SkyRover develops an edge-native integrated UAV–UGV system, where ground robots and drones cooperate through a heterogeneous 5G/4G and Wi-Fi network for telemetry, control, and low-latency video streaming. Computationally intensive operations—such as object detection using YOLOv11 and reinforcement learning algorithms for autonomous navigation—are delegated to containerized edge computing nodes managed via Docker, reducing onboard computational load and improving system scalability. The entire architecture is coordinated by ROS2, which integrates the various subsystems: from the rover’s LiDAR to the drone’s MAVLink–MAVROS pipeline, up to the linking component that converts the UAV’s real-time pose into Nav2 navigation goals for the rover. This conversion involves coordinate scaling, data filtering, and static TF transformations, ensuring consistency across the two robots’ reference frames.The system thus enables robust and reproducible multi-robot cooperation, where reinforcement learning training, environment mapping, and drone–rover tracking are coordinated within a distributed edge-node infrastructure.

STREAM PILOT addresses the challenge of orchestrating next-generation multimedia services across the entire cloud–edge continuum, with the goal of minimizing latency while optimizing costs and energy consumption. In this context, the IDAGO algorithm, developed by the research unit at the University of Naples Federico II, has been integrated into the SUPER Orchestrator to jointly optimize service function placement within the network, traffic routing, and the allocation of both computational and network resources. IDAGO determines service placement by leveraging, on the one hand, the formalization of application requirements through the service graph, and on the other, the description of infrastructure capabilities through the network graph. The REC-EXEC module enforces the decisions produced by IDAGO by automatically deploying the services onto the infrastructure. Students contributed by extending the resource allocation modules, automating the generation of the graphs, and preparing the Docker images required for deploying the services within the Catania testbed.

MEI investigates how conversational agents and Virtual Humans can share emotional episodic memories on sensitive topics—particularly in the medical field—to dynamically adapt dialogue. The architecture combines a Neo4j knowledge graph (with local and remote shards) integrated with Kafka, a high-performance message distribution system, and an Unreal Engine 5 avatar connected via the FANTASIA framework and LangGraph. When a real user reaction does not match the predicted response, a new experience is generated to retrain the models. Students developed the prediction module (SVR), the distributed database integrated with Apache Kafka, the Unreal Engine integration, and a procedure for evaluating errors made by LLMs, as part of efforts to consider ethical and explainability aspects from the design phase.