Detection of Simulated Fukushima Daichii Fuel Debris Using a Remotely Operated Vehicle at the Naraha Test Facility

Matthew Nancekievill, Jose Espinosa, Simon Watson, Barry Lennox, Ashley Jones, Malcolm John Joyce, Jun-ichi Katakura, Keisuke Okumura, So Kamada, Michio Katoh, Kazuya Nishimura2019 · DOI

TL;DR

This paper presents a novel, low-cost vision-based localization system for underwater robots operating in challenging environments like the Fukushima Daiichi nuclear disaster site. The system successfully enabled a Remotely Operated Vehicle (ROV) to create a 3D reconstruction of simulated nuclear fuel debris, demonstrating its potential for hazardous environment characterization and decommissioning.

What Problem Does This Paper Address?

The Fukushima Daiichi nuclear disaster in 2011 created highly radioactive and complex underwater environments within the reactor pressure vessels. Decommissioning these sites requires precise knowledge of the location and characteristics of submerged fuel debris, which is currently impossible for humans to access directly due to extreme radiation levels. Existing robotic solutions often struggle with localization accuracy, robustness in cluttered and low-visibility conditions, and cost-effectiveness. This paper tackles the challenge of developing a reliable and affordable underwater localization system that can accurately guide an ROV to map and characterize these hazardous submerged environments.

Methods

The core of the proposed solution is an External Vision Localization System (EVLS). This system utilizes a single camera mounted above the water surface, equipped with a wide-field-of-view (170°) fish-eye lens to capture a large area. The EVLS works by:

  • Image Calibration: Correcting for lens distortion, particularly the significant distortion introduced by the air-water interface.
  • Feature Identification: Detecting active markers (colored LED strips) on the ROV and passive landmark markers placed in the test environment.
  • Pixel to Real-World Coordinate Translation: Using the identified marker positions, a pressure sensor on the ROV for depth information, and a calibrated perspective transform matrix derived from the landmark markers to estimate the ROV's 3D position and attitude (pose).
  • Integration with Sonar: The localized ROV's position data is synchronized with sonar data from an IMAGENEX 831L pipe profiling sonar. This allows for the creation of 3D reconstructions by "stitching" together 2D sonar slices.
  • Robot Operating System (ROS): The entire system is implemented using ROS, with nodes programmed in Python for real-time operation.

The novelty lies in the low-cost implementation (under $150 USD), the external, air-to-water interface vision approach, and its adaptation for a specific, challenging application.

Datasets

Not discussed in the paper. The experiments were conducted in a controlled test facility (Naraha Test Facility) and a laboratory setup using mock ROVs and grids.

Key Insights and Ideas

  • Vision is Viable for Hazardous Underwater Localization: Despite challenges like water distortion and reflections, a vision-based system can achieve sufficient accuracy for mapping and characterization in relatively clear water, offering a cost-effective alternative to acoustic systems.
  • Active and Passive Markers are Key: The combination of active LED markers on the ROV and passive landmark markers in the environment provides robust tracking and allows for calibration in new environments.
  • Pressure Sensor is Crucial for 3D Accuracy: Integrating a pressure sensor on the ROV is essential for correcting depth distortion and achieving accurate 3D pose estimation, especially when dealing with the air-water interface.
  • Low-Cost Components Can Be Effective: The paper demonstrates that a functional and accurate localization system can be built using readily available, inexpensive components, making advanced robotics more accessible for hazardous applications.
  • 3D Reconstruction Bridges the Gap: Combining localization with sonar data is critical for creating a comprehensive understanding of submerged debris, moving beyond simple visual inspection to detailed mapping.

Key Results

  • Localization Accuracy: The system achieved a mean absolute error of 5 mm in a controlled grid experiment, with the most accurate position at the center of the camera's field of view (2.3 mm) and worst at the edges (8.1 mm). In the Naraha test tank, the localization system suggested an accuracy of approximately 100 mm.
  • 3D Reconstruction Accuracy: When combined with the sonar's accuracy (~50 mm), the overall 3D reconstruction accuracy was estimated to be around 150 mm.
  • Successful Demonstration: The system was successfully demonstrated at the Naraha Test Facility, enabling the creation of a 3D reconstruction of simulated fuel debris, providing a broad overview of the debris's shape and location.

Limitations and Future Work

  • Environmental Dependence: The system's accuracy can be affected by turbid water, which may require recalibration. Bright spot-lights also caused interference, though this is manageable in controlled deployments.
  • Calibration Requirement: The current system requires passive land-markers for calibration in new environments, which might be infeasible in some harsh scenarios.
  • Sonar Data Limitations: The sonar's relatively slow data acquisition rate (3 seconds per slice) can lead to inaccuracies in the 3D reconstruction if the ROV moves significantly during data capture. The "domed" nature of the sonar data also limits the clarity of individual metallic structures.
  • Future Work:
    • Moving the camera underwater in a waterproof housing to eliminate the need for external calibration markers.
  • Developing automated exposure calibration and brighter/larger active markers to improve performance in challenging lighting and water conditions.
  • Investigating non-linear approaches for depth estimation.
  • Improving data visualization to filter out acoustic noise and fuse data with IMU for better rotation handling.
  • Exploring alternative sonar systems for faster data acquisition.
  • Developing micro-ROVs and autonomous underwater vehicles that can leverage this low-cost positioning system in enclosed environments.

Summary updated

BibTeX
@misc{nancekievill2019detection,
  author = {Matthew Nancekievill and Jose Espinosa and Simon Watson and Barry Lennox and Ashley Jones and Malcolm John Joyce and Jun-ichi Katakura and Keisuke Okumura and So Kamada and Michio Katoh and Kazuya Nishimura},
  title = {{Detection of Simulated Fukushima Daichii Fuel Debris Using a Remotely Operated Vehicle at the Naraha Test Facility}},
  year = {2019},
  doi = {10.3390/s19204602},
}

Loading related content...

Figure