APEX-VISION
The Challenge
Name:
C8 – Open Challenge
Domain:
Autonomous Robotic Inspection, Domain Adaptive Deep Learning, Autonomous Navigation in GNSS-denied Confined Spaces, Decision Support
Challenge proposer:
Objectives:
To address the critical bottleneck of passive data collection in GNSS-denied environments by shifting robotic platforms into intelligent, autonomous decision-makers. The objective is to deploy an edge-based perception layer that filters irrelevant visual data, classifies structural defects (e.g., corrosion) on the fly, and dynamically generates inspection signals to adjust UAV navigation and prioritize structurally critical areas in real-time.
The Solution
Name:
A.I.S.A – Artificial Intelligence & Surveyor-infused Analysis for Unified Hull Integrity Assessment
The A.I.S.A. project addresses the gap between raw robotic data collection and actionable engineering intelligence by developing a software platform that acts as an intelligent “brain”. It combines qualitative (visual) and quantitative (UTM) datasets through a Spatio-Temporal Data Fusion Engine, while filtering the AI findings through a deterministic Expert System based on maritime rules, like CSR. By adding a 5-year historical data dimension, it predicts degradation rates to automate Class-ready condition evaluation reports and repair specifications.
The Solution
Name:
APEX-VISION – Autonomous Perception Engine for Real-Time Defect-Aware Robotic Inspection
APEX-VISION introduces a hardware-agnostic, edge-based perception engine for real-time, defect-aware robotic inspection. The system integrates Intelligent Frame Prioritization (IFP) and an Onboard Defect Classification Engine (ODCE) into a compact payload. This module processes visual data directly onboard, identifying and classifying corrosion types in real time, while filtering out low-quality or irrelevant frames. By generating immediate feedback signals (e.g., MAVLink-compatible), the system enables adaptive inspection behavior, guiding robotic platforms toward areas of higher structural relevance and significantly reducing data redundancy and post-processing efforts.
Unmanned Aerial Vehicle (UAV)-based inspections of maritime structures are rapidly advancing, yet current approaches remain largely inefficient. In confined, GNSS-denied environments (like ballast tanks and cargo holds), inspection missions generate vast volumes of visual data, much of which is blurred, redundant, or irrelevant. Moreover, drones operate on rigid, pre-programmed flight paths, lacking the capability to dynamically respond to detected structural anomalies. This results in overwhelmed communication bandwidth, delayed decision-making, increased post-processing effort, and suboptimal defect detection.
APEX-VISION addresses this limitation by introducing a fully integrated, edge-based perception engine that transforms this paradigm into a closed-loop, real-time perception system embedded directly at the edge. Centered around a high-performance edge compute module (e.g., NVIDIA Jetson Orin NX) running highly optimized AI models via TensorRT, the proposed solution combines three innovative aspects:
- Onboard intelligent data filtering (IFP): Real-time assessment of frame quality and relevance, ensuring that only high-value, sharp inspection data is retained. This significantly reduces data redundancy in bandwidth-constrained maritime environments.
- Embedded defect-aware perception (ODCE): Low-latency classification of corrosion types (e.g., as-new, rust, pitting, heavy scaling) and structural anomalies directly on the processing unit, utilizing domain-adaptive deep learning for the immediate interpretation of inspection data.
- Real-time adaptive inspection signaling: Generation of standardized control signals (e.g., MAVLink-compatible / ROS2) that allow robotic platforms to dynamically adjust their inspection behavior and autonomous navigation based on detected defects, introducing closed-loop autonomy into GNSS-denied inspection missions.
Unlike existing solutions, which treat perception and navigation as separate processes, APEX-VISION integrates perception-driven decision-making into a modular, hardware-agnostic payload. This approach enables seamless integration with diverse commercial robotic platforms without requiring the redesign of the host system. By shifting from passive data collection to active, defect-aware inspection, APEX-VISION significantly advances the state of the art, contributing to a more efficient, scalable, and intelligent structural integrity assessment framework for the AUTOASSESS ecosystem.
The Solution Provider
3DHUB is a Greek SME specializing in advanced digital manufacturing, computer vision, 3D scanning, artificial intelligence integration, and industrial digitalization solutions. The company has extensive experience in the development and deployment of innovative technologies for the maritime, industrial, aerospace, and defence sectors, while actively participating in numerous European R&D projects under Horizon Europe, EIT Manufacturing, EDF, and other innovation programmes. Combining expertise in embedded systems, digital inspection, AI-enabled data processing, and rapid prototyping, 3DHUB develops practical, industry-oriented solutions that bridge research and commercial deployment. Through APEX-VISION, the company leverages its multidisciplinary engineering capabilities to deliver a scalable edge-based perception system that enhances autonomous robotic inspection and supports the next generation of intelligent structural integrity assessment.
