Periodic Reporting for period 1 - edge air (On the edge AI-driven Autonomous Inspection Robots)
Période du rapport: 2023-04-01 au 2024-03-31
Key European industries, such as energy, power&utilities, chemical and oil&gas, require regular inspections and maintenance of their capital-intensive infrastructures. These are carried out by qualified operators and are very demanding and expensive. In hazardous environments, they are also risky and result every year in injuries and deadly accidents. The shortage of qualified human capacity due to the demographic change will lead to an increasing number of unstaffed positions. Inspections are also error-prone since the type, quality, and consistency of inspection data collected by humans are limited. Inadequate inspection and maintenance practices are the main reasons for major accidents, causing huge environmental, health, and safety (HSE) issues and costs (Kang 2018). The industry is ready for Autonomous Inspection Robots (AIRs) but lacks robots with the capacity to adapt to changing environments, homogeneous programming platforms/interfaces, and integrators working across OEMs, geographies, and industries (McKinsey 2021).
Energy Robotics (ER) proposes an end-to-end robot-agnostic AI-driven software platform that enables a diverse fleet of mobile robots to autonomously conduct routine inspections in industrial plants, including remote and hazardous facilities.
ER is already offering an SW platform for inspection that features several cloud-based and robot-based navigation and inspection skills. However, the growth would be limited, and the competitiveness with respect to new entrants would be at risk. The EIC funding will allow a completely new paradigm for AIR inspection. Thanks to advanced proprietary AI functions like semantic navigation, multi-modal data understanding, sense and react capabilities, and edge skill training, not only reliable and high-quality inspection data is ensured, but also autonomous navigation and inspection even in the event of significant changes in its surroundings and the occurrence of anomalies based on a semantic 3D understanding of the environment. These will enable AIRs to act similar to human inspection specialists in the event of deviations and anomalies and will enormously improve the competitiveness of ER’s scalability of the solution, making it applicable in any situation and different industries. The business plan considers manifold impact on revenues and margins in the medium-long term.
Energy Robotics (ER) proposes an end-to-end robot-agnostic AI-driven software platform that enables a diverse fleet of mobile robots to autonomously conduct routine inspections in industrial plants, including remote and hazardous facilities.
ER is already offering an SW platform for inspection that features several cloud-based and robot-based navigation and inspection skills. However, the growth would be limited, and the competitiveness with respect to new entrants would be at risk. The EIC funding will allow a completely new paradigm for AIR inspection. Thanks to advanced proprietary AI functions like semantic navigation, multi-modal data understanding, sense and react capabilities, and edge skill training, not only reliable and high-quality inspection data is ensured, but also autonomous navigation and inspection even in the event of significant changes in its surroundings and the occurrence of anomalies based on a semantic 3D understanding of the environment. These will enable AIRs to act similar to human inspection specialists in the event of deviations and anomalies and will enormously improve the competitiveness of ER’s scalability of the solution, making it applicable in any situation and different industries. The business plan considers manifold impact on revenues and margins in the medium-long term.
Technical aspects:
Energy Robotics developed AI models for semantic understanding and achieved the following results:
- Semantic Datasets for Industrial Scenes: Developed extensive datasets leveraging both real and synthetic data to facilitate a deeper understanding of environmental semantics. These datasets are crucial for enabling our robotic systems to interpret and navigate various industrial settings, aiding in both navigation and inspection tasks.
- Training Models for Semantic Understanding of Industrial Sites: Developed and trained different models designed to extract semantic information from 2D and 3D data that enables our robots to interact with and inspect various objects found in industrial settings.
- Distributed Training Infrastructure: This reliable infrastructure trains large-scale models on extensive datasets, catering to the increasing complexity of deep learning models.
Currently, ER is actively engaged in pursuing the following steps:
- Real-Time Edge Inference: Ensuring the effectiveness of our trained models' real-time edge inference.
- Allocentric Semantic Map Generation Module: Developing a joint semantic map compatible with various types of robots.
Energy Robotics developed AI models for semantic understanding and achieved the following results:
- Semantic Datasets for Industrial Scenes: Developed extensive datasets leveraging both real and synthetic data to facilitate a deeper understanding of environmental semantics. These datasets are crucial for enabling our robotic systems to interpret and navigate various industrial settings, aiding in both navigation and inspection tasks.
- Training Models for Semantic Understanding of Industrial Sites: Developed and trained different models designed to extract semantic information from 2D and 3D data that enables our robots to interact with and inspect various objects found in industrial settings.
- Distributed Training Infrastructure: This reliable infrastructure trains large-scale models on extensive datasets, catering to the increasing complexity of deep learning models.
Currently, ER is actively engaged in pursuing the following steps:
- Real-Time Edge Inference: Ensuring the effectiveness of our trained models' real-time edge inference.
- Allocentric Semantic Map Generation Module: Developing a joint semantic map compatible with various types of robots.
Regular inspections and monitoring by AIRs can contribute to structurally reducing capital cost while achieving maintenance excellence, providing significant advantages: - Cut inspection costs by >33% - More effective thanks to 24/7 inspection without the need for active control by an operator - Reduce exposure of humans to hazardous environments - More cost-effective plant maintenance, avoiding both costs of (too) early replacements, as well as costs of incidents resulting from (too) late replacement - More reliable inspections, through huge amounts of data-collection and intelligent AI-based analysis - A new level of robotic inspection capabilities more similar to human inspection specialists - Facilitates integration of further digitalization and innovative techniques, such as predictive maintenance - Hardware-agnostic, fully-scalable, end-to-end inspection solution providing best robot-/sensor-/skill-fit for the customer’s needs - Carry out inspections even in ATEX/IECEx Zone 1 areas with explosive atmospheres- Unified access to different types of mobile robots and inspection sensor payloads.