Edge AI Academy Summer School 2025 Successfully Held in Pisa
The Edge AI Academy Summer School took place on July 7–8, 2025, in Pisa, bringing together students, researchers, and professionals for two days of intensive training in edge AI, IoT, and edge computing. The event offered participants hands-on experience and expert-led sessions on cutting-edge technologies, including RISC-V architectures and ultra-low-power processors that are shaping the future of smart IoT and edge AI systems.
The Summer School has been very successful, with very interesting SmartEdge presentations by Andreas Stephan (Bosch), Aparna Saisree Thuluva (Siemens), Dave Raggett (W3C/ERCIM), Ayaan Verghese and Matthew Keating (DELL).
Andreas explored in his course “Scenario generation for autonomous driving using LLMs based on real-world edge data” scenario generation in the context of autonomous driving. It discussed the role of scenario generation in developing robust, and reliable autonomous driving systems and gave an overview of traditional and state-of-the-art scenario generation techniques. These include rule-based methods, statistical methods, deep-learning based methods, and Large Language Model (LLM) driven approaches.
The course “Generative AI to gain data insights from the shopfloor intuitively” held by Aparna demonstrated how AI technologies—such as Large Language Models, Natural Language Processing, and Knowledge Graphs—can address the challenges of handling diverse operational (OT) data from various machines on the factory floor. These machines, often from different vendors, generate data in inconsistent formats, units, and semantics. The session showcased how these AI tools were used to streamline the process of creating harmonized and uniform interfaces (e.g., Asset Administration Shell, OPC UA), enabling consistent and efficient use of OT data. Additionally, it illustrated how complex shopfloor information could be easily accessed and leveraged at the IT layer to extract valuable insights.
The lecture “Cognitive approaches to low-code real-time control of digital twins and swarm monitoring and control” held by Dave introduced cognitive approaches to low-code, real-time control of agent swarms using digital twins as abstractions, along with 2.5D and 3D methods for web-based monitoring and control. It highlighted how these cognitive strategies extend beyond traditional industrial robotics—typically limited to repetitive tasks—by enabling dynamic, context-aware behavior switching. This flexibility enhanced both human-robot and robot-robot collaboration. The session also demonstrated how web-based interfaces can simplify human oversight of factory operations.
The course “Autonomous mobile robots in smart manufacturing: A real-world approach to edge AI and computer vision” of Ayaan and Matthew examined how AI at the Edge and photogrammetric computer vision can support autonomous decision-making in swarm networks. It showcased how traditional, static manufacturing lines could evolve through the use of autonomous mobile robots and smart edge devices working collaboratively with minimal human oversight. Participants explored a low-code tool for training vision models, designed a reference architecture for edge deployments, and studied real-world applications of intelligent robotics in smart factory environments.
Attendees engaged in interactive lectures, practical workshops, and collaborative projects, gaining valuable insights into real-world applications of edge AI. The program was co-organized by several Chips JU and Horizon Europe projects, including EdgeAI, CLEVER, REBECCA, TRISTAN, NEUROKIT2E, SMARTY, dAIEDGE, and SMARTEDGE.