Funding for Groundbreaking Edge AI Research

The world of artificial intelligence (AI) is advancing at an unprecedented pace, and Europe is at the forefront of this evolution. If you're an AI researcher, startup, or organization focused on cutting-edge technology at the edge, the dAIEDGE 2nd Open Call offers an extraordinary opportunity to secure funding and technical support to bring your innovative ideas to life.

What is dAIEDGE?

dAIEDGE is a Network of Excellence (NoE) in Robotics and AI and seeks to strengthen and support the development of a dynamic European cutting-edge AI ecosystem under the umbrella of the European Lighthouse for AI, and to sustain the development of advanced AI. dAIEdge consortium partners are financially supported by the European Commission and the State Secretariat for Education, Research and Innovation, and this initiative seeks to enhance distributed, trustworthy, and scalable AI at the edge by funding projects that address key industrial and societal challenges.

Funding Highlights

• Total Fund: €600,000 (max grant per project €60,000)
• Support Duration: Up to 7 months
• Plan Phase (1 month, up to €9,000): Develop an Individual Mentoring Plan, including KPIs and milestones.
• Implementation Phase (6 months, up to €51,000): Deliver a proof of demonstration for your proposed solution.

Selected projects will also benefit from technical mentoring by leading dAIEDGE experts to help refine and scale their innovations.

Industry-Research Challenges to Tackle?

Participants are encouraged to propose solutions aligned with 20 predefined challenges. These challenges span a wide range of domains, technical issues, and problem types sourced by leading European industry and academic institutes with a research specialization on Edge AI. They address tasks such as detecting concealed weapons with mobile thermal imaging, implementing active inference on resource-constrained robotics platforms, and exploring domain-specific embeddings, federated learning frameworks, or specialized hardware architectures (FPGAs, SoCs, neuromorphic processors). By covering real-time object detection, energy-efficient neural networks, domain-adaptive solutions, and sustainable agriculture monitoring, this collection provides a comprehensive view of the critical fronts where research and industry should collaborate.

Challenge 1: Mobile Concealed Weapon Detection: Develop a thermal-imaging based concealed weapon detection system for mobile devices, aiming for real-time, high-accuracy performance. The solution must handle hardware constraints, ensure energy efficiency, and improve public safety in law enforcement scenarios.

Challenge 2: Continual Learning for Edge-Based 3D Environment Exploration: Implement gradient-free variational inference techniques for continual learning in robotics. By optimizing memory and computational efficiency on edge platforms, this challenge enhances autonomous navigation, ensuring models adapt continuously without catastrophic forgetting.?

Challenge 3: Real-Time Active Inference for IoT and Robotics: Design and deploy active inference models on edge devices to handle real-time decision-making in IoT and robotics. Emphasizing low latency, adaptability, and resource efficiency, this challenge fosters self-tuning AI agents that remain robust in changing environments.?

Challenge 4: Integration of AI-Powered Edge Device into a Virtual Lab: Integrate AI-powered edge devices into a remote benchmarking infrastructure (VLab) to simplify experimentation and performance comparison. This ensures reproducibility, broad accessibility, and accelerated adoption of edge AI solutions.

Challenge 5: Cloud-to-Edge AI for Plant Disease & Pest Detection with BioClip: Create a cloud-to-edge solution for plant disease and pest detection using BioClip and environmental data. The goal is real-time, on-device inference to improve sustainability, precision agriculture, and minimize chemical usage.

Challenge 6: FPGA-Based Energy-Efficient Edge AI via NAS: Develop an automated, hardware-aware NAS framework for FPGA-based edge AI, balancing accuracy, latency, and energy efficiency. The challenge targets rapid deployment, reduced engineering overhead, and scalable solutions for tasks like image classification.?

Challenge 7: SAFE AI with Synthetic Data and Domain Knowledge: Develop embeddings that integrate domain knowledge and synthetic data to increase robustness, explainability, and generalization in high-stakes tasks. This effort supports generalization, noise resilience, and efficient reuse of learned knowledge.

Challenge 8: Ultra-Low-Power Speech Enhancement at the Edge: Achieve ultra-low-power speech enhancement on edge hardware. By optimizing algorithms and leveraging multimodality, the challenge aims for significant energy savings and improved audio quality, suitable for drones and other resource-constrained platforms. Deploy speech enhancement AI on ultra-low power hardware (<50mW), suitable for drones and integrated with ONNX.

Challenge 9: COPD Home Monitoring Using Acoustic and Accelerometric Data: Build a home monitoring system that uses acoustic and accelerometric data to detect COPD exacerbations early. This healthcare-focused challenge seeks accurate, efficient inference on mobile devices, supporting personalized, preventative patient care.

Challenge 10: Off-Chip Weights Management on Streaming Architectures: Optimize off-chip weight management for large neural networks on streaming SoC architectures. The solution should minimize latency, energy inefficiency, and bandwidth constraints, ensuring smoother real-time edge processing.?

Challenge 11: 2D Vehicle Detection on PYNQ-Z1: Implement a resource-optimized vehicle detection network on the PYNQ-Z1 FPGA platform. Achieving real-time inference and efficient memory usage, this challenge addresses intelligent transportation systems needing low-power, fast decision-making.

Challenge 12: Real-Time Vision Transformer (ViT) on FPGA SoC: Deploy a real-time Vision Transformer (ViT) on FPGA SoCs for image classification. By optimizing attention mechanisms and memory transfers, the challenge aims for low-latency, energy-efficient performance in edge vision tasks.

Challenge 13: Spiking Neural Networks (SNNs) on the Edge: Demonstrate the capabilities of Spiking Neural Networks (SNNs) at the edge, potentially for speech enhancement. Leveraging sparse event-driven computation, SNNs promise ultra-low-power, high-efficiency performance across various modalities.?

Challenge 14: Energy-Efficient Text Embeddings at the Edge: Assess energy-efficient text embedding inference on edge hardware. The challenge involves testing multiple models and accelerators, aiming to reduce power consumption and bandwidth, and guiding selection of optimal NLP solutions for the edge.

Challenge 15: Automated Deployment & Tuning of Binary Neural Networks on FPGA: Automate deployment and performance evaluation of Binary Neural Networks on Zynq FPGA using DMWay middleware. This approach offers improved power efficiency and performance while simplifying workflows for edge AI developers.

Challenge 16: Deploying LLMs on Low-Memory Edge Devices: Enable large language model inference on devices with limited memory by dynamically loading weights and using FPGA acceleration. The solution should integrate FPGA accelerators and maintain throughput, expanding LLM applications where storage and bandwidth are constrained.

Challenge 17: Early Warning for Harmful Algal Blooms via EO Data: Use Earth Observation data and CNN models to detect harmful algal blooms (HABs) in inland and open waters, targeting compatibility with CogniSAT6. By creating HAB-focused datasets and CNN models compatible with satellite platforms, the challenge supports sustainable aquaculture management and environmental protection.

Challenge 18: Federated Machine Learning in Web Browsers: Build a federated learning framework using WebAssembly and WebGPU for decentralized ML in browsers, handling multiple data formats (text, images). This enables decentralized, privacy-preserving model training, supporting real-time decision-making without centralizing sensitive data.

Challenge 19: Machine Learning Benchmarks for On-Board Processing in Space: Extend the OBPMark-ML suite for benchmarking ML models in space applications. By optimizing performance across diverse hardware (FPGAs, AI accelerators), the challenge ensures efficient, real-time on-board processing for future space missions.?

Challenge 20: Incentivization Framework on Blockchain for AI Development: Design a blockchain-based incentivization framework for AI development. The solution must fairly compensate data, model, and compute resource providers, encouraging sustainable, collaborative AI ecosystems with transparent, decentralized governance.

For the full list of challenges and details, visit the official Open Call page.

Evaluation of Proposals

An independent evaluation committee will assess proposals based on three criteria—Excellence, Impact, and Implementation—with a consensus meeting of consortium members of dAIEdge making final determinations in cases of close scores.

Excellence:

• Ambition and European Dimension: Show how the project fits into the broader European AI and Edge Computing landscape, advancing beyond the current state-of-the-art and aligning with key industry challenges.
• Innovation: Highlight what makes the project unique compared to existing solutions, explaining how it introduces novel concepts, methods, or applications.
• Credibility of Methodology: Present a clear, well-structured approach and justifications for chosen methods.
• Gender Dimension: Demonstrate how the project’s R&I activities consider gender aspects, ensuring inclusivity and balanced perspectives.

Impact:

• Market Opportunity: Provide evidence that the resulting product, service, or technology has a clear market need and potential for commercialization.
• Competition and Differentiation: Illustrate how the innovation surpasses existing solutions, creating a distinctive competitive advantage.
• Commercial Strategy and Scalability: Show that the solution can address structural problems in the industry and scale to broader markets or sectors.
• Environmental and Social Benefits: Align with sustainability goals, European policy priorities, and deliver positive societal impact.

Implementation:

• Team Competence and Leadership: Introduce a cross-functional, well-qualified team with and a strong track record capable of moving from concept to market. Emphasize balanced skill sets, management capability, and gender diversity within the team.
• Resource Allocation: Match resources effectively to project goals, ensuring feasibility and efficient use of support.

Who Can Apply?

The dAIEDGE 2nd Open Call is more than just a funding opportunity. It’s a chance to contribute to Europe’s Edge AI leadership, collaborate with leading experts in the field, showcase your solutions on a global stage, and leverage cutting-edge resources.

This opportunity is open to Research and Technology Organizations (RTOs), Academia, SMEs and startups legally registered in EU Member States or Horizon Europe Associated Countries. Please read the fine print to check your eligibility or get in touch with the organizers of the call to check if you are eligible.

How to Apply?

Applications opened on January 10, 2025, and close on March 13, 2025, at 15:00 (Brussels Time). Submissions must be in English and completed via the online application form.

AI researchers and innovators, this is your moment to drive impactful change and shape the future of edge AI. Don’t miss the opportunity to be part of this transformative journey.