT addresses a critical gap in underwater asset inspection by enabling real-time, entanglement-aware planning in complex 3D environments. By integrating tether constraints directly into the planning loop, the framework eliminates the need for operators to carry excessive tether lengths to account for potential entanglements. Furthermore, it reduces overall operational time by avoiding the extensive post-mission disentanglement processes required by conventional planners. The results demonstrate that standard planners without tether awareness are insufficient for real-world underwater tasks, whereas REACT ensures mission completion and safety through adaptive, proactive replanning.