TL;DR
This paper presents a system for autonomous inspection of seaweed farms using an underwater robot (AUV). The system leverages sidescan sonar for navigation and detection in challenging underwater environments, demonstrating a practical approach to automating a crucial task for the growing seaweed aquaculture industry.
What Problem Does This Paper Address?
The global demand for food and biomass is increasing, and sustainable aquaculture, particularly of low-trophic species like seaweed, is seen as a key solution. Seaweed farming, especially large-scale operations, requires regular underwater inspection to monitor crop health, predict growth, and plan for harvesting. Traditional methods involving divers or surface vessels are costly, risky, and environmentally impactful due to fossil fuel consumption. This paper addresses the need for a cost-effective, safe, and automated solution for inspecting these farms, which are often located in dynamic and cluttered underwater environments where traditional navigation methods like GPS are unavailable.
Methods
The core of the system is an autonomous underwater vehicle (AUV) named SAM, designed for maneuverability in confined spaces. The autonomy stack runs on the Robot Operating System (ROS) and incorporates several key components:
- Mission Planning: A graphical user interface (GUI) called Neptus allows operators to plan inspection missions by defining waypoints.
- Mission Execution: A behavior tree (BT) framework ensures safe and compliant mission execution, handling sequences of actions and conditions.
- Navigation and Perception:
- Dead Reckoning with Extended Kalman Filter: Fuses inertial and external measurements to estimate the AUV's position and orientation underwater.
- Sidescan Sonar: Used for detecting ropes and buoys, which are crucial for farm localization and navigation. This is a key differentiator, as sidescan sonar is less affected by water conditions than cameras and provides range information.
- Rope and Buoy Detection: A window-sliding segmentation algorithm is employed to detect change points in the sidescan sonar signal, identifying ropes and buoys.
- Variational Gaussian Mixture Model (VGMM): Used to cluster buoy detections and create a robust, updated map of the seaweed farm, accounting for potential misdetections and drift.
- Line Following: A line-of-sight guidance law, based on look-ahead steering, is used to control the AUV along the seaweed lines, maintaining a desired offset for visual inspection.
- Simulation: The Stonefish simulator is used to validate the entire software pipeline and rehearse mission operations before deployment.
Datasets
The paper mentions preliminary field tests conducted at the Kristineberg marine field station on the Swedish west coast. Data were collected from this real-world test site during April, June, and November to train and test the algorithms, particularly for rope detection. The paper also describes the creation of a simulated seaweed farm environment within the Stonefish simulator, which was used to validate the system's control and mission execution pipeline.
Key Insights and Ideas
- Sidescan Sonar as a Primary Sensor: The paper highlights the effectiveness of sidescan sonar for navigating and detecting farm infrastructure (ropes, buoys) in challenging underwater conditions, overcoming limitations of cameras in murky water or at longer ranges.
- Robust Farm Initialization: The combination of VGMM for buoy clustering and piecewise linear fitting for line detection provides a robust method for building an accurate map of the seaweed farm, even with initial uncertainty or drift in buoy positions.
- Behavior Trees for Autonomous Control: The use of behavior trees offers a modular and safe way to manage complex autonomous missions, allowing for prioritization of actions and graceful handling of unexpected events or failures.
- Simulation for Validation: The integration of the Stonefish simulator is crucial for rapid development and testing of the AUV's software and control systems, reducing the need for extensive and costly sea trials.
Key Results
The paper demonstrates the successful implementation and testing of the proposed system.
- Rope Detection: The sidescan sonar-based rope detection algorithm was tested on real data, showing promising results for identifying farm structures.
- Farm Initialization: The VGMM-based buoy detection and line fitting were demonstrated in a simulated environment, producing an updated seaweed farm map.
- Line Following: The line-following inspection behavior was validated in simulation, showing effective control of the AUV along seaweed lines using the line-of-sight guidance law.
- Mission Execution: The entire system, from mission planning to execution, was tested in a simulated environment, validating the integrated pipeline.
Limitations and Future Work
The authors acknowledge several limitations and suggest future directions:
- Robust Detection with Varying Sizes: A key challenge is robustly detecting algae of varying sizes and in tightly spaced farm configurations.
- Sensor Robustness: Ensuring the robustness of acoustic sensors (used for mapping, navigation, and communication) to interference from onboard systems (motors, servos) is critical.
- Data-Driven Assessment: While the system can inspect the farm, obtaining actual numerical assessments of seaweed health and growth from sensor data (including cameras) is a future work item.
- Underwater Base Integration: Extending the system with an underwater docking station for recharging and data communication would enable long-term, continuous data collection.
- Broader Applications: The technologies developed have potential applications beyond seaweed farms, including infrastructure inspection, monitoring of other aquaculture species, and marine research.
- Collaboration: Future work will involve collaborations with local seaweed farming initiatives to further develop and deploy these underwater robots.
Summary updated