HyDRA Scorpion: A Cost-effective and Modular ROV for Real-Time Underwater Inspection, Intervention, and Object Detection

Technical Summary: HyDRA Scorpion

Problem Statement
The paper addresses the significant barrier to underwater research and intervention in Low- and Middle-Income Countries (LMICs): the high cost and logistical dependence on imported commercial Remotely Operated Vehicles (ROVs). Existing commercial platforms are often prohibitively expensive, lack integration with advanced intelligence (AI), and are ill-suited for local environmental conditions such as turbid water. Conversely, many low-cost or academic ROVs possess mechanical capabilities but lack "inspection-grade" perception, relying on simple image processing demos rather than real-time detection, quantitative measurement, or 3D mapping. There is a persistent gap between mechanically capable platforms and those with integrated, actionable intelligence.

Methodology
The authors developed HyDRA Scorpion, a modular, cost-effective, 6-Degree-of-Freedom (6-DoF) tethered ROV designed for real-time inspection, intervention, and object detection. The system architecture is hierarchical:

• Hardware: The vehicle features an open-frame chassis constructed from corrosion-resistant HDPE and a custom 6-inch acrylic pressure housing (tested to 4 bar, equivalent to ~300m depth). It utilizes eight T200 thrusters for full 6-DoF maneuverability. The core electronics include a Raspberry Pi 5 for high-level AI processing and a Pixhawk 4 flight controller for real-time stability and motor control. Power is delivered via a 48V tether with onboard buck conversion.
• Manipulation: A key mechanical innovation is the "Stinger," a 4-DoF manipulator arm featuring a novel coaxial dual-motor drive. This design allows the end-effector to rotate 360 degrees continuously without tangling power or control wires, solving a common issue in affordable underwater robotics. A secondary "Pincer" gripper assists in tasks.
• Software & Perception: The system employs a Python-based stack (PyQt6, OpenCV) with a client-server communication model. The vision pipeline integrates:
  • AI Object Detection: A YOLOv8 model optimized with TensorRT for real-time inference, trained on datasets including COCO and a specific jellyfish dataset.
  • In-Situ Measurement: A calibrated pixel-to-metric system allowing operators to measure object lengths and distances non-contact.
  • 3D Mapping: A 360° photosphere generation module using ORB feature detection and RANSAC for homography estimation.
  • Control: A cascaded PID controller with constrained weighted least-squares optimization for thrust allocation and station-keeping.

Key Contributions

1. Low-Cost, High-Performance Platform: The development of a fully functional ROV with a total system cost of approximately $4,542.98, significantly lower than commercial equivalents, while maintaining robustness and modularity.
2. Integrated Perception Stack: Bridging the gap between mechanical capability and intelligence by integrating real-time AI detection, quantitative length/distance measurement, and 3D environmental mapping into a single operational framework.
3. Novel Manipulator Design: The creation of a 4-DoF "Stinger" arm with a coaxial drive mechanism enabling 360-degree tangle-free rotation, a feature previously unavailable in low-cost commercial options.
4. Validated Performance: Comprehensive experimental validation of the system's pressure integrity, station-keeping accuracy, and vision-based measurement capabilities in both simulated and real-world environments.

Experimental Results
The system underwent rigorous testing yielding the following results:

• Pressure Integrity: The electronics housing successfully withstood 4 bar of pressure (simulating ~300m depth) with no moisture ingress.
• Station-Keeping: The vehicle maintained its position within a tight tolerance of ±0.15 meters under external disturbances.
• Object Detection: The YOLOv8 model achieved a mean Average Precision (mAP@0.5) of 0.89 on a held-out jellyfish dataset and an 88% recall on a specialized crack detection task.
• Measurement Accuracy: The interactive length measurement tool demonstrated a consistent error of less than 5% against ground-truth data. Color detection accuracy for markers reached 94%.
• Intervention: The system successfully executed a mock anode replacement task, utilizing the 360° map for planning and the manipulator for gripping, rotating, and retracting the object.

Significance
The paper claims that HyDRA Scorpion provides a practical, accessible framework for developing capable, vision-enhanced robotic systems with limited resources. By combining a locally manufacturable, low-cost platform with an end-to-end perception stack, the project addresses the critical need for indigenous underwater tools in developing nations. It demonstrates that high-fidelity underwater inspection, scientific measurement, and complex intervention are achievable without reliance on expensive foreign imports, thereby supporting research, search and rescue, and sustainable seabed monitoring. The authors position the system not just as a robot, but as a scalable solution that bridges the gap between theoretical AI algorithms and practical, field-deployable underwater operations.