Collision-Free Path Planning for Underwater Gliders in Time-Varying Currents via Dual-Archive Niching Differential Evolution
This repository hosts the official project website and supplementary visual materials for our paper:
Collision-Free Path Planning for Underwater Gliders in Time-Varying Currents via Dual-Archive Niching Differential Evolution
The paper has been accepted for publication in Ocean Engineering.
The study presents MadDE-NDA, a GPU-accelerated planning framework for long-range underwater-glider missions in complex, time-varying ocean environments. The framework jointly considers ocean-current-assisted navigation, dynamic reachability, three-dimensional seabed collision avoidance, and mission travel time.
Rather than directly optimizing a variable-length sequence of heading commands, the proposed method uses a fixed-dimensional B-spline representation and reconstructs the resulting route into executable dive–climb profiles. A dual-archive niching differential evolution algorithm is then employed to explore multiple feasible route corridors while maintaining both solution quality and population diversity.
The complete project page is available at:
The website provides:
- detailed descriptions of the planning framework;
- B-spline trajectory encoding and waypoint reconstruction;
- GPU-based RK4 dynamic-feasibility evaluation;
- ESDF-based continuous collision checking;
- MadDE-NDA optimization mechanisms;
- experimental results for five real-world ocean regions;
- two-dimensional route animations;
- three-dimensional dive–climb trajectory visualizations;
- GPU acceleration and collision-checking benchmarks.
The horizontal route is represented using a fixed number of B-spline control-point offsets. This avoids the variable-dimensional optimization problem caused by different numbers of underwater-glider dive–climb cycles.
Candidate routes are evaluated through forward numerical simulation using fourth-order Runge–Kutta integration. The evaluation considers the spatial and temporal variations of real ocean-current data.
A three-dimensional Euclidean Signed Distance Field is constructed from seabed bathymetry. ESDF-guided sphere tracing is used to detect potential collisions along the continuous dive–climb trajectory.
MadDE-NDA extends differential evolution with:
- a quality archive for retaining historically competitive solutions;
- a diversity archive for preserving structurally different search information;
- stagnation-triggered niching for escaping inferior route corridors;
- adaptive archive-source sampling for balancing exploitation and exploration.
The computationally intensive trajectory reconstruction, RK4 integration, reachability analysis, and collision checking procedures are evaluated in parallel on the GPU.
The project website contains supplementary animations for five planning cases.
For each case, we provide:
- a 2D horizontal-route animation over the time-varying ocean-current field;
- a 3D trajectory animation showing the reconstructed dive–climb motion relative to the seabed terrain.
These materials demonstrate the current-aware, terrain-aware, and physically executable characteristics of the planned routes.
RAPID-route.github.io/
├── assets/ # Figures, animations, and website resources
├── real_data/ # Data used by the project website
├── index.markdown # English project page
├── index_cn.markdown # Chinese project page
├── 2_about.markdown # About page
├── _config.yml # GitHub Pages configuration
└── README.md # Repository introduction