Researchers Johny J. Lopez, Md Meftahul Ferdaus, and Mahdi Abdelguerfi from the University of Texas at Arlington have developed a novel approach to improve the autonomous inspection of underground infrastructure, such as sewer and culvert systems. Their work, published in the IEEE Internet of Things Journal, focuses on generating human-readable summaries from visual data collected by robotic platforms, enabling more efficient and scalable infrastructure maintenance.
The team presents a two-stage pipeline designed to work on resource-constrained edge devices. The first stage employs a lightweight model called RAPID-SCAN, which efficiently segments and identifies structural deficiencies in underground infrastructure. RAPID-SCAN achieves a high F1-score of 0.834 with only 0.64 million parameters, making it suitable for edge deployment. The second stage utilizes a fine-tuned Vision-Language Model (VLM) called Phi-3.5, which generates concise, domain-specific summaries in natural language from the segmentation outputs. To ensure real-time performance, the researchers employed post-training quantization with hardware-specific optimization, significantly reducing model size and inference latency without compromising summarization quality.
The researchers introduced a curated dataset of inspection images with manually verified descriptions to fine-tune and evaluate the VLM. They deployed and tested their complete pipeline on a mobile robotic platform, demonstrating its effectiveness in real-world inspection scenarios. The integration of AI systems like this one can bridge the gap between automated defect detection and actionable insights, paving the way for more scalable and autonomous inspection solutions in the energy and infrastructure sectors.
This research highlights the potential of edge-deployable integrated AI systems to enhance the autonomous inspection of underground infrastructure. By providing human-readable summaries of structural deficiencies, these systems can support more informed decision-making and improve public safety and urban sustainability. The practical applications for the energy sector include more efficient maintenance of critical infrastructure, reduced downtime, and improved resource allocation.
Source: IEEE Internet of Things Journal
This article is based on research available at arXiv.