Periodic Reporting for period 2 - edge air (On the edge AI-driven Autonomous Inspection Robots)
Reporting period: 2024-04-01 to 2025-03-31
The Edge AIR project, carried out by Energy Robotics, addresses the growing need for safer, more efficient, and automated inspection solutions in industrial environments such as chemical plants and energy infrastructures.
These industries depend on frequent inspections for safety and operational continuity, but traditional methods are labor-intensive, costly, and expose personnel to hazardous conditions. Demographic shifts have further reduced the availability of qualified workers, while increasingly dynamic, remote, and complex sites require more flexible solutions.
Edge AIR aimed to address these challenges by developing a robot-agnostic, AI-driven software platform that enables fleets of mobile robots to autonomously perform routine inspections. Unlike vendor-specific solutions, the platform was designed to integrate with robots from different manufacturers and connect with industrial systems through open interfaces. This ensures interoperability, scalability, and adaptability across a wide range of industrial use cases.
A core innovation lies in its use of novel capabilities for on-the-edge Artificial Intelligence, including semantic scene understanding, advanced multi-modal data analysis, and real-time sense & react capabilities. These innovations give robots an artificial form of "common sense," allowing them to perceive, interpret, and adapt to complex and dynamic environments in real time.
The project’s main objectives were to:
- Develop a robot-agnostic software platform for managing heterogeneous robotic fleets.
- Implement edge-based AI modules for autonomous perception and decision-making.
- Enable semantic navigation and adaptive behavior control for operation in complex industrial environments.
- Build a scalable AI ecosystem for continuous learning and extension.
- Demonstrate the platform’s applicability across multiple industries and use cases.
Thereby, Edge AIR contributes to EU priorities in digitalization, industrial safety, competitiveness, and sustainability. It reduces human exposure to risk, improves inspection quality, and supports the transition toward automated, data-driven infrastructure management. The approach also accounts for human-robot interaction and workforce adaptation, ensuring automation complements—not replaces—human roles.
These industries depend on frequent inspections for safety and operational continuity, but traditional methods are labor-intensive, costly, and expose personnel to hazardous conditions. Demographic shifts have further reduced the availability of qualified workers, while increasingly dynamic, remote, and complex sites require more flexible solutions.
Edge AIR aimed to address these challenges by developing a robot-agnostic, AI-driven software platform that enables fleets of mobile robots to autonomously perform routine inspections. Unlike vendor-specific solutions, the platform was designed to integrate with robots from different manufacturers and connect with industrial systems through open interfaces. This ensures interoperability, scalability, and adaptability across a wide range of industrial use cases.
A core innovation lies in its use of novel capabilities for on-the-edge Artificial Intelligence, including semantic scene understanding, advanced multi-modal data analysis, and real-time sense & react capabilities. These innovations give robots an artificial form of "common sense," allowing them to perceive, interpret, and adapt to complex and dynamic environments in real time.
The project’s main objectives were to:
- Develop a robot-agnostic software platform for managing heterogeneous robotic fleets.
- Implement edge-based AI modules for autonomous perception and decision-making.
- Enable semantic navigation and adaptive behavior control for operation in complex industrial environments.
- Build a scalable AI ecosystem for continuous learning and extension.
- Demonstrate the platform’s applicability across multiple industries and use cases.
Thereby, Edge AIR contributes to EU priorities in digitalization, industrial safety, competitiveness, and sustainability. It reduces human exposure to risk, improves inspection quality, and supports the transition toward automated, data-driven infrastructure management. The approach also accounts for human-robot interaction and workforce adaptation, ensuring automation complements—not replaces—human roles.
The project delivered and validated a robot-agnostic software platform capable of managing autonomous robots across diverse industrial environments. The system was tested in operational environments, such as energy facilities and chemical plants, and successfully integrated with multiple robotic platforms and customer IT systems.
Edge AIR’s on-the-edge AI modules enable robots to interpret and navigate their environments autonomously. Key outputs include 2D/3D semantic segmentation models, a 3D semantic scene graph for structured environmental representation, and anomaly detection models using multi-modal sensor data. A temporal data analysis framework enhances the accuracy of anomaly detection by integrating sensor inputs over time.
Major advancements were achieved in semantic navigation and adaptive behavior control. Highlights include the development of an evergreen digital twin of industrial sites, monocular depth estimation for low-visibility conditions, object-aware path planning, and sense & react modules for real-time response to environmental changes.
To ensure continuous development, a scalable AI ecosystem was established, including tools and infrastructure for training models, managing data pipelines, and deploying updates. Domain-specific datasets were collected to improve the accuracy and robustness of the system across a range of conditions.
Edge AIR’s on-the-edge AI modules enable robots to interpret and navigate their environments autonomously. Key outputs include 2D/3D semantic segmentation models, a 3D semantic scene graph for structured environmental representation, and anomaly detection models using multi-modal sensor data. A temporal data analysis framework enhances the accuracy of anomaly detection by integrating sensor inputs over time.
Major advancements were achieved in semantic navigation and adaptive behavior control. Highlights include the development of an evergreen digital twin of industrial sites, monocular depth estimation for low-visibility conditions, object-aware path planning, and sense & react modules for real-time response to environmental changes.
To ensure continuous development, a scalable AI ecosystem was established, including tools and infrastructure for training models, managing data pipelines, and deploying updates. Domain-specific datasets were collected to improve the accuracy and robustness of the system across a range of conditions.
Edge AIR represents a substantial leap forward in autonomous robotics for industrial inspection. By combining robot-agnostic system architecture with on-the-edge AI, the project breaks new ground in enabling robots to operate autonomously across diverse, real-world industrial environments.
At the core of this innovation is a hardware-agnostic AI platform that can manage heterogeneous fleets of inspection robots across various industrial environments. This overcomes the fragmentation and interoperability limitations posed by vendor-specific systems, allowing seamless integration of multiple robot types into a single operational ecosystem.
Edge AIR also introduces breakthroughs in semantic understanding of industrial sites. By leveraging advanced AI models and real-time sensor fusion, the system enables robots to build and maintain a persistent, semantically rich digital twin of the environment. This allows for deeper situational awareness and insight into the state of assets and surroundings. Complementing this are real-time sense & react behavior modules, which allow robots to autonomously respond to unexpected changes or anomalies during inspection missions.
These innovations position Energy Robotics as a leader in AI-powered industrial inspection. The project successfully transitioned from research to application, with pilot deployments validating system performance under real-world conditions. Work is continuing with existing and new customers to scale the solution and adapt it to additional industrial sectors.
Ongoing efforts focus on commercialization, standardization, and certification to support broader adoption. Future development will extend capabilities toward generalized industrial AI, capable of interpreting complex semantic relationships in industrial environments providing even more advanced insights to site operators. This will also enable new functionalities such as robotic manipulation, opening the door to fully autonomous inspection and interaction in more sectors, and reinforcing Europe’s position at the forefront of industrial AI and robotics.
At the core of this innovation is a hardware-agnostic AI platform that can manage heterogeneous fleets of inspection robots across various industrial environments. This overcomes the fragmentation and interoperability limitations posed by vendor-specific systems, allowing seamless integration of multiple robot types into a single operational ecosystem.
Edge AIR also introduces breakthroughs in semantic understanding of industrial sites. By leveraging advanced AI models and real-time sensor fusion, the system enables robots to build and maintain a persistent, semantically rich digital twin of the environment. This allows for deeper situational awareness and insight into the state of assets and surroundings. Complementing this are real-time sense & react behavior modules, which allow robots to autonomously respond to unexpected changes or anomalies during inspection missions.
These innovations position Energy Robotics as a leader in AI-powered industrial inspection. The project successfully transitioned from research to application, with pilot deployments validating system performance under real-world conditions. Work is continuing with existing and new customers to scale the solution and adapt it to additional industrial sectors.
Ongoing efforts focus on commercialization, standardization, and certification to support broader adoption. Future development will extend capabilities toward generalized industrial AI, capable of interpreting complex semantic relationships in industrial environments providing even more advanced insights to site operators. This will also enable new functionalities such as robotic manipulation, opening the door to fully autonomous inspection and interaction in more sectors, and reinforcing Europe’s position at the forefront of industrial AI and robotics.