Powering the Green Deal – AI-Driven Solutions for Resource Optimisation

The transition to a greener, more sustainable Europe demands not only bold policies but also innovative technologies capable of driving real-world change. This was the focus of the second webinar of the series “AI Innovation From Research to Market” on the theme of “Powering the Green Deal: AI-Driven Solutions for Resource Optimisation,” which brought together experts from academia, industry, and applied research to showcase how AI technologies are being leveraged to meet the ambitious goals of the European Green Deal.

The session featured four distinguished speakers who showcased cutting-edge projects that are transforming sectors such as manufacturing, agriculture, and waste management.

• Andrea Borghesi, Assistant Professor (Tenure-track) at the University of Bologna, opened the session with a presentation on the AI4Europe Benchmarking Guide, offering a structured approach to evaluating AI systems with transparency, accountability, and real-world applicability in mind.
  
• Gianpietro Bontempi, R&D Manager at Fonderia di Torbole, shared key results and tools developed within the ALCHIMIA project, which aims to integrate AI for more sustainable industrial processes.
  
• Erato Lazarou, Project Manager and Research Associate at the Smart Farming Technology Group of the Agricultural University of Athens, presented valuable lessons from the Smart Droplets project pilots, highlighting how AI-driven technologies are enhancing resource efficiency in agriculture.
  
• Finally, Fredy Raptopoulos, Robotics Engineer at ROBENSO, gave a hands-on look at the real-world deployment of the portable robotic material recovery facility (prMRF) on the Greek island of Kefalonia, developed through the RECLAIM project to support circular economy practices in waste management.

Benchmarking AI for a sustainable future: insights from the AI4Europe guide

Delivering on the ambitions of the European Green Deal requires not only cutting-edge AI technologies but also reliable methods to evaluate their real-world impact. Benchmarking plays a critical role in this process, offering a structured way to assess how AI systems perform under various constraints—such as time, cost, energy use, and fairness. It ensures that AI solutions deployed for sustainability and resource optimisation are effective, trustworthy, and aligned with broader environmental and social goals.

In his presentation, Andrea Borghesi, introduced the AI4Europe Benchmarking Guide—a key resource developed to support the transparent and consistent evaluation of AI applications across domains. The guide helps answer two central questions: Which algorithm performs best for a given task and context? and Which hardware platform is most suitable for its deployment?

Andrea explained that benchmarking is more than just measuring accuracy—it involves a formalised process of selecting relevant tasks, datasets, and evaluation metrics. The guide categorises metrics across several dimensions: performance (e.g., accuracy, precision, latency), efficiency (e.g., model size, inference cost), robustness (e.g., resilience to noisy inputs), fairness (e.g., performance across different user groups), and sustainability (e.g., energy consumption and carbon footprint). These comprehensive benchmarks are crucial to ensure that AI systems contribute positively to environmental and social sustainability.

He also described different types of benchmarks—component-level, application-level, synthetic, and real-world—each serving distinct goals such as model selection, deployment testing, or stress analysis. Borghesi emphasized that combining these benchmark types often provides the most meaningful insights.

To ensure scalability and reproducibility, the guide promotes automation and the use of standard APIs, making benchmarking processes more efficient and easier to integrate into development pipelines. Andreai also highlighted practical challenges—such as lack of standards for emerging AI domains, dataset bias, and over-optimisation on specific benchmarks—and suggested improvements through community collaboration and integration with MLOps tools.

Benchmarking is a cornerstone for deploying AI responsibly. The AI4Europe Benchmarking Guide supports the development of AI systems that are not only high-performing but also sustainable, fair, and aligned with the Green Deal’s vision of a greener, more equitable Europe.

AI for smarter manufacturing: The ALCHIMIA project at Fonderia di Torbole

In his presentation, Gianpietro Bontempi, R&D Manager at Fonderia di Torbole, offered a compelling look at how artificial intelligence can transform traditional industries like metal casting, bringing them in line with the goals of the European Green Deal through improved efficiency, sustainability, and waste reduction.

Fonderia di Torbole, one of Europe’s leading cast iron foundries for automotive brake components, produces around 150,000 tons of parts annually for major clients such as Stellantis, Toyota, and Kia, with a workforce of over 500 employees. The company has already taken significant steps toward sustainability: 90% of materials used are recycled, 95% of waste is reused, 25% of energy is self-generated, and a roadmap is in place for full decarbonisation by 2030.

However, challenges remain in the casting process, particularly regarding quality control. Parts must cool for up to three hours before their quality can be verified, and any failure to meet specifications—such as porosity or insufficient mechanical resistance—results in scrapping large volumes of material, energy, and time.

To address this, the company has implemented the ALCHIMIA project, an AI-powered platform designed to predict the final quality of cast parts using only initial production data. By correlating variables like chemical composition, thermal analysis, and production timings with final part quality, ALCHIMIA enables real-time decision-making. Operators can adjust the production process before the three-hour cooling period ends, avoiding waste and increasing efficiency.

The system also incorporates environmental impact monitoring and supports federated learning, enabling scalability across multiple facilities. After extensive data collection and integration, the platform is currently in the evaluation phase, with the goal of optimising its predictive accuracy and integrating it fully into daily operations.

Gianpietro’s presentation illustrated how traditional manufacturing can harness AI not just to improve productivity, but also to align industrial practices with the broader environmental objectives of the Green Deal.

Smart farming in action: Lessons from the Smart Droplets Project

Erato Lazarou, Project Manager and Research Associate at the Smart Farming Technology Group of the Agricultural University of Athens, presented the Smart Droplets project, a European-funded initiative using AI, robotics, and digital twins to revolutionize how pesticides and fertilizers are applied in agriculture. Aligned with the European Green Deal, the project aims to cut pesticide use by 50%, nutrient loss by 20%, and promote sustainable farming on at least 25% of European land by 2030.

Running since September 2022, Smart Droplets addresses key challenges in agriculture—ranging from overuse of chemicals and robotic safety to data utilisation and public acceptance. The core innovation lies in an integrated smart spraying system that combines AI-driven digital twins, autonomous robotic sprayers, and real-time data analytics to optimize crop treatment, reduce waste, and minimize environmental impact.

At the heart of the system are digital twins—virtual models of plants that simulate growth, detect diseases, and predict treatment needs. These are supported by real-time vision systems that detect weeds and validate the digital models. The autonomous robotic tractors use GPS and sensors to precisely navigate fields, applying chemicals only where needed through a direct injection system that mixes up to five products on-the-fly. The result: up to 40% reduction in pesticides, 15% less nutrient use, and 50% water savings.

Erato shared insights from two pilot sites: a 155-hectare apple orchard in Spain and a wheat field in Lithuania. The Spanish pilot tackled the complexities of orchard topography with sensor-based 3D modelling and real-time navigation, while the Lithuanian pilot demonstrated efficient coverage-based spraying in large, open fields. Both showed promising results for real-world deployment.

Beyond technology, the project emphasizes education and stakeholder engagement. Through the Smart Droplets Academy, it offers hands-on training, webinars, and online resources for farmers, agronomists, and technicians. This outreach is crucial to ensure that cutting-edge technologies are adopted effectively and responsibly by the agricultural community.

Erato concluded by highlighting that Smart Droplets not only supports sustainability goals but also helps farmers make smarter, data-informed decisions—bridging the gap between innovation and impact in European agriculture.

AI-powered waste sorting on the edge: The RECLAIM project’s portable recycling Facility

Fredy Raptopoulos, Co-founder and Technical Lead at ROBENSO, presented a groundbreaking innovation developed under the RECLAIM project—a portable, AI-powered material recovery facility (prMRF) designed for deployment in remote and island regions of Greece. This compact, autonomous unit brings advanced recycling capabilities directly to underserved municipalities, eliminating the need for long and inefficient waste transport chains.

The idea stemmed from a pressing need: many isolated areas in Greece, such as the island of Kefalonia, lacked access to proper recycling infrastructure. Previously, recyclables had to be stored for days, shipped via ferry to the mainland, processed, and then returned—an energy-intensive process with limited recovery efficiency. The prMRF radically simplifies this by enabling on-site sorting, compacting, and preparation for upcycling.

Housed in a standard 14-meter transport container, the prMRF integrates robotic arms, AI-based image recognition systems, and advanced material handling technologies. Inside, recyclable materials pass through a vibrating feeder and conveyor system, where RGB and hyperspectral cameras identify materials. Custom 1.5- and 2.5-degrees-of-freedom robotic arms—engineered for affordability and scalability—then sort the waste into 12 bins using vacuum grippers and magnetic tools.

The system can currently classify seven types of recyclable materials, with sorting accuracy reaching up to 94%. It achieves a processing rate of 120 items per minute, recovering up to 4 tons of material per day. Smart design elements, such as retractable legs for easy deployment, and automation features, allow operation by a single trained operator and one worker for material loading—making it ideal for local municipalities with limited technical expertise.

To optimize performance, RECLAIM leverages a digital twin developed in collaboration with KU Leuven. This virtual model allows the team to simulate and test various scenarios, such as different waste compositions or changes in market prices, adjusting bin allocations and sorting priorities in real time. A target planner further enhances efficiency by determining the most profitable sorting strategy based on current conditions.

Moreover, an intuitive monitoring interface and sensor feedback help detect blockages or performance issues, while municipalities can use built-in analytics tools to assess the composition of local waste streams and even incentivize or penalize communities based on sorting quality.

Now in the final testing and benchmarking phase, the prMRF is poised to offer a scalable, cost-effective solution for sustainable waste management in remote areas—bringing Europe one step closer to a circular economy that leaves no region behind.

The “Powering the Green Deal: AI-Driven Solutions for Resource Optimisation” webinar showcased how artificial intelligence is already delivering tangible impact across sectors—from manufacturing and agriculture to waste management. Through insightful presentations from leading researchers and innovators, participants saw how AI can drive sustainability, boost efficiency, and support Europe’s transition to a greener, more resilient economy. Whether it’s predicting quality in industrial casting, enabling precision farming, or transforming recycling in remote communities, these use cases prove that AI is not just a tool for digital advancement—it’s a catalyst for real-world environmental progress. As these projects continue to evolve, they offer a clear roadmap for integrating AI into the heart of the European Green Deal.