The Mini TORTUGA is a compact and highly manoeuvrable Remotely Operated Vehicle (ROV) deployed by the SeaCAT to map the seafloor and identify debris before collection operations.
- Max. depth: 300 m
- Dimensions: L 672 mm x W 310 mm x H 381 mm
- Weight: 19 kg (ROV only)
- Max. speed: 4 knots
- Mission: Underwater mapping, debris detection, and environmental monitoring
- Sensors for litter detection (integrated by Subsea Tech): Full HD video, high resolution imaging sonar, side scan sonar, magnetometer
- Sensors for positioning (integrated by Subsea Tech): Inertial measurement unit, pressure sensor, Doppler velocity logger, acoustic positioning.
KEY FEATURES
- Hydrodynamic Design: The optimized form factor, based on the design of an autonomous vehicle, reduces drag and provides better resistance to ocean currents.
- Power Line Communication (PLC): The PLC communication protocol uses the power pair to transmit high-speed data via umbilical cable up to 500 m. It is more economical, more robust, and easier to repair than fibre optics.
- Portability: Weighing just under 20 kg, the Mini Tortuga can be transported, deployed, and recovered by a single operator. Such feature also reduces mob/demob costs.
- Integrated Sensors: The initial design already considered the addition of standard sensors such as an imaging sonar and an acoustic positioning system, allowing for simplified integration and effective sensor protection.
METHODS AND SOFTWARE FOR POSITIONING AND MAPPING
KEY FEATURES
- Underwater positioning based on nonlinear, extended Kalman filtering that fuses the sensors to obtain an estimate of the pose (position and attitude) of the ROV at each point in time. Position errors below 1.7m demonstrated in field experiments.
- AI-based, deep-learning litter detection from camera images, currently being extended to sonar images.
- Seafloor litter mapping software, placing detected litter on a seafloor map represented as a 3D point cloud, by labeling points that belong to litter objects. Accuracy of 0.9m demonstrated in field experiments.
- Software developed in the industry-standard Robotic Operating System for easy integration and extension.
LITTER DETECTION AND IDENTIFICATION SOFTWARE
KEY FEATURES
- Based on labeled synced camera-sonar dataset constructed within SeaClear2.0 containing litter items as well as marine flora and fauna.
- Litter detection and classification via a multi-modal camera-sonar pipeline based on deep learning object detection architectures.
- Estimation of the prediction uncertainty of the detections and classifications.
- Learning from multiple underwater image data sources by clustering similar data points and training an ensemble consisting of an object detection model per cluster.
