Neuigkeiten
RPTU and Fraunhofer ITWM Present Joint Embedded AI Demonstrator at Embedded World 2026
At the Embedded World Exhibition 2026, the Rhineland-Palatinate Technical University Kaiserslautern-Landau (RPTU) and the Fraunhofer Institute for Industrial Mathematics (ITWM) jointly presented a demonstrator showcasing the next generation of embedded artificial intelligence development. The exhibit highlighted the strong cooperation between both partners and demonstrated how tightly integrated software and hardware innovation can significantly accelerate AI deployment on edge devices.
https://www.embedded-world.de/en
Demonstrator Validates 10× Faster Face Detection
A key highlight of the demonstrator was its ability to achieve up to ten times faster face detection compared to state-of-the-art baseline implementation. This performance gain illustrates the effectiveness of the jointly developed solution, which combines advanced neural architecture optimization with real hardware-based evaluation.
The demonstrator provided clear, measurable proof of how tailored AI models can outperform generic approaches when optimized directly for specific embedded platforms.
Combining Expertise: NASE and Hardware-in-the-Loop
The collaboration builds on complementary expertise from both institutions:
Fraunhofer ITWM is responsible for the development of the Neural Architecture Search Engine (NASE), an intelligent system that automatically designs and optimizes neural networks for embedded and edge environments.
RPTU Kaiserslautern-Landau contributes its expertise in Hardware-in-the-Loop (HiL), enabling the execution and evaluation of AI models directly on real hardware platforms.
This integration ensures that optimization is not based on theoretical estimates but on actual performance measurements. The feedback from real hardware execution is continuously incorporated into the model design process, resulting in highly efficient and application-specific AI solutions.
Accelerating Embedded AI Development
The joint system significantly reduces development time while improving model performance. Instead of adapting pre-existing models to hardware constraints, the solution generates architectures that are inherently aligned with the target platform.
By combining automated architecture search with real-time hardware evaluation, developers benefit from:
Faster development cycles
Improved runtime efficiency
Reduced energy consumption
Higher reliability in deployment scenarios
Strong Partnership for Future AI Systems
The successful presentation at Embedded World Exhibition 2026 underlines the impact of the collaboration between RPTU Kaiserslautern-Landau and Fraunhofer ITWM. Together, the partners are advancing the development of efficient, hardware-aware AI systems and contributing to the future of embedded intelligence in industry and research.
Acknowledgment
The presented research and demonstrator were supported by the Carl Zeiss Stiftung, Germany, under the Sustainable Embedded AI project (P2021-02-009). The demonstrator was exhibited at the booth of the Fraunhofer-Gesellschaft at the Embedded World Exhibition 2026.
Further Reading
Agentic Artificial Intelligence Accelerates Embedded Development
Neuronale Architektursuche mit Hardware in the Loop
Kaiserslautern. Agentic Artificial Intelligence Accelerates Embedded Development